Add readme

.dockerignore

@@ -1,4 +0,0 @@
-*/saved_models
-!vocoder/saved_models/pretrained/**
-!encoder/saved_models/pretrained.pt
-/datasets

.gitignore

@@ -14,11 +14,8 @@
 *.bcf
 *.toc
 *.sh
-data/ckpt
+*/saved_models
-!data/ckpt/vocoder/pretrained/**
+!vocoder/saved_models/pretrained/**
-!data/ckpt/encoder/pretrained.pt
+!encoder/saved_models/pretrained.pt
 wavs
 log
-!/docker-entrypoint.sh
-!/datasets_download/*.sh
-/datasets

.vscode/launch.json

@@ -15,8 +15,7 @@
       "name": "Python: Vocoder Preprocess",
       "type": "python",
       "request": "launch",
-      "program": "control\\cli\\vocoder_preprocess.py",
+      "program": "vocoder_preprocess.py",
-      "cwd": "${workspaceFolder}",
       "console": "integratedTerminal",
       "args": ["..\\audiodata"]
     },
@@ -24,8 +23,7 @@
       "name": "Python: Vocoder Train",
       "type": "python",
       "request": "launch",
-      "program": "control\\cli\\vocoder_train.py",
+      "program": "vocoder_train.py",
-      "cwd": "${workspaceFolder}",
       "console": "integratedTerminal",
       "args": ["dev", "..\\audiodata"]
     },
@@ -34,7 +32,6 @@
       "type": "python",
       "request": "launch",
       "program": "demo_toolbox.py",
-      "cwd": "${workspaceFolder}",
       "console": "integratedTerminal",
       "args": ["-d","..\\audiodata"]
     },
@@ -43,7 +40,6 @@
       "type": "python",
       "request": "launch",
       "program": "demo_toolbox.py",
-      "cwd": "${workspaceFolder}",
       "console": "integratedTerminal",
       "args": ["-d","..\\audiodata","-vc"]
     },
@@ -51,9 +47,9 @@
       "name": "Python: Synth Train",
       "type": "python",
       "request": "launch",
-      "program": "train.py",
+      "program": "synthesizer_train.py",
       "console": "integratedTerminal",
-      "args": ["--type", "synth", "..\\audiodata\\SV2TTS\\synthesizer"]
+      "args": ["my_run", "..\\"]
     },
     {
       "name": "Python: PPG Convert",
@@ -64,6 +60,14 @@
       "args": ["-c", ".\\ppg2mel\\saved_models\\seq2seq_mol_ppg2mel_vctk_libri_oneshotvc_r4_normMel_v2.yaml",
         "-m", ".\\ppg2mel\\saved_models\\best_loss_step_304000.pth", "--wav_dir", ".\\wavs\\input", "--ref_wav_path", ".\\wavs\\pkq.mp3", "-o", ".\\wavs\\output\\"
       ]
-    }
+    },
+    {
+      "name": "GUI",
+      "type": "python",
+      "request": "launch",
+      "program": "mkgui\\base\\_cli.py",
+      "console": "integratedTerminal",
+      "args": []
+    },
   ]
 }

Dockerfile

@@ -1,17 +0,0 @@
-FROM pytorch/pytorch:latest
-
-RUN apt-get update && apt-get install -y build-essential ffmpeg parallel aria2 && apt-get clean
-
-COPY ./requirements.txt /workspace/requirements.txt
-
-RUN pip install -r requirements.txt && pip install webrtcvad-wheels
-
-COPY . /workspace
-
-VOLUME [ "/datasets", "/workspace/synthesizer/saved_models/" ]
-
-ENV DATASET_MIRROR=default FORCE_RETRAIN=false TRAIN_DATASETS=aidatatang_200zh\ magicdata\ aishell3\ data_aishell TRAIN_SKIP_EXISTING=true
-
-EXPOSE 8080
-
-ENTRYPOINT [ "/workspace/docker-entrypoint.sh" ]

README-CN.md

@@ -20,15 +20,10 @@
 
 ### 进行中的工作
 *  GUI/客户端大升级与合并
-
+[X] 初始化框架 `./mkgui` （基于streamlit + fastapi）和 [技术设计](https://vaj2fgg8yn.feishu.cn/docs/doccnvotLWylBub8VJIjKzoEaee)
-- [x] 初始化框架 `./mkgui` （基于streamlit + fastapi）和 [技术设计](https://vaj2fgg8yn.feishu.cn/docs/doccnvotLWylBub8VJIjKzoEaee)
+[X] 增加 Voice Cloning and Conversion的演示页面
-
+[X] 增加Voice Conversion的预处理preprocessing 和训练 training 页面 
-- [x] 增加 Voice Cloning and Conversion的演示页面
+[ ] 增加其他的的预处理preprocessing 和训练 training 页面 
-
-- [x] 增加Voice Conversion的预处理preprocessing 和训练 training 页面 
-
-- [ ] 增加其他的的预处理preprocessing 和训练 training 页面 
-
 * 模型后端基于ESPnet2升级
 
 
@@ -82,7 +77,7 @@
 对效果影响不大，已经预置3款，如果希望自己训练可以参考以下命令。
 * 预处理数据:
 `python vocoder_preprocess.py <datasets_root> -m <synthesizer_model_path>`
-> `<datasets_root>`替换为你的数据集目录，`<synthesizer_model_path>`替换为一个你最好的synthesizer模型目录，例如 *sythensizer\saved_models\xxx*
+> `<datasets_root>`替换为你的数据集目录，`<synthesizer_model_path>`替换为一个你最好的synthesizer模型目录，例如 *sythensizer\saved_mode\xxx*
 
 
 * 训练wavernn声码器:
@@ -92,10 +87,7 @@
 * 训练hifigan声码器:
 `python vocoder_train.py <trainid> <datasets_root> hifigan`
 > `<trainid>`替换为你想要的标识，同一标识再次训练时会延续原模型
-* 训练fregan声码器:
+
-`python vocoder_train.py <trainid> <datasets_root> --config config.json fregan`
-> `<trainid>`替换为你想要的标识，同一标识再次训练时会延续原模型
-* 将GAN声码器的训练切换为多GPU模式：修改GAN文件夹下.json文件中的"num_gpus"参数
 ### 3. 启动程序或工具箱
 您可以尝试使用以下命令：
 
@@ -113,12 +105,12 @@
 ### 4. 番外：语音转换Voice Conversion(PPG based)
 想像柯南拿着变声器然后发出毛利小五郎的声音吗？本项目现基于PPG-VC，引入额外两个模块（PPG extractor + PPG2Mel）, 可以实现变声功能。（文档不全，尤其是训练部分，正在努力补充中）
 #### 4.0 准备环境
-* 确保项目以上环境已经安装ok，运行`pip install espnet` 来安装剩余的必要包。
+* 确保项目以上环境已经安装ok，运行`pip install -r requirements_vc.txt` 来安装剩余的必要包。
 * 下载以下模型 链接：https://pan.baidu.com/s/1bl_x_DHJSAUyN2fma-Q_Wg 
 提取码：gh41
-  * 24K采样率专用的vocoder（hifigan）到 *vocoder\saved_models\xxx*
+  * 24K采样率专用的vocoder（hifigan）到 *vocoder\saved_mode\xxx*
-  * 预训练的ppg特征encoder(ppg_extractor)到 *ppg_extractor\saved_models\xxx*
+  * 预训练的ppg特征encoder(ppg_extractor)到 *ppg_extractor\saved_mode\xxx*
-  * 预训练的PPG2Mel到 *ppg2mel\saved_models\xxx*
+  * 预训练的PPG2Mel到 *ppg2mel\saved_mode\xxx*
 
 #### 4.1 使用数据集自己训练PPG2Mel模型 (可选)
 
@@ -127,7 +119,7 @@
 `python pre4ppg.py <datasets_root> -d {dataset} -n {number}`
 可传入参数：
 * `-d {dataset}` 指定数据集，支持 aidatatang_200zh, 不传默认为aidatatang_200zh
-* `-n {number}` 指定并行数，CPU 11700k在8的情况下，需要运行12到18小时！待优化
+* `-n {number}` 指定并行数，CPU 11770k在8的情况下，需要运行12到18小时！待优化
 > 假如你下载的 `aidatatang_200zh`文件放在D盘，`train`文件路径为 `D:\data\aidatatang_200zh\corpus\train` , 你的`datasets_root`就是 `D:\data\`
 
 * 训练合成器, 注意在上一步先下载好`ppg2mel.yaml`, 修改里面的地址指向预训练好的文件夹：
@@ -136,7 +128,7 @@
 
 #### 4.2 启动工具箱VC模式
 您可以尝试使用以下命令：
-`python demo_toolbox.py -vc -d <datasets_root>`
+`python demo_toolbox.py vc -d <datasets_root>`
 > 请指定一个可用的数据集文件路径，如果有支持的数据集则会自动加载供调试，也同时会作为手动录制音频的存储目录。
 <img width="971" alt="微信图片_20220305005351" src="https://user-images.githubusercontent.com/7423248/156805733-2b093dbc-d989-4e68-8609-db11f365886a.png">
 
@@ -147,36 +139,35 @@
 | --- | ----------- | ----- | --------------------- |
 | [1803.09017](https://arxiv.org/abs/1803.09017) | GlobalStyleToken (synthesizer)| Style Tokens: Unsupervised Style Modeling, Control and Transfer in End-to-End Speech Synthesis | 本代码库 |
 | [2010.05646](https://arxiv.org/abs/2010.05646) | HiFi-GAN (vocoder)| Generative Adversarial Networks for Efficient and High Fidelity Speech Synthesis | 本代码库 |
-| [2106.02297](https://arxiv.org/abs/2106.02297) | Fre-GAN (vocoder)| Fre-GAN: Adversarial Frequency-consistent Audio Synthesis | 本代码库 |
 |[**1806.04558**](https://arxiv.org/pdf/1806.04558.pdf) | SV2TTS | Transfer Learning from Speaker Verification to Multispeaker Text-To-Speech Synthesis | 本代码库 |
 |[1802.08435](https://arxiv.org/pdf/1802.08435.pdf) | WaveRNN (vocoder) | Efficient Neural Audio Synthesis | [fatchord/WaveRNN](https://github.com/fatchord/WaveRNN) |
 |[1703.10135](https://arxiv.org/pdf/1703.10135.pdf) | Tacotron (synthesizer) | Tacotron: Towards End-to-End Speech Synthesis | [fatchord/WaveRNN](https://github.com/fatchord/WaveRNN)
 |[1710.10467](https://arxiv.org/pdf/1710.10467.pdf) | GE2E (encoder)| Generalized End-To-End Loss for Speaker Verification | 本代码库 |
 
-## 常见问题(FQ&A)
+## 常見問題(FQ&A)
-#### 1.数据集在哪里下载?
+#### 1.數據集哪裡下載?
 | 数据集 | OpenSLR地址 | 其他源 (Google Drive, Baidu网盘等) |
 | --- | ----------- | ---------------|
 | aidatatang_200zh | [OpenSLR](http://www.openslr.org/62/) | [Google Drive](https://drive.google.com/file/d/110A11KZoVe7vy6kXlLb6zVPLb_J91I_t/view?usp=sharing) |
 | magicdata | [OpenSLR](http://www.openslr.org/68/) | [Google Drive (Dev set)](https://drive.google.com/file/d/1g5bWRUSNH68ycC6eNvtwh07nX3QhOOlo/view?usp=sharing) |
 | aishell3 | [OpenSLR](https://www.openslr.org/93/) | [Google Drive](https://drive.google.com/file/d/1shYp_o4Z0X0cZSKQDtFirct2luFUwKzZ/view?usp=sharing) |
 | data_aishell | [OpenSLR](https://www.openslr.org/33/) |  |
-> 解压 aidatatang_200zh 后，还需将 `aidatatang_200zh\corpus\train`下的文件全选解压缩
+> 解壓 aidatatang_200zh 後，還需將 `aidatatang_200zh\corpus\train`下的檔案全選解壓縮
 
 #### 2.`<datasets_root>`是什麼意思?
-假如数据集路径为 `D:\data\aidatatang_200zh`，那么 `<datasets_root>`就是 `D:\data`
+假如數據集路徑為 `D:\data\aidatatang_200zh`，那麼 `<datasets_root>`就是 `D:\data`
 
-#### 3.训练模型显存不足
+#### 3.訓練模型顯存不足
-训练合成器时：将 `synthesizer/hparams.py`中的batch_size参数调小
+訓練合成器時：將 `synthesizer/hparams.py`中的batch_size參數調小
 ```
-//调整前
+//調整前
 tts_schedule = [(2,  1e-3,  20_000,  12),   # Progressive training schedule
                 (2,  5e-4,  40_000,  12),   # (r, lr, step, batch_size)
                 (2,  2e-4,  80_000,  12),   #
                 (2,  1e-4, 160_000,  12),   # r = reduction factor (# of mel frames
                 (2,  3e-5, 320_000,  12),   #     synthesized for each decoder iteration)
                 (2,  1e-5, 640_000,  12)],  # lr = learning rate
-//调整后
+//調整後
 tts_schedule = [(2,  1e-3,  20_000,  8),   # Progressive training schedule
                 (2,  5e-4,  40_000,  8),   # (r, lr, step, batch_size)
                 (2,  2e-4,  80_000,  8),   #
@@ -185,15 +176,15 @@ tts_schedule = [(2,  1e-3,  20_000,  8),   # Progressive training schedule
                 (2,  1e-5, 640_000,  8)],  # lr = learning rate
 ```
 
-声码器-预处理数据集时：将 `synthesizer/hparams.py`中的batch_size参数调小
+聲碼器-預處理數據集時：將 `synthesizer/hparams.py`中的batch_size參數調小
 ```
-//调整前
+//調整前
 ### Data Preprocessing
         max_mel_frames = 900,
         rescale = True,
         rescaling_max = 0.9,
         synthesis_batch_size = 16,                  # For vocoder preprocessing and inference.
-//调整后
+//調整後
 ### Data Preprocessing
         max_mel_frames = 900,
         rescale = True,
@@ -201,16 +192,16 @@ tts_schedule = [(2,  1e-3,  20_000,  8),   # Progressive training schedule
         synthesis_batch_size = 8,                  # For vocoder preprocessing and inference.
 ```
 
-声码器-训练声码器时：将 `vocoder/wavernn/hparams.py`中的batch_size参数调小
+聲碼器-訓練聲碼器時：將 `vocoder/wavernn/hparams.py`中的batch_size參數調小
 ```
-//调整前
+//調整前
 # Training
 voc_batch_size = 100
 voc_lr = 1e-4
 voc_gen_at_checkpoint = 5
 voc_pad = 2
 
-//调整后
+//調整後
 # Training
 voc_batch_size = 6
 voc_lr = 1e-4
@@ -219,13 +210,13 @@ voc_pad =2
 ```
 
 #### 4.碰到`RuntimeError: Error(s) in loading state_dict for Tacotron: size mismatch for encoder.embedding.weight: copying a param with shape torch.Size([70, 512]) from checkpoint, the shape in current model is torch.Size([75, 512]).`
-请参照 issue [#37](https://github.com/babysor/MockingBird/issues/37)
+請參照 issue [#37](https://github.com/babysor/MockingBird/issues/37)
 
-#### 5.如何改善CPU、GPU占用率?
+#### 5.如何改善CPU、GPU佔用率?
-视情况调整batch_size参数来改善
+適情況調整batch_size參數來改善
 
-#### 6.发生 `页面文件太小，无法完成操作`
+#### 6.發生 `頁面文件太小，無法完成操作`
-请参考这篇[文章](https://blog.csdn.net/qq_17755303/article/details/112564030)，将虚拟内存更改为100G(102400)，例如:文件放置D盘就更改D盘的虚拟内存
+請參考這篇[文章](https://blog.csdn.net/qq_17755303/article/details/112564030)，將虛擬內存更改為100G(102400)，例如:档案放置D槽就更改D槽的虚拟内存
 
 #### 7.什么时候算训练完成？
 首先一定要出现注意力模型，其次是loss足够低，取决于硬件设备和数据集。拿本人的供参考，我的注意力是在 18k 步之后出现的，并且在 50k 步之后损失变得低于 0.4

README.md

@@ -37,7 +37,7 @@
 * Install [ffmpeg](https://ffmpeg.org/download.html#get-packages).
 * Run `pip install -r requirements.txt` to install the remaining necessary packages.
 * Install webrtcvad `pip install webrtcvad-wheels`(If you need)
-> Note that we are using the pretrained encoder/vocoder but synthesizer since the original model is incompatible with the Chinese symbols. It means the demo_cli is not working at this moment.
+> Note that we are using the pretrained encoder/vocoder but synthesizer, since the original model is incompatible with the Chinese sympols. It means the demo_cli is not working at this moment.
 ### 2. Prepare your models
 You can either train your models or use existing ones:
 
@@ -68,7 +68,7 @@ Allowing parameter `--dataset {dataset}` to support aidatatang_200zh, magicdata,
 | @author | https://pan.baidu.com/s/1iONvRxmkI-t1nHqxKytY3g  [Baidu](https://pan.baidu.com/s/1iONvRxmkI-t1nHqxKytY3g) 4j5d  |  | 75k steps trained by multiple datasets
 | @author | https://pan.baidu.com/s/1fMh9IlgKJlL2PIiRTYDUvw  [Baidu](https://pan.baidu.com/s/1fMh9IlgKJlL2PIiRTYDUvw) code：om7f  |  | 25k steps trained by multiple datasets, only works under version 0.0.1
 |@FawenYo | https://drive.google.com/file/d/1H-YGOUHpmqKxJ9FRc6vAjPuqQki24UbC/view?usp=sharing https://u.teknik.io/AYxWf.pt  | [input](https://github.com/babysor/MockingBird/wiki/audio/self_test.mp3) [output](https://github.com/babysor/MockingBird/wiki/audio/export.wav) | 200k steps with local accent of Taiwan, only works under version 0.0.1
-|@miven| https://pan.baidu.com/s/1PI-hM3sn5wbeChRryX-RCQ code: 2021 https://www.aliyundrive.com/s/AwPsbo8mcSP code: z2m0 | https://www.bilibili.com/video/BV1uh411B7AD/ | only works under version 0.0.1
+|@miven| https://pan.baidu.com/s/1PI-hM3sn5wbeChRryX-RCQ code：2021 | https://www.bilibili.com/video/BV1uh411B7AD/ | only works under version 0.0.1
 
 #### 2.4 Train vocoder (Optional)
 > note: vocoder has little difference in effect, so you may not need to train a new one.
@@ -90,11 +90,6 @@ You can then try to run:`python web.py` and open it in browser, default as `http
 You can then try the toolbox:
 `python demo_toolbox.py -d <datasets_root>`
 
-#### 3.3 Using the command line
-You can then try the command:
-`python gen_voice.py <text_file.txt> your_wav_file.wav`
-you may need to install cn2an by "pip install cn2an" for better digital number result.
-
 ## Reference
 > This repository is forked from [Real-Time-Voice-Cloning](https://github.com/CorentinJ/Real-Time-Voice-Cloning) which only support English.
 
@@ -102,7 +97,6 @@ you may need to install cn2an by "pip install cn2an" for better digital number r
 | --- | ----------- | ----- | --------------------- |
 | [1803.09017](https://arxiv.org/abs/1803.09017) | GlobalStyleToken (synthesizer)| Style Tokens: Unsupervised Style Modeling, Control and Transfer in End-to-End Speech Synthesis | This repo |
 | [2010.05646](https://arxiv.org/abs/2010.05646) | HiFi-GAN (vocoder)| Generative Adversarial Networks for Efficient and High Fidelity Speech Synthesis | This repo |
-| [2106.02297](https://arxiv.org/abs/2106.02297) | Fre-GAN (vocoder)| Fre-GAN: Adversarial Frequency-consistent Audio Synthesis | This repo |
 |[**1806.04558**](https://arxiv.org/pdf/1806.04558.pdf) | **SV2TTS** | **Transfer Learning from Speaker Verification to Multispeaker Text-To-Speech Synthesis** | This repo |
 |[1802.08435](https://arxiv.org/pdf/1802.08435.pdf) | WaveRNN (vocoder) | Efficient Neural Audio Synthesis | [fatchord/WaveRNN](https://github.com/fatchord/WaveRNN) |
 |[1703.10135](https://arxiv.org/pdf/1703.10135.pdf) | Tacotron (synthesizer) | Tacotron: Towards End-to-End Speech Synthesis | [fatchord/WaveRNN](https://github.com/fatchord/WaveRNN)

archived_untest_files/demo_cli.py

@@ -1,9 +1,9 @@
-from models.encoder.params_model import model_embedding_size as speaker_embedding_size
+from encoder.params_model import model_embedding_size as speaker_embedding_size
 from utils.argutils import print_args
 from utils.modelutils import check_model_paths
-from models.synthesizer.inference import Synthesizer
+from synthesizer.inference import Synthesizer
-from models.encoder import inference as encoder
+from encoder import inference as encoder
-from models.vocoder import inference as vocoder
+from vocoder import inference as vocoder
 from pathlib import Path
 import numpy as np
 import soundfile as sf

control/cli/train_ppg2mel.py

@@ -1,66 +0,0 @@
-import sys
-import torch
-import argparse
-import numpy as np
-from utils.hparams import HpsYaml
-from models.ppg2mel.train.train_linglf02mel_seq2seq_oneshotvc import Solver
-
-# For reproducibility, comment these may speed up training
-torch.backends.cudnn.deterministic = True
-torch.backends.cudnn.benchmark = False
-
-def main():
-    # Arguments
-    parser = argparse.ArgumentParser(description=
-            'Training PPG2Mel VC model.')
-    parser.add_argument('--config', type=str, 
-                        help='Path to experiment config, e.g., config/vc.yaml')
-    parser.add_argument('--name', default=None, type=str, help='Name for logging.')
-    parser.add_argument('--logdir', default='log/', type=str,
-                        help='Logging path.', required=False)
-    parser.add_argument('--ckpdir', default='ppg2mel/saved_models/', type=str,
-                        help='Checkpoint path.', required=False)
-    parser.add_argument('--outdir', default='result/', type=str,
-                        help='Decode output path.', required=False)
-    parser.add_argument('--load', default=None, type=str,
-                        help='Load pre-trained model (for training only)', required=False)
-    parser.add_argument('--warm_start', action='store_true',
-                        help='Load model weights only, ignore specified layers.')
-    parser.add_argument('--seed', default=0, type=int,
-                        help='Random seed for reproducable results.', required=False)
-    parser.add_argument('--njobs', default=8, type=int,
-                        help='Number of threads for dataloader/decoding.', required=False)
-    parser.add_argument('--cpu', action='store_true', help='Disable GPU training.')
-    parser.add_argument('--no-pin', action='store_true',
-                        help='Disable pin-memory for dataloader')
-    parser.add_argument('--test', action='store_true', help='Test the model.')
-    parser.add_argument('--no-msg', action='store_true', help='Hide all messages.')
-    parser.add_argument('--finetune', action='store_true', help='Finetune model')
-    parser.add_argument('--oneshotvc', action='store_true', help='Oneshot VC model')
-    parser.add_argument('--bilstm', action='store_true', help='BiLSTM VC model')
-    parser.add_argument('--lsa', action='store_true', help='Use location-sensitive attention (LSA)')
-
-    ###
-    paras = parser.parse_args()
-    setattr(paras, 'gpu', not paras.cpu)
-    setattr(paras, 'pin_memory', not paras.no_pin)
-    setattr(paras, 'verbose', not paras.no_msg)
-    # Make the config dict dot visitable
-    config = HpsYaml(paras.config)
-
-    np.random.seed(paras.seed)
-    torch.manual_seed(paras.seed)
-    if torch.cuda.is_available():
-        torch.cuda.manual_seed_all(paras.seed)
-
-    print(">>> OneShot VC training ...")
-    mode = "train"
-    solver = Solver(config, paras, mode)
-    solver.load_data()
-    solver.set_model()
-    solver.exec()
-    print(">>> Oneshot VC train finished!")
-    sys.exit(0)
-
-if __name__ == "__main__":
-    main()   

control/mkgui/train.py

@@ -1,106 +0,0 @@
-from pydantic import BaseModel, Field
-import os
-from pathlib import Path
-from enum import Enum
-from typing import Any
-from models.synthesizer.hparams import hparams
-from models.synthesizer.train import train as synt_train
-
-# Constants
-SYN_MODELS_DIRT = f"data{os.sep}ckpt{os.sep}synthesizer"
-ENC_MODELS_DIRT = f"data{os.sep}ckpt{os.sep}encoder"
-
-
-# EXT_MODELS_DIRT = f"data{os.sep}ckpt{os.sep}ppg_extractor"
-# CONV_MODELS_DIRT = f"data{os.sep}ckpt{os.sep}ppg2mel"
-# ENC_MODELS_DIRT = f"data{os.sep}ckpt{os.sep}encoder"
-
-# Pre-Load models
-if os.path.isdir(SYN_MODELS_DIRT):    
-    synthesizers =  Enum('synthesizers', list((file.name, file) for file in Path(SYN_MODELS_DIRT).glob("**/*.pt")))
-    print("Loaded synthesizer models: " + str(len(synthesizers)))
-else:
-    raise Exception(f"Model folder {SYN_MODELS_DIRT} doesn't exist.")
-
-if os.path.isdir(ENC_MODELS_DIRT):    
-    encoders =  Enum('encoders', list((file.name, file) for file in Path(ENC_MODELS_DIRT).glob("**/*.pt")))
-    print("Loaded encoders models: " + str(len(encoders)))
-else:
-    raise Exception(f"Model folder {ENC_MODELS_DIRT} doesn't exist.")
-
-class Model(str, Enum):
-    DEFAULT = "default"
-
-class Input(BaseModel):
-    model: Model = Field(
-        Model.DEFAULT, title="模型类型",
-    )
-    # datasets_root: str = Field(
-    #     ..., alias="预处理数据根目录", description="输入目录（相对/绝对）,不适用于ppg2mel模型",
-    #     format=True,
-    #     example="..\\trainning_data\\"
-    # )
-    input_root: str = Field(
-        ..., alias="输入目录", description="预处理数据根目录",
-        format=True,
-        example=f"..{os.sep}audiodata{os.sep}SV2TTS{os.sep}synthesizer"
-    )
-    run_id: str = Field(
-        "", alias="新模型名/运行ID", description="使用新ID进行重新训练，否则选择下面的模型进行继续训练",
-    )
-    synthesizer: synthesizers = Field(
-        ..., alias="已有合成模型", 
-        description="选择语音合成模型文件."
-    )
-    gpu: bool = Field(
-        True, alias="GPU训练", description="选择“是”，则使用GPU训练",
-    )
-    verbose: bool = Field(
-        True, alias="打印详情", description="选择“是”，输出更多详情",
-    )
-    encoder: encoders = Field(
-        ..., alias="语音编码模型", 
-        description="选择语音编码模型文件."
-    )
-    save_every: int = Field(
-        1000, alias="更新间隔", description="每隔n步则更新一次模型",
-    )
-    backup_every: int = Field(
-        10000, alias="保存间隔", description="每隔n步则保存一次模型",
-    )
-    log_every: int = Field(
-        500, alias="打印间隔", description="每隔n步则打印一次训练统计",
-    )
-
-class AudioEntity(BaseModel):
-    content: bytes
-    mel: Any
-
-class Output(BaseModel):
-    __root__: int
-
-    def render_output_ui(self, streamlit_app) -> None:  # type: ignore
-        """Custom output UI.
-        If this method is implmeneted, it will be used instead of the default Output UI renderer.
-        """
-        streamlit_app.subheader(f"Training started with code: {self.__root__}")
-
-def train(input: Input) -> Output:
-    """Train(训练)"""
-
-    print(">>> Start training ...")
-    force_restart = len(input.run_id) > 0
-    if not force_restart:
-        input.run_id = Path(input.synthesizer.value).name.split('.')[0]
-    
-    synt_train(
-        input.run_id, 
-        input.input_root, 
-        f"data{os.sep}ckpt{os.sep}synthesizer", 
-        input.save_every, 
-        input.backup_every, 
-        input.log_every, 
-        force_restart,
-        hparams
-    )
-    return Output(__root__=0)

datasets_download/CN.txt

@@ -1,8 +0,0 @@
-https://openslr.magicdatatech.com/resources/62/aidatatang_200zh.tgz
-  out=download/aidatatang_200zh.tgz
-https://openslr.magicdatatech.com/resources/68/train_set.tar.gz
-  out=download/magicdata.tgz
-https://openslr.magicdatatech.com/resources/93/data_aishell3.tgz
-  out=download/aishell3.tgz
-https://openslr.magicdatatech.com/resources/33/data_aishell.tgz
-  out=download/data_aishell.tgz

datasets_download/EU.txt

@@ -1,8 +0,0 @@
-https://openslr.elda.org/resources/62/aidatatang_200zh.tgz
-  out=download/aidatatang_200zh.tgz
-https://openslr.elda.org/resources/68/train_set.tar.gz
-  out=download/magicdata.tgz
-https://openslr.elda.org/resources/93/data_aishell3.tgz
-  out=download/aishell3.tgz
-https://openslr.elda.org/resources/33/data_aishell.tgz
-  out=download/data_aishell.tgz

datasets_download/US.txt

@@ -1,8 +0,0 @@
-https://us.openslr.org/resources/62/aidatatang_200zh.tgz
-  out=download/aidatatang_200zh.tgz
-https://us.openslr.org/resources/68/train_set.tar.gz
-  out=download/magicdata.tgz
-https://us.openslr.org/resources/93/data_aishell3.tgz
-  out=download/aishell3.tgz
-https://us.openslr.org/resources/33/data_aishell.tgz
-  out=download/data_aishell.tgz

datasets_download/datasets.sha256sum

@@ -1,4 +0,0 @@
-0c0ace77fe8ee77db8d7542d6eb0b7ddf09b1bfb880eb93a7fbdbf4611e9984b  /datasets/download/aidatatang_200zh.tgz
-be2507d431ad59419ec871e60674caedb2b585f84ffa01fe359784686db0e0cc  /datasets/download/aishell3.tgz
-a4a0313cde0a933e0e01a451f77de0a23d6c942f4694af5bb7f40b9dc38143fe  /datasets/download/data_aishell.tgz
-1d2647c614b74048cfe16492570cc5146d800afdc07483a43b31809772632143  /datasets/download/magicdata.tgz

datasets_download/default.txt

@@ -1,8 +0,0 @@
-https://www.openslr.org/resources/62/aidatatang_200zh.tgz
-  out=download/aidatatang_200zh.tgz
-https://www.openslr.org/resources/68/train_set.tar.gz
-  out=download/magicdata.tgz
-https://www.openslr.org/resources/93/data_aishell3.tgz
-  out=download/aishell3.tgz
-https://www.openslr.org/resources/33/data_aishell.tgz
-  out=download/data_aishell.tgz

datasets_download/download.sh

@@ -1,8 +0,0 @@
-#!/usr/bin/env bash
-
-set -Eeuo pipefail
-
-aria2c -x 10 --disable-ipv6 --input-file /workspace/datasets_download/${DATASET_MIRROR}.txt --dir /datasets --continue
-
-echo "Verifying sha256sum..."
-parallel --will-cite -a /workspace/datasets_download/datasets.sha256sum "echo -n {} | sha256sum -c"

datasets_download/extract.sh

@@ -1,29 +0,0 @@
-#!/usr/bin/env bash
-
-set -Eeuo pipefail
-
-mkdir -p /datasets/aidatatang_200zh
-if [ -z "$(ls -A /datasets/aidatatang_200zh)" ] ; then
-    tar xvz --directory /datasets/ -f /datasets/download/aidatatang_200zh.tgz --exclude 'aidatatang_200zh/corpus/dev/*' --exclude 'aidatatang_200zh/corpus/test/*'
-    cd /datasets/aidatatang_200zh/corpus/train/
-    cat *.tar.gz | tar zxvf - -i
-    rm -f *.tar.gz
-fi
-
-mkdir -p /datasets/magicdata
-if [ -z "$(ls -A /datasets/magicdata)" ] ; then
-    tar xvz --directory /datasets/magicdata -f /datasets/download/magicdata.tgz train/
-fi
-
-mkdir -p /datasets/aishell3
-if [ -z "$(ls -A /datasets/aishell3)" ] ; then
-    tar xvz --directory /datasets/aishell3 -f /datasets/download/aishell3.tgz train/
-fi
-
-mkdir -p /datasets/data_aishell
-if [ -z "$(ls -A /datasets/data_aishell)" ] ; then
-    tar xvz --directory /datasets/ -f /datasets/download/data_aishell.tgz
-    cd /datasets/data_aishell/wav/
-    cat *.tar.gz | tar zxvf - -i --exclude 'dev/*' --exclude 'test/*'
-    rm -f *.tar.gz
-fi

demo_toolbox.py

@@ -1,5 +1,5 @@
 from pathlib import Path
-from control.toolbox import Toolbox
+from toolbox import Toolbox
 from utils.argutils import print_args
 from utils.modelutils import check_model_paths
 import argparse
@@ -17,15 +17,15 @@ if __name__ == '__main__':
         "supported datasets.", default=None)
     parser.add_argument("-vc", "--vc_mode", action="store_true", 
                         help="Voice Conversion Mode(PPG based)")
-    parser.add_argument("-e", "--enc_models_dir", type=Path, default=f"data{os.sep}ckpt{os.sep}encoder", 
+    parser.add_argument("-e", "--enc_models_dir", type=Path, default="encoder/saved_models", 
                         help="Directory containing saved encoder models")
-    parser.add_argument("-s", "--syn_models_dir", type=Path, default=f"data{os.sep}ckpt{os.sep}synthesizer", 
+    parser.add_argument("-s", "--syn_models_dir", type=Path, default="synthesizer/saved_models", 
                         help="Directory containing saved synthesizer models")
-    parser.add_argument("-v", "--voc_models_dir", type=Path, default=f"data{os.sep}ckpt{os.sep}vocoder", 
+    parser.add_argument("-v", "--voc_models_dir", type=Path, default="vocoder/saved_models", 
                         help="Directory containing saved vocoder models")
-    parser.add_argument("-ex", "--extractor_models_dir", type=Path, default=f"data{os.sep}ckpt{os.sep}ppg_extractor", 
+    parser.add_argument("-ex", "--extractor_models_dir", type=Path, default="ppg_extractor/saved_models", 
                         help="Directory containing saved extrator models")
-    parser.add_argument("-cv", "--convertor_models_dir", type=Path, default=f"data{os.sep}ckpt{os.sep}ppg2mel", 
+    parser.add_argument("-cv", "--convertor_models_dir", type=Path, default="ppg2mel/saved_models", 
                         help="Directory containing saved convert models")
     parser.add_argument("--cpu", action="store_true", help=\
         "If True, processing is done on CPU, even when a GPU is available.")

docker-compose.yml

@@ -1,23 +0,0 @@
-version: '3.8'
-
-services:
-  server:
-    image: mockingbird:latest
-    build: .
-    volumes:
-      - ./datasets:/datasets
-      - ./synthesizer/saved_models:/workspace/synthesizer/saved_models
-    environment:
-      - DATASET_MIRROR=US
-      - FORCE_RETRAIN=false
-      - TRAIN_DATASETS=aidatatang_200zh magicdata aishell3 data_aishell
-      - TRAIN_SKIP_EXISTING=true
-    ports:
-      - 8080:8080
-    deploy:
-      resources:
-        reservations:
-          devices:
-            - driver: nvidia
-              device_ids: [ '0' ]
-              capabilities: [ gpu ]

docker-entrypoint.sh

@@ -1,17 +0,0 @@
-#!/usr/bin/env bash
-
-if [ -z "$(ls -A /workspace/synthesizer/saved_models)" ] || [ "$FORCE_RETRAIN" = true ] ; then
-    /workspace/datasets_download/download.sh
-    /workspace/datasets_download/extract.sh
-    for DATASET in ${TRAIN_DATASETS}
-    do
-        if [ "$TRAIN_SKIP_EXISTING" = true ] ; then
-            python pre.py /datasets -d ${DATASET} -n $(nproc) --skip_existing
-        else
-            python pre.py /datasets -d ${DATASET} -n $(nproc)
-        fi
-    done
-    python synthesizer_train.py mandarin /datasets/SV2TTS/synthesizer
-fi
-
-python web.py

encoder/audio.py

@@ -1,5 +1,5 @@
 from scipy.ndimage.morphology import binary_dilation
-from models.encoder.params_data import *
+from encoder.params_data import *
 from pathlib import Path
 from typing import Optional, Union
 from warnings import warn
@@ -56,8 +56,8 @@ def wav_to_mel_spectrogram(wav):
     Note: this not a log-mel spectrogram.
     """
     frames = librosa.feature.melspectrogram(
-        y=wav,
+        wav,
-        sr=sampling_rate,
+        sampling_rate,
         n_fft=int(sampling_rate * mel_window_length / 1000),
         hop_length=int(sampling_rate * mel_window_step / 1000),
         n_mels=mel_n_channels

encoder/data_objects/__init__.py

@@ -0,0 +1,2 @@
+from encoder.data_objects.speaker_verification_dataset import SpeakerVerificationDataset
+from encoder.data_objects.speaker_verification_dataset import SpeakerVerificationDataLoader

encoder/data_objects/speaker.py

@@ -1,5 +1,5 @@
-from models.encoder.data_objects.random_cycler import RandomCycler
+from encoder.data_objects.random_cycler import RandomCycler
-from models.encoder.data_objects.utterance import Utterance
+from encoder.data_objects.utterance import Utterance
 from pathlib import Path
 
 # Contains the set of utterances of a single speaker

encoder/data_objects/speaker_batch.py

@@ -1,6 +1,6 @@
 import numpy as np
 from typing import List
-from models.encoder.data_objects.speaker import Speaker
+from encoder.data_objects.speaker import Speaker
 
 class SpeakerBatch:
     def __init__(self, speakers: List[Speaker], utterances_per_speaker: int, n_frames: int):

encoder/data_objects/speaker_verification_dataset.py

@@ -1,7 +1,7 @@
-from models.encoder.data_objects.random_cycler import RandomCycler
+from encoder.data_objects.random_cycler import RandomCycler
-from models.encoder.data_objects.speaker_batch import SpeakerBatch
+from encoder.data_objects.speaker_batch import SpeakerBatch
-from models.encoder.data_objects.speaker import Speaker
+from encoder.data_objects.speaker import Speaker
-from models.encoder.params_data import partials_n_frames
+from encoder.params_data import partials_n_frames
 from torch.utils.data import Dataset, DataLoader
 from pathlib import Path
 

encoder/inference.py

@@ -1,8 +1,8 @@
-from models.encoder.params_data import *
+from encoder.params_data import *
-from models.encoder.model import SpeakerEncoder
+from encoder.model import SpeakerEncoder
-from models.encoder.audio import preprocess_wav   # We want to expose this function from here
+from encoder.audio import preprocess_wav   # We want to expose this function from here
 from matplotlib import cm
-from models.encoder import audio
+from encoder import audio
 from pathlib import Path
 import matplotlib.pyplot as plt
 import numpy as np

encoder/model.py

@@ -1,5 +1,5 @@
-from models.encoder.params_model import *
+from encoder.params_model import *
-from models.encoder.params_data import *
+from encoder.params_data import *
 from scipy.interpolate import interp1d
 from sklearn.metrics import roc_curve
 from torch.nn.utils import clip_grad_norm_

encoder/preprocess.py

@@ -1,8 +1,8 @@
 from multiprocess.pool import ThreadPool
-from models.encoder.params_data import *
+from encoder.params_data import *
-from models.encoder.config import librispeech_datasets, anglophone_nationalites
+from encoder.config import librispeech_datasets, anglophone_nationalites
 from datetime import datetime
-from models.encoder import audio
+from encoder import audio
 from pathlib import Path
 from tqdm import tqdm
 import numpy as np
@@ -22,7 +22,7 @@ class DatasetLog:
         self._log_params()
         
     def _log_params(self):
-        from models.encoder import params_data
+        from encoder import params_data
         self.write_line("Parameter values:")
         for param_name in (p for p in dir(params_data) if not p.startswith("__")):
             value = getattr(params_data, param_name)

encoder/train.py

@@ -1,7 +1,7 @@
-from models.encoder.visualizations import Visualizations
+from encoder.visualizations import Visualizations
-from models.encoder.data_objects import SpeakerVerificationDataLoader, SpeakerVerificationDataset
+from encoder.data_objects import SpeakerVerificationDataLoader, SpeakerVerificationDataset
-from models.encoder.params_model import *
+from encoder.params_model import *
-from models.encoder.model import SpeakerEncoder
+from encoder.model import SpeakerEncoder
 from utils.profiler import Profiler
 from pathlib import Path
 import torch

encoder/visualizations.py

@@ -1,4 +1,4 @@
-from models.encoder.data_objects.speaker_verification_dataset import SpeakerVerificationDataset
+from encoder.data_objects.speaker_verification_dataset import SpeakerVerificationDataset
 from datetime import datetime
 from time import perf_counter as timer
 import matplotlib.pyplot as plt
@@ -65,8 +65,8 @@ class Visualizations:
     def log_params(self):
         if self.disabled:
             return 
-        from models.encoder import params_data
+        from encoder import params_data
-        from models.encoder import params_model
+        from encoder import params_model
         param_string = "<b>Model parameters</b>:<br>"
         for param_name in (p for p in dir(params_model) if not p.startswith("__")):
             value = getattr(params_model, param_name)

encoder_preprocess.py

@@ -1,10 +1,7 @@
-import argparse
+from encoder.preprocess import preprocess_librispeech, preprocess_voxceleb1, preprocess_voxceleb2, preprocess_aidatatang_200zh
-from pathlib import Path
-
-from models.encoder.preprocess import (preprocess_aidatatang_200zh,
-                                preprocess_librispeech, preprocess_voxceleb1,
-                                preprocess_voxceleb2)
 from utils.argutils import print_args
+from pathlib import Path
+import argparse
 
 if __name__ == "__main__":
     class MyFormatter(argparse.ArgumentDefaultsHelpFormatter, argparse.RawDescriptionHelpFormatter):

encoder_train.py

@@ -1,5 +1,5 @@
 from utils.argutils import print_args
-from models.encoder.train import train
+from encoder.train import train
 from pathlib import Path
 import argparse
 

gen_voice.py

@@ -1,120 +0,0 @@
-from models.synthesizer.inference import Synthesizer
-from models.encoder import inference as encoder
-from models.vocoder.hifigan import inference as gan_vocoder
-from pathlib import Path
-import numpy as np
-import soundfile as sf
-import torch
-import sys
-import os
-import re
-import cn2an
-
-vocoder = gan_vocoder
-
-def gen_one_wav(synthesizer, in_fpath, embed, texts, file_name, seq):
-    embeds = [embed] * len(texts)
-    # If you know what the attention layer alignments are, you can retrieve them here by
-    # passing return_alignments=True
-    specs = synthesizer.synthesize_spectrograms(texts, embeds, style_idx=-1, min_stop_token=4, steps=400)
-    #spec = specs[0]
-    breaks = [spec.shape[1] for spec in specs]
-    spec = np.concatenate(specs, axis=1)
-
-    # If seed is specified, reset torch seed and reload vocoder
-    # Synthesizing the waveform is fairly straightforward. Remember that the longer the
-    # spectrogram, the more time-efficient the vocoder.
-    generated_wav, output_sample_rate = vocoder.infer_waveform(spec)
-    
-    # Add breaks
-    b_ends = np.cumsum(np.array(breaks) * synthesizer.hparams.hop_size)
-    b_starts = np.concatenate(([0], b_ends[:-1]))
-    wavs = [generated_wav[start:end] for start, end, in zip(b_starts, b_ends)]
-    breaks = [np.zeros(int(0.15 * synthesizer.sample_rate))] * len(breaks)
-    generated_wav = np.concatenate([i for w, b in zip(wavs, breaks) for i in (w, b)])
-    
-    ## Post-generation
-    # There's a bug with sounddevice that makes the audio cut one second earlier, so we
-    # pad it.
-
-    # Trim excess silences to compensate for gaps in spectrograms (issue #53)
-    generated_wav = encoder.preprocess_wav(generated_wav)
-    generated_wav = generated_wav / np.abs(generated_wav).max() * 0.97
-        
-    # Save it on the disk
-    model=os.path.basename(in_fpath)
-    filename = "%s_%d_%s.wav" %(file_name, seq, model)
-    sf.write(filename, generated_wav, synthesizer.sample_rate)
-
-    print("\nSaved output as %s\n\n" % filename)
-    
-    
-def generate_wav(enc_model_fpath, syn_model_fpath, voc_model_fpath, in_fpath, input_txt, file_name): 
-    if torch.cuda.is_available():
-        device_id = torch.cuda.current_device()
-        gpu_properties = torch.cuda.get_device_properties(device_id)
-        ## Print some environment information (for debugging purposes)
-        print("Found %d GPUs available. Using GPU %d (%s) of compute capability %d.%d with "
-            "%.1fGb total memory.\n" % 
-            (torch.cuda.device_count(),
-            device_id,
-            gpu_properties.name,
-            gpu_properties.major,
-            gpu_properties.minor,
-            gpu_properties.total_memory / 1e9))
-    else:
-        print("Using CPU for inference.\n")
-
-    print("Preparing the encoder, the synthesizer and the vocoder...")
-    encoder.load_model(enc_model_fpath)
-    synthesizer = Synthesizer(syn_model_fpath)
-    vocoder.load_model(voc_model_fpath)
-
-    encoder_wav = synthesizer.load_preprocess_wav(in_fpath)
-    embed, partial_embeds, _ = encoder.embed_utterance(encoder_wav, return_partials=True)
-
-    texts = input_txt.split("\n")
-    seq=0
-    each_num=1500
-    
-    punctuation = '！，。、,' # punctuate and split/clean text
-    processed_texts = []
-    cur_num = 0
-    for text in texts:
-      for processed_text in re.sub(r'[{}]+'.format(punctuation), '\n', text).split('\n'):
-        if processed_text:
-            processed_texts.append(processed_text.strip())
-            cur_num += len(processed_text.strip())
-      if cur_num > each_num:
-        seq = seq +1
-        gen_one_wav(synthesizer, in_fpath, embed, processed_texts, file_name, seq)
-        processed_texts = []
-        cur_num = 0
-    
-    if len(processed_texts)>0:
-      seq = seq +1
-      gen_one_wav(synthesizer, in_fpath, embed, processed_texts, file_name, seq)
-
-if (len(sys.argv)>=3):
-    my_txt = ""
-    print("reading from :", sys.argv[1])
-    with open(sys.argv[1], "r") as f:
-        for line in f.readlines():
-            #line = line.strip('\n')
-            my_txt += line
-    txt_file_name = sys.argv[1]
-    wav_file_name = sys.argv[2]
-
-    output = cn2an.transform(my_txt, "an2cn")
-    print(output)
-    generate_wav(
-    Path("encoder/saved_models/pretrained.pt"),
-    Path("synthesizer/saved_models/mandarin.pt"),
-    Path("vocoder/saved_models/pretrained/g_hifigan.pt"), wav_file_name, output, txt_file_name
-    )
-
-else:
-    print("please input the file name")
-    exit(1)
-
-

mkgui/app.py

@@ -1,27 +1,26 @@
+from asyncio.windows_events import NULL
 from pydantic import BaseModel, Field
 import os
 from pathlib import Path
 from enum import Enum
-from models.encoder import inference as encoder
+from encoder import inference as encoder
 import librosa
 from scipy.io.wavfile import write
 import re
 import numpy as np
-from control.mkgui.base.components.types import FileContent
+from mkgui.base.components.types import FileContent
-from models.vocoder.hifigan import inference as gan_vocoder
+from vocoder.hifigan import inference as gan_vocoder
-from models.synthesizer.inference import Synthesizer
+from synthesizer.inference import Synthesizer
-from typing import Any, Tuple
+from typing import Any
 import matplotlib.pyplot as plt
 
 # Constants
-AUDIO_SAMPLES_DIR = f"data{os.sep}samples{os.sep}"
+AUDIO_SAMPLES_DIR = 'samples\\'
-SYN_MODELS_DIRT = f"data{os.sep}ckpt{os.sep}synthesizer"
+SYN_MODELS_DIRT = "synthesizer\\saved_models"
-ENC_MODELS_DIRT = f"data{os.sep}ckpt{os.sep}encoder"
+ENC_MODELS_DIRT = "encoder\\saved_models"
-VOC_MODELS_DIRT = f"data{os.sep}ckpt{os.sep}vocoder"
+VOC_MODELS_DIRT = "vocoder\\saved_models"
-TEMP_SOURCE_AUDIO = f"wavs{os.sep}temp_source.wav"
+TEMP_SOURCE_AUDIO = "wavs/temp_source.wav"
-TEMP_RESULT_AUDIO = f"wavs{os.sep}temp_result.wav"
+TEMP_RESULT_AUDIO = "wavs/temp_result.wav"
-if not os.path.isdir("wavs"):
-    os.makedirs("wavs")
 
 # Load local sample audio as options TODO: load dataset 
 if os.path.isdir(AUDIO_SAMPLES_DIR):
@@ -31,7 +30,7 @@ if os.path.isdir(SYN_MODELS_DIRT):
     synthesizers =  Enum('synthesizers', list((file.name, file) for file in Path(SYN_MODELS_DIRT).glob("**/*.pt")))
     print("Loaded synthesizer models: " + str(len(synthesizers)))
 else:
-    raise Exception(f"Model folder {SYN_MODELS_DIRT} doesn't exist. 请将模型文件位置移动到上述位置中进行重试！")
+    raise Exception(f"Model folder {SYN_MODELS_DIRT} doesn't exist.")
 
 if os.path.isdir(ENC_MODELS_DIRT):    
     encoders =  Enum('encoders', list((file.name, file) for file in Path(ENC_MODELS_DIRT).glob("**/*.pt")))
@@ -51,11 +50,9 @@ class Input(BaseModel):
         ..., example="欢迎使用工具箱, 现已支持中文输入！", alias="文本内容"
     )
     local_audio_file: audio_input_selection = Field(
-        ..., alias="选择语音（本地wav）",
+        ..., alias="输入语音（本地wav）",
         description="选择本地语音文件."
     )
-    record_audio_file: FileContent = Field(default=None, alias="录制语音",
-        description="录音.", is_recorder=True, mime_type="audio/wav")
     upload_audio_file: FileContent = Field(default=None, alias="或上传语音",
         description="拖拽或点击上传.", mime_type="audio/wav")
     encoder: encoders = Field(
@@ -76,7 +73,7 @@ class AudioEntity(BaseModel):
     mel: Any
 
 class Output(BaseModel):
-    __root__: Tuple[AudioEntity, AudioEntity]
+    __root__: tuple[AudioEntity, AudioEntity]
 
     def render_output_ui(self, streamlit_app, input) -> None:  # type: ignore
         """Custom output UI.
@@ -105,12 +102,7 @@ def synthesize(input: Input) -> Output:
     gan_vocoder.load_model(Path(input.vocoder.value))
 
     # load file
-    if input.record_audio_file != None:
+    if input.upload_audio_file != None:
-        with open(TEMP_SOURCE_AUDIO, "w+b") as f:
-            f.write(input.record_audio_file.as_bytes())
-            f.seek(0)
-        wav, sample_rate = librosa.load(TEMP_SOURCE_AUDIO)
-    elif input.upload_audio_file != None:
         with open(TEMP_SOURCE_AUDIO, "w+b") as f:
             f.write(input.upload_audio_file.as_bytes())
             f.seek(0)

mkgui/app_vc.py

@@ -1,29 +1,31 @@
-import os
+from asyncio.windows_events import NULL
-from enum import Enum
+from synthesizer.inference import Synthesizer
-from pathlib import Path
-from typing import Any, Tuple
-
-import librosa
-import matplotlib.pyplot as plt
-import torch
 from pydantic import BaseModel, Field
+from encoder import inference as speacker_encoder
+import torch
+import os
+from pathlib import Path
+from enum import Enum
+import ppg_extractor as Extractor
+import ppg2mel as Convertor
+import librosa
 from scipy.io.wavfile import write
+import re
+import numpy as np
+from mkgui.base.components.types import FileContent
+from vocoder.hifigan import inference as gan_vocoder
+from typing import Any
+import matplotlib.pyplot as plt
 
-import models.ppg2mel as Convertor
-import models.ppg_extractor as Extractor
-from control.mkgui.base.components.types import FileContent
-from models.encoder import inference as speacker_encoder
-from models.synthesizer.inference import Synthesizer
-from models.vocoder.hifigan import inference as gan_vocoder
 
 # Constants
-AUDIO_SAMPLES_DIR = f'data{os.sep}samples{os.sep}'
+AUDIO_SAMPLES_DIR = 'samples\\'
-EXT_MODELS_DIRT = f'data{os.sep}ckpt{os.sep}ppg_extractor'
+EXT_MODELS_DIRT = "ppg_extractor\\saved_models"
-CONV_MODELS_DIRT = f'data{os.sep}ckpt{os.sep}ppg2mel'
+CONV_MODELS_DIRT = "ppg2mel\\saved_models"
-VOC_MODELS_DIRT = f'data{os.sep}ckpt{os.sep}vocoder'
+VOC_MODELS_DIRT = "vocoder\\saved_models"
-TEMP_SOURCE_AUDIO = f'wavs{os.sep}temp_source.wav'
+TEMP_SOURCE_AUDIO = "wavs/temp_source.wav"
-TEMP_TARGET_AUDIO = f'wavs{os.sep}temp_target.wav'
+TEMP_TARGET_AUDIO = "wavs/temp_target.wav"
-TEMP_RESULT_AUDIO = f'wavs{os.sep}temp_result.wav'
+TEMP_RESULT_AUDIO = "wavs/temp_result.wav"
 
 # Load local sample audio as options TODO: load dataset 
 if os.path.isdir(AUDIO_SAMPLES_DIR):
@@ -69,7 +71,7 @@ class Input(BaseModel):
         description="选择语音转换模型文件."
     )
     vocoder: vocoders = Field(
-        ..., alias="语音解码模型", 
+        ..., alias="语音编码模型", 
         description="选择语音解码模型文件(目前只支持HifiGan类型)."
     )
 
@@ -78,7 +80,7 @@ class AudioEntity(BaseModel):
     mel: Any
 
 class Output(BaseModel):
-    __root__: Tuple[AudioEntity, AudioEntity, AudioEntity]
+    __root__: tuple[AudioEntity, AudioEntity, AudioEntity]
 
     def render_output_ui(self, streamlit_app, input) -> None:  # type: ignore
         """Custom output UI.
@@ -131,10 +133,9 @@ def convert(input: Input) -> Output:
 
     ppg = extractor.extract_from_wav(src_wav)
     # Import necessary dependency of Voice Conversion
-    from utils.f0_utils import (compute_f0, compute_mean_std, f02lf0,
+    from utils.f0_utils import compute_f0, f02lf0, compute_mean_std, get_converted_lf0uv   
-                                get_converted_lf0uv)   
     ref_lf0_mean, ref_lf0_std = compute_mean_std(f02lf0(compute_f0(ref_wav)))
-    speacker_encoder.load_model(Path(f"data{os.sep}ckpt{os.sep}encoder{os.sep}pretrained_bak_5805000.pt"))
+    speacker_encoder.load_model(Path("encoder/saved_models/pretrained_bak_5805000.pt"))
     embed = speacker_encoder.embed_utterance(ref_wav)
     lf0_uv = get_converted_lf0uv(src_wav, ref_lf0_mean, ref_lf0_std, convert=True)
     min_len = min(ppg.shape[1], len(lf0_uv))

mkgui/base/ui/schema_utils.py

@@ -37,12 +37,6 @@ def is_single_file_property(property: Dict) -> bool:
     # TODO: binary?
     return property.get("format") == "byte"
 
-def is_single_autio_property(property: Dict) -> bool:
-    if property.get("type") != "string":
-        return False
-    # TODO: binary?
-    return property.get("format") == "bytes"
-
 
 def is_single_directory_property(property: Dict) -> bool:
     if property.get("type") != "string":

mkgui/base/ui/streamlit_ui.py

@@ -2,7 +2,7 @@ import datetime
 import inspect
 import mimetypes
 import sys
-from os import getcwd, unlink, path
+from os import getcwd, unlink
 from platform import system
 from tempfile import NamedTemporaryFile
 from typing import Any, Callable, Dict, List, Type
@@ -14,13 +14,14 @@ from fastapi.encoders import jsonable_encoder
 from loguru import logger
 from pydantic import BaseModel, ValidationError, parse_obj_as
 
-from control.mkgui.base import Opyrator
+from mkgui.base import Opyrator
-from control.mkgui.base.core import name_to_title
+from mkgui.base.core import name_to_title
-from . import schema_utils
+from mkgui.base.ui import schema_utils
-from .streamlit_utils import CUSTOM_STREAMLIT_CSS
+from mkgui.base.ui.streamlit_utils import CUSTOM_STREAMLIT_CSS
 
 STREAMLIT_RUNNER_SNIPPET = """
-from control.mkgui.base.ui import render_streamlit_ui
+from mkgui.base.ui import render_streamlit_ui
+from mkgui.base import Opyrator
 
 import streamlit as st
 
@@ -243,13 +244,6 @@ class InputUI:
         if "mime_type" in property:
             file_extension = mimetypes.guess_extension(property["mime_type"])
 
-        if "is_recorder" in property:
-            from audio_recorder_streamlit import audio_recorder
-            audio_bytes = audio_recorder()
-            if audio_bytes:
-                streamlit_app.audio(audio_bytes, format="audio/wav")
-            return audio_bytes
-        
         uploaded_file = streamlit_app.file_uploader(
             **streamlit_kwargs, accept_multiple_files=False, type=file_extension
         )
@@ -269,39 +263,6 @@ class InputUI:
                 streamlit_app.video(bytes, format=property.get("mime_type"))
         return bytes
 
-    def _render_single_audio_input(
-            self, streamlit_app: st, key: str, property: Dict
-        ) -> Any:
-            # streamlit_kwargs = self._get_default_streamlit_input_kwargs(key, property)
-            from audio_recorder_streamlit import audio_recorder
-            audio_bytes = audio_recorder()
-            if audio_bytes:
-                streamlit_app.audio(audio_bytes, format="audio/wav")
-            return audio_bytes
-
-            # file_extension = None
-            # if "mime_type" in property:
-            #     file_extension = mimetypes.guess_extension(property["mime_type"])
-
-            # uploaded_file = streamlit_app.file_uploader(
-            #     **streamlit_kwargs, accept_multiple_files=False, type=file_extension
-            # )
-            # if uploaded_file is None:
-            #     return None
-
-            # bytes = uploaded_file.getvalue()
-            # if property.get("mime_type"):
-            #     if is_compatible_audio(property["mime_type"]):
-            #         # Show audio
-            #         streamlit_app.audio(bytes, format=property.get("mime_type"))
-            #     if is_compatible_image(property["mime_type"]):
-            #         # Show image
-            #         streamlit_app.image(bytes)
-            #     if is_compatible_video(property["mime_type"]):
-            #         # Show video
-            #         streamlit_app.video(bytes, format=property.get("mime_type"))
-            # return bytes
-
     def _render_single_string_input(
         self, streamlit_app: st, key: str, property: Dict
     ) -> Any:
@@ -846,20 +807,18 @@ class OutputUI:
 
 def getOpyrator(mode: str) -> Opyrator:
     if mode == None or mode.startswith('VC'):
-        from control.mkgui.app_vc import convert
+        from mkgui.app_vc import convert
         return  Opyrator(convert)
     if mode == None or mode.startswith('预处理'):
-        from control.mkgui.preprocess import preprocess
+        from mkgui.preprocess import preprocess
         return  Opyrator(preprocess)
     if mode == None or mode.startswith('模型训练'):
-        from control.mkgui.train import train
+        from mkgui.train import train
         return  Opyrator(train)
-    if mode == None or mode.startswith('模型训练(VC)'):
+    from mkgui.app import synthesize
-        from control.mkgui.train_vc import train_vc
-        return  Opyrator(train_vc)
-    from control.mkgui.app import synthesize
     return Opyrator(synthesize)
     
+
 def render_streamlit_ui() -> None:
     # init
     session_state = st.session_state
@@ -870,7 +829,7 @@ def render_streamlit_ui() -> None:
     with st.spinner("Loading MockingBird GUI. Please wait..."):
         session_state.mode = st.sidebar.selectbox(
             '模式选择', 
-            ( "AI拟音", "VC拟音", "预处理", "模型训练", "模型训练(VC)")
+            ( "AI拟音", "VC拟音", "预处理", "模型训练")
         )
         if "mode" in session_state:
             mode = session_state.mode
@@ -883,7 +842,7 @@ def render_streamlit_ui() -> None:
     col2.title(title)
     col2.markdown("欢迎使用MockingBird Web 2")
 
-    image = Image.open(path.join('control','mkgui', 'static', 'mb.png'))
+    image = Image.open('.\\mkgui\\static\\mb.png')
     col1.image(image)
 
     st.markdown("---")
@@ -891,13 +850,6 @@ def render_streamlit_ui() -> None:
 
     with left:
         st.header("Control 控制")
-        # if session_state.mode in ["AI拟音", "VC拟音"] :
-            # from audiorecorder import audiorecorder
-            # audio = audiorecorder("Click to record", "Recording...")
-            # if len(audio) > 0:
-            #     # To play audio in frontend:
-            #     st.audio(audio.tobytes())
-            
         InputUI(session_state=session_state, input_class=opyrator.input_type).render_ui(st)
         execute_selected = st.button(opyrator.action)
         if execute_selected:

mkgui/preprocess.py

@@ -2,12 +2,12 @@ from pydantic import BaseModel, Field
 import os
 from pathlib import Path
 from enum import Enum
-from typing import Any, Tuple
+from typing import Any
 
 
 # Constants
-EXT_MODELS_DIRT = f"data{os.sep}ckpt{os.sep}ppg_extractor"
+EXT_MODELS_DIRT = "ppg_extractor\\saved_models"
-ENC_MODELS_DIRT = f"data{os.sep}ckpt{os.sep}encoder"
+ENC_MODELS_DIRT = "encoder\\saved_models"
 
 
 if os.path.isdir(EXT_MODELS_DIRT):    
@@ -70,7 +70,7 @@ class AudioEntity(BaseModel):
     mel: Any
 
 class Output(BaseModel):
-    __root__: Tuple[str, int]
+    __root__: tuple[str, int]
 
     def render_output_ui(self, streamlit_app, input) -> None:  # type: ignore
         """Custom output UI.
@@ -83,7 +83,7 @@ def preprocess(input: Input) -> Output:
     """Preprocess(预处理)"""
     finished = 0
     if input.model == Model.VC_PPG2MEL:
-        from models.ppg2mel.preprocess import preprocess_dataset
+        from ppg2mel.preprocess import preprocess_dataset
         finished = preprocess_dataset(
             datasets_root=Path(input.datasets_root),
             dataset=input.dataset,

mkgui/train.py

@@ -2,16 +2,17 @@ from pydantic import BaseModel, Field
 import os
 from pathlib import Path
 from enum import Enum
-from typing import Any, Tuple
+from typing import Any
 import numpy as np
-from utils.hparams import HpsYaml
+from utils.load_yaml import HpsYaml
 from utils.util import AttrDict
 import torch
 
+# TODO: seperator for *unix systems
 # Constants
-EXT_MODELS_DIRT = f"data{os.sep}ckpt{os.sep}ppg_extractor"
+EXT_MODELS_DIRT = "ppg_extractor\\saved_models"
-CONV_MODELS_DIRT = f"data{os.sep}ckpt{os.sep}ppg2mel"
+CONV_MODELS_DIRT = "ppg2mel\\saved_models"
-ENC_MODELS_DIRT = f"data{os.sep}ckpt{os.sep}encoder"
+ENC_MODELS_DIRT = "encoder\\saved_models"
 
 
 if os.path.isdir(EXT_MODELS_DIRT):    
@@ -101,7 +102,7 @@ class AudioEntity(BaseModel):
     mel: Any
 
 class Output(BaseModel):
-    __root__: Tuple[str, int]
+    __root__: tuple[str, int]
 
     def render_output_ui(self, streamlit_app, input) -> None:  # type: ignore
         """Custom output UI.
@@ -110,8 +111,8 @@ class Output(BaseModel):
         sr, count = self.__root__
         streamlit_app.subheader(f"Dataset {sr} done processed total of {count}")
 
-def train_vc(input: Input) -> Output:
+def train(input: Input) -> Output:
-    """Train VC(训练 VC)"""
+    """Train(训练)"""
 
     print(">>> OneShot VC training ...")
     params = AttrDict()
@@ -144,7 +145,7 @@ def train_vc(input: Input) -> Output:
     if torch.cuda.is_available():
         torch.cuda.manual_seed_all(input.seed)
     mode = "train"
-    from models.ppg2mel.train.train_linglf02mel_seq2seq_oneshotvc import Solver
+    from ppg2mel.train.train_linglf02mel_seq2seq_oneshotvc import Solver
     solver = Solver(config, params, mode)
     solver.load_data()
     solver.set_model()

models/encoder/data_objects/__init__.py

@@ -1,2 +0,0 @@
-from models.encoder.data_objects.speaker_verification_dataset import SpeakerVerificationDataset
-from models.encoder.data_objects.speaker_verification_dataset import SpeakerVerificationDataLoader

models/synthesizer/models/base.py

@@ -1,77 +0,0 @@
-import torch
-import torch.nn as nn
-import imp
-import numpy as np
-
-class Base(nn.Module):
-    def __init__(self, stop_threshold):
-        super().__init__()
-
-        self.init_model()
-        self.num_params()
-
-        self.register_buffer("step", torch.zeros(1, dtype=torch.long))
-        self.register_buffer("stop_threshold", torch.tensor(stop_threshold, dtype=torch.float32))
-
-    @property
-    def r(self):
-        return self.decoder.r.item()
-
-    @r.setter
-    def r(self, value):
-        self.decoder.r = self.decoder.r.new_tensor(value, requires_grad=False)
-
-    def init_model(self):
-        for p in self.parameters():
-            if p.dim() > 1: nn.init.xavier_uniform_(p)
-
-    def finetune_partial(self, whitelist_layers):
-        self.zero_grad()
-        for name, child in self.named_children():
-            if name in whitelist_layers:
-                print("Trainable Layer: %s" % name)
-                print("Trainable Parameters: %.3f" % sum([np.prod(p.size()) for p in child.parameters()]))
-                for param in child.parameters():
-                    param.requires_grad = False
-
-    def get_step(self):
-        return self.step.data.item()
-
-    def reset_step(self):
-        # assignment to parameters or buffers is overloaded, updates internal dict entry
-        self.step = self.step.data.new_tensor(1)
-
-    def log(self, path, msg):
-        with open(path, "a") as f:
-            print(msg, file=f)
-
-    def load(self, path, device, optimizer=None):
-        # Use device of model params as location for loaded state
-        checkpoint = torch.load(str(path), map_location=device)
-        if "model_state" in checkpoint:
-            state = checkpoint["model_state"]
-        else:
-            state = checkpoint["model"]
-        self.load_state_dict(state, strict=False)
-
-        if "optimizer_state" in checkpoint and optimizer is not None:
-            optimizer.load_state_dict(checkpoint["optimizer_state"])
-
-    def save(self, path, optimizer=None):
-        if optimizer is not None:
-            torch.save({
-                "model_state": self.state_dict(),
-                "optimizer_state": optimizer.state_dict(),
-            }, str(path))
-        else:
-            torch.save({
-                "model_state": self.state_dict(),
-            }, str(path))
-
-
-    def num_params(self, print_out=True):
-        parameters = filter(lambda p: p.requires_grad, self.parameters())
-        parameters = sum([np.prod(p.size()) for p in parameters]) / 1_000_000
-        if print_out:
-            print("Trainable Parameters: %.3fM" % parameters)
-        return parameters

models/synthesizer/models/sublayer/cbhg.py

@@ -1,85 +0,0 @@
-import torch
-import torch.nn as nn
-from .common.batch_norm_conv import BatchNormConv
-from .common.highway_network import HighwayNetwork
-
-class CBHG(nn.Module):
-    def __init__(self, K, in_channels, channels, proj_channels, num_highways):
-        super().__init__()
-
-        # List of all rnns to call `flatten_parameters()` on
-        self._to_flatten = []
-
-        self.bank_kernels = [i for i in range(1, K + 1)]
-        self.conv1d_bank = nn.ModuleList()
-        for k in self.bank_kernels:
-            conv = BatchNormConv(in_channels, channels, k)
-            self.conv1d_bank.append(conv)
-
-        self.maxpool = nn.MaxPool1d(kernel_size=2, stride=1, padding=1)
-
-        self.conv_project1 = BatchNormConv(len(self.bank_kernels) * channels, proj_channels[0], 3)
-        self.conv_project2 = BatchNormConv(proj_channels[0], proj_channels[1], 3, relu=False)
-
-        # Fix the highway input if necessary
-        if proj_channels[-1] != channels:
-            self.highway_mismatch = True
-            self.pre_highway = nn.Linear(proj_channels[-1], channels, bias=False)
-        else:
-            self.highway_mismatch = False
-
-        self.highways = nn.ModuleList()
-        for i in range(num_highways):
-            hn = HighwayNetwork(channels)
-            self.highways.append(hn)
-
-        self.rnn = nn.GRU(channels, channels // 2, batch_first=True, bidirectional=True)
-        self._to_flatten.append(self.rnn)
-
-        # Avoid fragmentation of RNN parameters and associated warning
-        self._flatten_parameters()
-
-    def forward(self, x):
-        # Although we `_flatten_parameters()` on init, when using DataParallel
-        # the model gets replicated, making it no longer guaranteed that the
-        # weights are contiguous in GPU memory. Hence, we must call it again
-        self.rnn.flatten_parameters()
-
-        # Save these for later
-        residual = x
-        seq_len = x.size(-1)
-        conv_bank = []
-
-        # Convolution Bank
-        for conv in self.conv1d_bank:
-            c = conv(x) # Convolution
-            conv_bank.append(c[:, :, :seq_len])
-
-        # Stack along the channel axis
-        conv_bank = torch.cat(conv_bank, dim=1)
-
-        # dump the last padding to fit residual
-        x = self.maxpool(conv_bank)[:, :, :seq_len]
-
-        # Conv1d projections
-        x = self.conv_project1(x)
-        x = self.conv_project2(x)
-
-        # Residual Connect
-        x = x + residual
-
-        # Through the highways
-        x = x.transpose(1, 2)
-        if self.highway_mismatch is True:
-            x = self.pre_highway(x)
-        for h in self.highways: x = h(x)
-
-        # And then the RNN
-        x, _ = self.rnn(x)
-        return x
-
-    def _flatten_parameters(self):
-        """Calls `flatten_parameters` on all the rnns used by the WaveRNN. Used
-        to improve efficiency and avoid PyTorch yelling at us."""
-        [m.flatten_parameters() for m in self._to_flatten]
-

models/synthesizer/models/sublayer/common/batch_norm_conv.py

@@ -1,14 +0,0 @@
-import torch.nn as nn
-import torch.nn.functional as F
-
-class BatchNormConv(nn.Module):
-    def __init__(self, in_channels, out_channels, kernel, relu=True):
-        super().__init__()
-        self.conv = nn.Conv1d(in_channels, out_channels, kernel, stride=1, padding=kernel // 2, bias=False)
-        self.bnorm = nn.BatchNorm1d(out_channels)
-        self.relu = relu
-
-    def forward(self, x):
-        x = self.conv(x)
-        x = F.relu(x) if self.relu is True else x
-        return self.bnorm(x)

models/synthesizer/models/sublayer/common/highway_network.py

@@ -1,17 +0,0 @@
-import torch
-import torch.nn as nn
-import torch.nn.functional as F
-
-class HighwayNetwork(nn.Module):
-    def __init__(self, size):
-        super().__init__()
-        self.W1 = nn.Linear(size, size)
-        self.W2 = nn.Linear(size, size)
-        self.W1.bias.data.fill_(0.)
-
-    def forward(self, x):
-        x1 = self.W1(x)
-        x2 = self.W2(x)
-        g = torch.sigmoid(x2)
-        y = g * F.relu(x1) + (1. - g) * x
-        return y

models/synthesizer/models/sublayer/common/transforms.py

@@ -1,193 +0,0 @@
-import torch
-from torch.nn import functional as F
-
-import numpy as np
-
-
-DEFAULT_MIN_BIN_WIDTH = 1e-3
-DEFAULT_MIN_BIN_HEIGHT = 1e-3
-DEFAULT_MIN_DERIVATIVE = 1e-3
-
-
-def piecewise_rational_quadratic_transform(inputs, 
-                                           unnormalized_widths,
-                                           unnormalized_heights,
-                                           unnormalized_derivatives,
-                                           inverse=False,
-                                           tails=None, 
-                                           tail_bound=1.,
-                                           min_bin_width=DEFAULT_MIN_BIN_WIDTH,
-                                           min_bin_height=DEFAULT_MIN_BIN_HEIGHT,
-                                           min_derivative=DEFAULT_MIN_DERIVATIVE):
-
-    if tails is None:
-        spline_fn = rational_quadratic_spline
-        spline_kwargs = {}
-    else:
-        spline_fn = unconstrained_rational_quadratic_spline
-        spline_kwargs = {
-            'tails': tails,
-            'tail_bound': tail_bound
-        }
-
-    outputs, logabsdet = spline_fn(
-            inputs=inputs,
-            unnormalized_widths=unnormalized_widths,
-            unnormalized_heights=unnormalized_heights,
-            unnormalized_derivatives=unnormalized_derivatives,
-            inverse=inverse,
-            min_bin_width=min_bin_width,
-            min_bin_height=min_bin_height,
-            min_derivative=min_derivative,
-            **spline_kwargs
-    )
-    return outputs, logabsdet
-
-
-def searchsorted(bin_locations, inputs, eps=1e-6):
-    bin_locations[..., -1] += eps
-    return torch.sum(
-        inputs[..., None] >= bin_locations,
-        dim=-1
-    ) - 1
-
-
-def unconstrained_rational_quadratic_spline(inputs,
-                                            unnormalized_widths,
-                                            unnormalized_heights,
-                                            unnormalized_derivatives,
-                                            inverse=False,
-                                            tails='linear',
-                                            tail_bound=1.,
-                                            min_bin_width=DEFAULT_MIN_BIN_WIDTH,
-                                            min_bin_height=DEFAULT_MIN_BIN_HEIGHT,
-                                            min_derivative=DEFAULT_MIN_DERIVATIVE):
-    inside_interval_mask = (inputs >= -tail_bound) & (inputs <= tail_bound)
-    outside_interval_mask = ~inside_interval_mask
-
-    outputs = torch.zeros_like(inputs)
-    logabsdet = torch.zeros_like(inputs)
-
-    if tails == 'linear':
-        unnormalized_derivatives = F.pad(unnormalized_derivatives, pad=(1, 1))
-        constant = np.log(np.exp(1 - min_derivative) - 1)
-        unnormalized_derivatives[..., 0] = constant
-        unnormalized_derivatives[..., -1] = constant
-
-        outputs[outside_interval_mask] = inputs[outside_interval_mask]
-        logabsdet[outside_interval_mask] = 0
-    else:
-        raise RuntimeError('{} tails are not implemented.'.format(tails))
-
-    outputs[inside_interval_mask], logabsdet[inside_interval_mask] = rational_quadratic_spline(
-        inputs=inputs[inside_interval_mask],
-        unnormalized_widths=unnormalized_widths[inside_interval_mask, :],
-        unnormalized_heights=unnormalized_heights[inside_interval_mask, :],
-        unnormalized_derivatives=unnormalized_derivatives[inside_interval_mask, :],
-        inverse=inverse,
-        left=-tail_bound, right=tail_bound, bottom=-tail_bound, top=tail_bound,
-        min_bin_width=min_bin_width,
-        min_bin_height=min_bin_height,
-        min_derivative=min_derivative
-    )
-
-    return outputs, logabsdet
-
-def rational_quadratic_spline(inputs,
-                              unnormalized_widths,
-                              unnormalized_heights,
-                              unnormalized_derivatives,
-                              inverse=False,
-                              left=0., right=1., bottom=0., top=1.,
-                              min_bin_width=DEFAULT_MIN_BIN_WIDTH,
-                              min_bin_height=DEFAULT_MIN_BIN_HEIGHT,
-                              min_derivative=DEFAULT_MIN_DERIVATIVE):
-    if torch.min(inputs) < left or torch.max(inputs) > right:
-        raise ValueError('Input to a transform is not within its domain')
-
-    num_bins = unnormalized_widths.shape[-1]
-
-    if min_bin_width * num_bins > 1.0:
-        raise ValueError('Minimal bin width too large for the number of bins')
-    if min_bin_height * num_bins > 1.0:
-        raise ValueError('Minimal bin height too large for the number of bins')
-
-    widths = F.softmax(unnormalized_widths, dim=-1)
-    widths = min_bin_width + (1 - min_bin_width * num_bins) * widths
-    cumwidths = torch.cumsum(widths, dim=-1)
-    cumwidths = F.pad(cumwidths, pad=(1, 0), mode='constant', value=0.0)
-    cumwidths = (right - left) * cumwidths + left
-    cumwidths[..., 0] = left
-    cumwidths[..., -1] = right
-    widths = cumwidths[..., 1:] - cumwidths[..., :-1]
-
-    derivatives = min_derivative + F.softplus(unnormalized_derivatives)
-
-    heights = F.softmax(unnormalized_heights, dim=-1)
-    heights = min_bin_height + (1 - min_bin_height * num_bins) * heights
-    cumheights = torch.cumsum(heights, dim=-1)
-    cumheights = F.pad(cumheights, pad=(1, 0), mode='constant', value=0.0)
-    cumheights = (top - bottom) * cumheights + bottom
-    cumheights[..., 0] = bottom
-    cumheights[..., -1] = top
-    heights = cumheights[..., 1:] - cumheights[..., :-1]
-
-    if inverse:
-        bin_idx = searchsorted(cumheights, inputs)[..., None]
-    else:
-        bin_idx = searchsorted(cumwidths, inputs)[..., None]
-
-    input_cumwidths = cumwidths.gather(-1, bin_idx)[..., 0]
-    input_bin_widths = widths.gather(-1, bin_idx)[..., 0]
-
-    input_cumheights = cumheights.gather(-1, bin_idx)[..., 0]
-    delta = heights / widths
-    input_delta = delta.gather(-1, bin_idx)[..., 0]
-
-    input_derivatives = derivatives.gather(-1, bin_idx)[..., 0]
-    input_derivatives_plus_one = derivatives[..., 1:].gather(-1, bin_idx)[..., 0]
-
-    input_heights = heights.gather(-1, bin_idx)[..., 0]
-
-    if inverse:
-        a = (((inputs - input_cumheights) * (input_derivatives
-                                             + input_derivatives_plus_one
-                                             - 2 * input_delta)
-              + input_heights * (input_delta - input_derivatives)))
-        b = (input_heights * input_derivatives
-             - (inputs - input_cumheights) * (input_derivatives
-                                              + input_derivatives_plus_one
-                                              - 2 * input_delta))
-        c = - input_delta * (inputs - input_cumheights)
-
-        discriminant = b.pow(2) - 4 * a * c
-        assert (discriminant >= 0).all()
-
-        root = (2 * c) / (-b - torch.sqrt(discriminant))
-        outputs = root * input_bin_widths + input_cumwidths
-
-        theta_one_minus_theta = root * (1 - root)
-        denominator = input_delta + ((input_derivatives + input_derivatives_plus_one - 2 * input_delta)
-                                     * theta_one_minus_theta)
-        derivative_numerator = input_delta.pow(2) * (input_derivatives_plus_one * root.pow(2)
-                                                     + 2 * input_delta * theta_one_minus_theta
-                                                     + input_derivatives * (1 - root).pow(2))
-        logabsdet = torch.log(derivative_numerator) - 2 * torch.log(denominator)
-
-        return outputs, -logabsdet
-    else:
-        theta = (inputs - input_cumwidths) / input_bin_widths
-        theta_one_minus_theta = theta * (1 - theta)
-
-        numerator = input_heights * (input_delta * theta.pow(2)
-                                     + input_derivatives * theta_one_minus_theta)
-        denominator = input_delta + ((input_derivatives + input_derivatives_plus_one - 2 * input_delta)
-                                     * theta_one_minus_theta)
-        outputs = input_cumheights + numerator / denominator
-
-        derivative_numerator = input_delta.pow(2) * (input_derivatives_plus_one * theta.pow(2)
-                                                     + 2 * input_delta * theta_one_minus_theta
-                                                     + input_derivatives * (1 - theta).pow(2))
-        logabsdet = torch.log(derivative_numerator) - 2 * torch.log(denominator)
-
-        return outputs, logabsdet

models/synthesizer/models/sublayer/lsa.py

@@ -1,42 +0,0 @@
-import torch
-import torch.nn as nn
-import torch.nn.functional as F
-
-class LSA(nn.Module):
-    def __init__(self, attn_dim, kernel_size=31, filters=32):
-        super().__init__()
-        self.conv = nn.Conv1d(1, filters, padding=(kernel_size - 1) // 2, kernel_size=kernel_size, bias=True)
-        self.L = nn.Linear(filters, attn_dim, bias=False)
-        self.W = nn.Linear(attn_dim, attn_dim, bias=True) # Include the attention bias in this term
-        self.v = nn.Linear(attn_dim, 1, bias=False)
-        self.cumulative = None
-        self.attention = None
-
-    def init_attention(self, encoder_seq_proj):
-        device = encoder_seq_proj.device  # use same device as parameters
-        b, t, c = encoder_seq_proj.size()
-        self.cumulative = torch.zeros(b, t, device=device)
-        self.attention = torch.zeros(b, t, device=device)
-
-    def forward(self, encoder_seq_proj, query, times, chars):
-
-        if times == 0: self.init_attention(encoder_seq_proj)
-
-        processed_query = self.W(query).unsqueeze(1)
-
-        location = self.cumulative.unsqueeze(1)
-        processed_loc = self.L(self.conv(location).transpose(1, 2))
-
-        u = self.v(torch.tanh(processed_query + encoder_seq_proj + processed_loc))
-        u = u.squeeze(-1)
-
-        # Mask zero padding chars
-        u = u * (chars != 0).float()
-
-        # Smooth Attention
-        # scores = torch.sigmoid(u) / torch.sigmoid(u).sum(dim=1, keepdim=True)
-        scores = F.softmax(u, dim=1)
-        self.attention = scores
-        self.cumulative = self.cumulative + self.attention
-
-        return scores.unsqueeze(-1).transpose(1, 2)

models/synthesizer/models/sublayer/pre_net.py

@@ -1,27 +0,0 @@
-import torch.nn as nn
-import torch.nn.functional as F
-
-class PreNet(nn.Module):
-    def __init__(self, in_dims, fc1_dims=256, fc2_dims=128, dropout=0.5):
-        super().__init__()
-        self.fc1 = nn.Linear(in_dims, fc1_dims)
-        self.fc2 = nn.Linear(fc1_dims, fc2_dims)
-        self.p = dropout
-
-    def forward(self, x):
-        """forward
-
-        Args:
-            x (3D tensor with size `[batch_size, num_chars, tts_embed_dims]`): input texts list
-
-        Returns:
-            3D tensor with size `[batch_size, num_chars, encoder_dims]`
-            
-        """        
-        x = self.fc1(x)
-        x = F.relu(x)
-        x = F.dropout(x, self.p, training=True)
-        x = self.fc2(x)
-        x = F.relu(x)
-        x = F.dropout(x, self.p, training=True)
-        return x

models/synthesizer/models/sublayer/vits_modules.py

@@ -1,675 +0,0 @@
-import math
-import torch
-from torch import nn
-from torch.nn import functional as F
-
-from torch.nn import Conv1d
-from torch.nn.utils import weight_norm, remove_weight_norm
-from utils.util import init_weights, get_padding, convert_pad_shape, convert_pad_shape, subsequent_mask, fused_add_tanh_sigmoid_multiply
-from .common.transforms import piecewise_rational_quadratic_transform
-
-LRELU_SLOPE = 0.1
-
-class LayerNorm(nn.Module):
-  def __init__(self, channels, eps=1e-5):
-    super().__init__()
-    self.channels = channels
-    self.eps = eps
-
-    self.gamma = nn.Parameter(torch.ones(channels))
-    self.beta = nn.Parameter(torch.zeros(channels))
-
-  def forward(self, x):
-    x = x.transpose(1, -1)
-    x = F.layer_norm(x, (self.channels,), self.gamma, self.beta, self.eps)
-    return x.transpose(1, -1)
-
- 
-class ConvReluNorm(nn.Module):
-  def __init__(self, in_channels, hidden_channels, out_channels, kernel_size, n_layers, p_dropout):
-    super().__init__()
-    self.in_channels = in_channels
-    self.hidden_channels = hidden_channels
-    self.out_channels = out_channels
-    self.kernel_size = kernel_size
-    self.n_layers = n_layers
-    self.p_dropout = p_dropout
-    assert n_layers > 1, "Number of layers should be larger than 0."
-
-    self.conv_layers = nn.ModuleList()
-    self.norm_layers = nn.ModuleList()
-    self.conv_layers.append(nn.Conv1d(in_channels, hidden_channels, kernel_size, padding=kernel_size//2))
-    self.norm_layers.append(LayerNorm(hidden_channels))
-    self.relu_drop = nn.Sequential(
-        nn.ReLU(),
-        nn.Dropout(p_dropout))
-    for _ in range(n_layers-1):
-      self.conv_layers.append(nn.Conv1d(hidden_channels, hidden_channels, kernel_size, padding=kernel_size//2))
-      self.norm_layers.append(LayerNorm(hidden_channels))
-    self.proj = nn.Conv1d(hidden_channels, out_channels, 1)
-    self.proj.weight.data.zero_()
-    self.proj.bias.data.zero_()
-
-  def forward(self, x, x_mask):
-    x_org = x
-    for i in range(self.n_layers):
-      x = self.conv_layers[i](x * x_mask)
-      x = self.norm_layers[i](x)
-      x = self.relu_drop(x)
-    x = x_org + self.proj(x)
-    return x * x_mask
-
-
-class DDSConv(nn.Module):
-  """
-  Dilated and Depth-Separable Convolution
-  """
-  def __init__(self, channels, kernel_size, n_layers, p_dropout=0.):
-    super().__init__()
-    self.channels = channels
-    self.kernel_size = kernel_size
-    self.n_layers = n_layers
-    self.p_dropout = p_dropout
-
-    self.drop = nn.Dropout(p_dropout)
-    self.convs_sep = nn.ModuleList()
-    self.convs_1x1 = nn.ModuleList()
-    self.norms_1 = nn.ModuleList()
-    self.norms_2 = nn.ModuleList()
-    for i in range(n_layers):
-      dilation = kernel_size ** i
-      padding = (kernel_size * dilation - dilation) // 2
-      self.convs_sep.append(nn.Conv1d(channels, channels, kernel_size, 
-          groups=channels, dilation=dilation, padding=padding
-      ))
-      self.convs_1x1.append(nn.Conv1d(channels, channels, 1))
-      self.norms_1.append(LayerNorm(channels))
-      self.norms_2.append(LayerNorm(channels))
-
-  def forward(self, x, x_mask, g=None):
-    if g is not None:
-      x = x + g
-    for i in range(self.n_layers):
-      y = self.convs_sep[i](x * x_mask)
-      y = self.norms_1[i](y)
-      y = F.gelu(y)
-      y = self.convs_1x1[i](y)
-      y = self.norms_2[i](y)
-      y = F.gelu(y)
-      y = self.drop(y)
-      x = x + y
-    return x * x_mask
-
-
-class WN(torch.nn.Module):
-  def __init__(self, hidden_channels, kernel_size, dilation_rate, n_layers, gin_channels=0, p_dropout=0):
-    super(WN, self).__init__()
-    assert(kernel_size % 2 == 1)
-    self.hidden_channels =hidden_channels
-    self.kernel_size = kernel_size,
-    self.dilation_rate = dilation_rate
-    self.n_layers = n_layers
-    self.gin_channels = gin_channels
-    self.p_dropout = p_dropout
-
-    self.in_layers = torch.nn.ModuleList()
-    self.res_skip_layers = torch.nn.ModuleList()
-    self.drop = nn.Dropout(p_dropout)
-
-    if gin_channels != 0:
-      cond_layer = torch.nn.Conv1d(gin_channels, 2*hidden_channels*n_layers, 1)
-      self.cond_layer = torch.nn.utils.weight_norm(cond_layer, name='weight')
-
-    for i in range(n_layers):
-      dilation = dilation_rate ** i
-      padding = int((kernel_size * dilation - dilation) / 2)
-      in_layer = torch.nn.Conv1d(hidden_channels, 2*hidden_channels, kernel_size,
-                                 dilation=dilation, padding=padding)
-      in_layer = torch.nn.utils.weight_norm(in_layer, name='weight')
-      self.in_layers.append(in_layer)
-
-      # last one is not necessary
-      if i < n_layers - 1:
-        res_skip_channels = 2 * hidden_channels
-      else:
-        res_skip_channels = hidden_channels
-
-      res_skip_layer = torch.nn.Conv1d(hidden_channels, res_skip_channels, 1)
-      res_skip_layer = torch.nn.utils.weight_norm(res_skip_layer, name='weight')
-      self.res_skip_layers.append(res_skip_layer)
-
-  def forward(self, x, x_mask, g=None, **kwargs):
-    output = torch.zeros_like(x)
-    n_channels_tensor = torch.IntTensor([self.hidden_channels])
-
-    if g is not None:
-      g = self.cond_layer(g)
-
-    for i in range(self.n_layers):
-      x_in = self.in_layers[i](x)
-      if g is not None:
-        cond_offset = i * 2 * self.hidden_channels
-        g_l = g[:,cond_offset:cond_offset+2*self.hidden_channels,:]
-      else:
-        g_l = torch.zeros_like(x_in)
-
-      acts = fused_add_tanh_sigmoid_multiply(
-          x_in,
-          g_l,
-          n_channels_tensor)
-      acts = self.drop(acts)
-
-      res_skip_acts = self.res_skip_layers[i](acts)
-      if i < self.n_layers - 1:
-        res_acts = res_skip_acts[:,:self.hidden_channels,:]
-        x = (x + res_acts) * x_mask
-        output = output + res_skip_acts[:,self.hidden_channels:,:]
-      else:
-        output = output + res_skip_acts
-    return output * x_mask
-
-  def remove_weight_norm(self):
-    if self.gin_channels != 0:
-      torch.nn.utils.remove_weight_norm(self.cond_layer)
-    for l in self.in_layers:
-      torch.nn.utils.remove_weight_norm(l)
-    for l in self.res_skip_layers:
-     torch.nn.utils.remove_weight_norm(l)
-
-
-class ResBlock1(torch.nn.Module):
-    def __init__(self, channels, kernel_size=3, dilation=(1, 3, 5)):
-        super(ResBlock1, self).__init__()
-        self.convs1 = nn.ModuleList([
-            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=dilation[0],
-                               padding=get_padding(kernel_size, dilation[0]))),
-            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=dilation[1],
-                               padding=get_padding(kernel_size, dilation[1]))),
-            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=dilation[2],
-                               padding=get_padding(kernel_size, dilation[2])))
-        ])
-        self.convs1.apply(init_weights)
-
-        self.convs2 = nn.ModuleList([
-            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=1,
-                               padding=get_padding(kernel_size, 1))),
-            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=1,
-                               padding=get_padding(kernel_size, 1))),
-            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=1,
-                               padding=get_padding(kernel_size, 1)))
-        ])
-        self.convs2.apply(init_weights)
-
-    def forward(self, x, x_mask=None):
-        for c1, c2 in zip(self.convs1, self.convs2):
-            xt = F.leaky_relu(x, LRELU_SLOPE)
-            if x_mask is not None:
-                xt = xt * x_mask
-            xt = c1(xt)
-            xt = F.leaky_relu(xt, LRELU_SLOPE)
-            if x_mask is not None:
-                xt = xt * x_mask
-            xt = c2(xt)
-            x = xt + x
-        if x_mask is not None:
-            x = x * x_mask
-        return x
-
-    def remove_weight_norm(self):
-        for l in self.convs1:
-            remove_weight_norm(l)
-        for l in self.convs2:
-            remove_weight_norm(l)
-
-
-class ResBlock2(torch.nn.Module):
-    def __init__(self, channels, kernel_size=3, dilation=(1, 3)):
-        super(ResBlock2, self).__init__()
-        self.convs = nn.ModuleList([
-            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=dilation[0],
-                               padding=get_padding(kernel_size, dilation[0]))),
-            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=dilation[1],
-                               padding=get_padding(kernel_size, dilation[1])))
-        ])
-        self.convs.apply(init_weights)
-
-    def forward(self, x, x_mask=None):
-        for c in self.convs:
-            xt = F.leaky_relu(x, LRELU_SLOPE)
-            if x_mask is not None:
-                xt = xt * x_mask
-            xt = c(xt)
-            x = xt + x
-        if x_mask is not None:
-            x = x * x_mask
-        return x
-
-    def remove_weight_norm(self):
-        for l in self.convs:
-            remove_weight_norm(l)
-
-
-class Log(nn.Module):
-  def forward(self, x, x_mask, reverse=False, **kwargs):
-    if not reverse:
-      y = torch.log(torch.clamp_min(x, 1e-5)) * x_mask
-      logdet = torch.sum(-y, [1, 2])
-      return y, logdet
-    else:
-      x = torch.exp(x) * x_mask
-      return x
-    
-
-class Flip(nn.Module):
-  def forward(self, x, *args, reverse=False, **kwargs):
-    x = torch.flip(x, [1])
-    if not reverse:
-      logdet = torch.zeros(x.size(0)).to(dtype=x.dtype, device=x.device)
-      return x, logdet
-    else:
-      return x
-
-
-class ElementwiseAffine(nn.Module):
-  def __init__(self, channels):
-    super().__init__()
-    self.channels = channels
-    self.m = nn.Parameter(torch.zeros(channels,1))
-    self.logs = nn.Parameter(torch.zeros(channels,1))
-
-  def forward(self, x, x_mask, reverse=False, **kwargs):
-    if not reverse:
-      y = self.m + torch.exp(self.logs) * x
-      y = y * x_mask
-      logdet = torch.sum(self.logs * x_mask, [1,2])
-      return y, logdet
-    else:
-      x = (x - self.m) * torch.exp(-self.logs) * x_mask
-      return x
-
-
-class ResidualCouplingLayer(nn.Module):
-  def __init__(self,
-      channels,
-      hidden_channels,
-      kernel_size,
-      dilation_rate,
-      n_layers,
-      p_dropout=0,
-      gin_channels=0,
-      mean_only=False):
-    assert channels % 2 == 0, "channels should be divisible by 2"
-    super().__init__()
-    self.channels = channels
-    self.hidden_channels = hidden_channels
-    self.kernel_size = kernel_size
-    self.dilation_rate = dilation_rate
-    self.n_layers = n_layers
-    self.half_channels = channels // 2
-    self.mean_only = mean_only
-
-    self.pre = nn.Conv1d(self.half_channels, hidden_channels, 1)
-    self.enc = WN(hidden_channels, kernel_size, dilation_rate, n_layers, p_dropout=p_dropout, gin_channels=gin_channels)
-    self.post = nn.Conv1d(hidden_channels, self.half_channels * (2 - mean_only), 1)
-    self.post.weight.data.zero_()
-    self.post.bias.data.zero_()
-
-  def forward(self, x, x_mask, g=None, reverse=False):
-    x0, x1 = torch.split(x, [self.half_channels]*2, 1)
-    h = self.pre(x0) * x_mask
-    h = self.enc(h, x_mask, g=g)
-    stats = self.post(h) * x_mask
-    if not self.mean_only:
-      m, logs = torch.split(stats, [self.half_channels]*2, 1)
-    else:
-      m = stats
-      logs = torch.zeros_like(m)
-
-    if not reverse:
-      x1 = m + x1 * torch.exp(logs) * x_mask
-      x = torch.cat([x0, x1], 1)
-      logdet = torch.sum(logs, [1,2])
-      return x, logdet
-    else:
-      x1 = (x1 - m) * torch.exp(-logs) * x_mask
-      x = torch.cat([x0, x1], 1)
-      return x
-
-
-class ConvFlow(nn.Module):
-  def __init__(self, in_channels, filter_channels, kernel_size, n_layers, num_bins=10, tail_bound=5.0):
-    super().__init__()
-    self.in_channels = in_channels
-    self.filter_channels = filter_channels
-    self.kernel_size = kernel_size
-    self.n_layers = n_layers
-    self.num_bins = num_bins
-    self.tail_bound = tail_bound
-    self.half_channels = in_channels // 2
-
-    self.pre = nn.Conv1d(self.half_channels, filter_channels, 1)
-    self.convs = DDSConv(filter_channels, kernel_size, n_layers, p_dropout=0.)
-    self.proj = nn.Conv1d(filter_channels, self.half_channels * (num_bins * 3 - 1), 1)
-    self.proj.weight.data.zero_()
-    self.proj.bias.data.zero_()
-
-  def forward(self, x, x_mask, g=None, reverse=False):
-    x0, x1 = torch.split(x, [self.half_channels]*2, 1)
-    h = self.pre(x0)
-    h = self.convs(h, x_mask, g=g)
-    h = self.proj(h) * x_mask
-
-    b, c, t = x0.shape
-    h = h.reshape(b, c, -1, t).permute(0, 1, 3, 2) # [b, cx?, t] -> [b, c, t, ?]
-
-    unnormalized_widths = h[..., :self.num_bins] / math.sqrt(self.filter_channels)
-    unnormalized_heights = h[..., self.num_bins:2*self.num_bins] / math.sqrt(self.filter_channels)
-    unnormalized_derivatives = h[..., 2 * self.num_bins:]
-
-    x1, logabsdet = piecewise_rational_quadratic_transform(x1,
-        unnormalized_widths,
-        unnormalized_heights,
-        unnormalized_derivatives,
-        inverse=reverse,
-        tails='linear',
-        tail_bound=self.tail_bound
-    )
-
-    x = torch.cat([x0, x1], 1) * x_mask
-    logdet = torch.sum(logabsdet * x_mask, [1,2])
-    if not reverse:
-        return x, logdet
-    else:
-        return x
-
-class Encoder(nn.Module):
-  def __init__(self, hidden_channels, filter_channels, n_heads, n_layers, kernel_size=1, p_dropout=0., window_size=4, **kwargs):
-    super().__init__()
-    self.hidden_channels = hidden_channels
-    self.filter_channels = filter_channels
-    self.n_heads = n_heads
-    self.n_layers = n_layers
-    self.kernel_size = kernel_size
-    self.p_dropout = p_dropout
-    self.window_size = window_size
-
-    self.drop = nn.Dropout(p_dropout)
-    self.attn_layers = nn.ModuleList()
-    self.norm_layers_1 = nn.ModuleList()
-    self.ffn_layers = nn.ModuleList()
-    self.norm_layers_2 = nn.ModuleList()
-    for i in range(self.n_layers):
-      self.attn_layers.append(MultiHeadAttention(hidden_channels, hidden_channels, n_heads, p_dropout=p_dropout, window_size=window_size))
-      self.norm_layers_1.append(LayerNorm(hidden_channels))
-      self.ffn_layers.append(FFN(hidden_channels, hidden_channels, filter_channels, kernel_size, p_dropout=p_dropout))
-      self.norm_layers_2.append(LayerNorm(hidden_channels))
-
-  def forward(self, x, x_mask):
-    attn_mask = x_mask.unsqueeze(2) * x_mask.unsqueeze(-1)
-    x = x * x_mask
-    for i in range(self.n_layers):
-      y = self.attn_layers[i](x, x, attn_mask)
-      y = self.drop(y)
-      x = self.norm_layers_1[i](x + y)
-
-      y = self.ffn_layers[i](x, x_mask)
-      y = self.drop(y)
-      x = self.norm_layers_2[i](x + y)
-    x = x * x_mask
-    return x
-
-
-class Decoder(nn.Module):
-  def __init__(self, hidden_channels, filter_channels, n_heads, n_layers, kernel_size=1, p_dropout=0., proximal_bias=False, proximal_init=True, **kwargs):
-    super().__init__()
-    self.hidden_channels = hidden_channels
-    self.filter_channels = filter_channels
-    self.n_heads = n_heads
-    self.n_layers = n_layers
-    self.kernel_size = kernel_size
-    self.p_dropout = p_dropout
-    self.proximal_bias = proximal_bias
-    self.proximal_init = proximal_init
-
-    self.drop = nn.Dropout(p_dropout)
-    self.self_attn_layers = nn.ModuleList()
-    self.norm_layers_0 = nn.ModuleList()
-    self.encdec_attn_layers = nn.ModuleList()
-    self.norm_layers_1 = nn.ModuleList()
-    self.ffn_layers = nn.ModuleList()
-    self.norm_layers_2 = nn.ModuleList()
-    for i in range(self.n_layers):
-      self.self_attn_layers.append(MultiHeadAttention(hidden_channels, hidden_channels, n_heads, p_dropout=p_dropout, proximal_bias=proximal_bias, proximal_init=proximal_init))
-      self.norm_layers_0.append(LayerNorm(hidden_channels))
-      self.encdec_attn_layers.append(MultiHeadAttention(hidden_channels, hidden_channels, n_heads, p_dropout=p_dropout))
-      self.norm_layers_1.append(LayerNorm(hidden_channels))
-      self.ffn_layers.append(FFN(hidden_channels, hidden_channels, filter_channels, kernel_size, p_dropout=p_dropout, causal=True))
-      self.norm_layers_2.append(LayerNorm(hidden_channels))
-
-  def forward(self, x, x_mask, h, h_mask):
-    """
-    x: decoder input
-    h: encoder output
-    """
-    self_attn_mask = subsequent_mask(x_mask.size(2)).to(device=x.device, dtype=x.dtype)
-    encdec_attn_mask = h_mask.unsqueeze(2) * x_mask.unsqueeze(-1)
-    x = x * x_mask
-    for i in range(self.n_layers):
-      y = self.self_attn_layers[i](x, x, self_attn_mask)
-      y = self.drop(y)
-      x = self.norm_layers_0[i](x + y)
-
-      y = self.encdec_attn_layers[i](x, h, encdec_attn_mask)
-      y = self.drop(y)
-      x = self.norm_layers_1[i](x + y)
-      
-      y = self.ffn_layers[i](x, x_mask)
-      y = self.drop(y)
-      x = self.norm_layers_2[i](x + y)
-    x = x * x_mask
-    return x
-
-
-class MultiHeadAttention(nn.Module):
-  def __init__(self, channels, out_channels, n_heads, p_dropout=0., window_size=None, heads_share=True, block_length=None, proximal_bias=False, proximal_init=False):
-    super().__init__()
-    assert channels % n_heads == 0
-
-    self.channels = channels
-    self.out_channels = out_channels
-    self.n_heads = n_heads
-    self.p_dropout = p_dropout
-    self.window_size = window_size
-    self.heads_share = heads_share
-    self.block_length = block_length
-    self.proximal_bias = proximal_bias
-    self.proximal_init = proximal_init
-    self.attn = None
-
-    self.k_channels = channels // n_heads
-    self.conv_q = nn.Conv1d(channels, channels, 1)
-    self.conv_k = nn.Conv1d(channels, channels, 1)
-    self.conv_v = nn.Conv1d(channels, channels, 1)
-    self.conv_o = nn.Conv1d(channels, out_channels, 1)
-    self.drop = nn.Dropout(p_dropout)
-
-    if window_size is not None:
-      n_heads_rel = 1 if heads_share else n_heads
-      rel_stddev = self.k_channels**-0.5
-      self.emb_rel_k = nn.Parameter(torch.randn(n_heads_rel, window_size * 2 + 1, self.k_channels) * rel_stddev)
-      self.emb_rel_v = nn.Parameter(torch.randn(n_heads_rel, window_size * 2 + 1, self.k_channels) * rel_stddev)
-
-    nn.init.xavier_uniform_(self.conv_q.weight)
-    nn.init.xavier_uniform_(self.conv_k.weight)
-    nn.init.xavier_uniform_(self.conv_v.weight)
-    if proximal_init:
-      with torch.no_grad():
-        self.conv_k.weight.copy_(self.conv_q.weight)
-        self.conv_k.bias.copy_(self.conv_q.bias)
-      
-  def forward(self, x, c, attn_mask=None):
-    q = self.conv_q(x)
-    k = self.conv_k(c)
-    v = self.conv_v(c)
-    
-    x, self.attn = self.attention(q, k, v, mask=attn_mask)
-
-    x = self.conv_o(x)
-    return x
-
-  def attention(self, query, key, value, mask=None):
-    # reshape [b, d, t] -> [b, n_h, t, d_k]
-    b, d, t_s, t_t = (*key.size(), query.size(2))
-    query = query.view(b, self.n_heads, self.k_channels, t_t).transpose(2, 3)
-    key = key.view(b, self.n_heads, self.k_channels, t_s).transpose(2, 3)
-    value = value.view(b, self.n_heads, self.k_channels, t_s).transpose(2, 3)
-
-    scores = torch.matmul(query / math.sqrt(self.k_channels), key.transpose(-2, -1))
-    if self.window_size is not None:
-      assert t_s == t_t, "Relative attention is only available for self-attention."
-      key_relative_embeddings = self._get_relative_embeddings(self.emb_rel_k, t_s)
-      rel_logits = self._matmul_with_relative_keys(query /math.sqrt(self.k_channels), key_relative_embeddings)
-      scores_local = self._relative_position_to_absolute_position(rel_logits)
-      scores = scores + scores_local
-    if self.proximal_bias:
-      assert t_s == t_t, "Proximal bias is only available for self-attention."
-      scores = scores + self._attention_bias_proximal(t_s).to(device=scores.device, dtype=scores.dtype)
-    if mask is not None:
-      scores = scores.masked_fill(mask == 0, -1e4)
-      if self.block_length is not None:
-        assert t_s == t_t, "Local attention is only available for self-attention."
-        block_mask = torch.ones_like(scores).triu(-self.block_length).tril(self.block_length)
-        scores = scores.masked_fill(block_mask == 0, -1e4)
-    p_attn = F.softmax(scores, dim=-1) # [b, n_h, t_t, t_s]
-    p_attn = self.drop(p_attn)
-    output = torch.matmul(p_attn, value)
-    if self.window_size is not None:
-      relative_weights = self._absolute_position_to_relative_position(p_attn)
-      value_relative_embeddings = self._get_relative_embeddings(self.emb_rel_v, t_s)
-      output = output + self._matmul_with_relative_values(relative_weights, value_relative_embeddings)
-    output = output.transpose(2, 3).contiguous().view(b, d, t_t) # [b, n_h, t_t, d_k] -> [b, d, t_t]
-    return output, p_attn
-
-  def _matmul_with_relative_values(self, x, y):
-    """
-    x: [b, h, l, m]
-    y: [h or 1, m, d]
-    ret: [b, h, l, d]
-    """
-    ret = torch.matmul(x, y.unsqueeze(0))
-    return ret
-
-  def _matmul_with_relative_keys(self, x, y):
-    """
-    x: [b, h, l, d]
-    y: [h or 1, m, d]
-    ret: [b, h, l, m]
-    """
-    ret = torch.matmul(x, y.unsqueeze(0).transpose(-2, -1))
-    return ret
-
-  def _get_relative_embeddings(self, relative_embeddings, length):
-    max_relative_position = 2 * self.window_size + 1
-    # Pad first before slice to avoid using cond ops.
-    pad_length = max(length - (self.window_size + 1), 0)
-    slice_start_position = max((self.window_size + 1) - length, 0)
-    slice_end_position = slice_start_position + 2 * length - 1
-    if pad_length > 0:
-      padded_relative_embeddings = F.pad(
-          relative_embeddings,
-          convert_pad_shape([[0, 0], [pad_length, pad_length], [0, 0]]))
-    else:
-      padded_relative_embeddings = relative_embeddings
-    used_relative_embeddings = padded_relative_embeddings[:,slice_start_position:slice_end_position]
-    return used_relative_embeddings
-
-  def _relative_position_to_absolute_position(self, x):
-    """
-    x: [b, h, l, 2*l-1]
-    ret: [b, h, l, l]
-    """
-    batch, heads, length, _ = x.size()
-    # Concat columns of pad to shift from relative to absolute indexing.
-    x = F.pad(x, convert_pad_shape([[0,0],[0,0],[0,0],[0,1]]))
-
-    # Concat extra elements so to add up to shape (len+1, 2*len-1).
-    x_flat = x.view([batch, heads, length * 2 * length])
-    x_flat = F.pad(x_flat, convert_pad_shape([[0,0],[0,0],[0,length-1]]))
-
-    # Reshape and slice out the padded elements.
-    x_final = x_flat.view([batch, heads, length+1, 2*length-1])[:, :, :length, length-1:]
-    return x_final
-
-  def _absolute_position_to_relative_position(self, x):
-    """
-    x: [b, h, l, l]
-    ret: [b, h, l, 2*l-1]
-    """
-    batch, heads, length, _ = x.size()
-    # padd along column
-    x = F.pad(x, convert_pad_shape([[0, 0], [0, 0], [0, 0], [0, length-1]]))
-    x_flat = x.view([batch, heads, length**2 + length*(length -1)])
-    # add 0's in the beginning that will skew the elements after reshape
-    x_flat = F.pad(x_flat, convert_pad_shape([[0, 0], [0, 0], [length, 0]]))
-    x_final = x_flat.view([batch, heads, length, 2*length])[:,:,:,1:]
-    return x_final
-
-  def _attention_bias_proximal(self, length):
-    """Bias for self-attention to encourage attention to close positions.
-    Args:
-      length: an integer scalar.
-    Returns:
-      a Tensor with shape [1, 1, length, length]
-    """
-    r = torch.arange(length, dtype=torch.float32)
-    diff = torch.unsqueeze(r, 0) - torch.unsqueeze(r, 1)
-    return torch.unsqueeze(torch.unsqueeze(-torch.log1p(torch.abs(diff)), 0), 0)
-
-
-class FFN(nn.Module):
-  def __init__(self, in_channels, out_channels, filter_channels, kernel_size, p_dropout=0., activation=None, causal=False):
-    super().__init__()
-    self.in_channels = in_channels
-    self.out_channels = out_channels
-    self.filter_channels = filter_channels
-    self.kernel_size = kernel_size
-    self.p_dropout = p_dropout
-    self.activation = activation
-    self.causal = causal
-
-    if causal:
-      self.padding = self._causal_padding
-    else:
-      self.padding = self._same_padding
-
-    self.conv_1 = nn.Conv1d(in_channels, filter_channels, kernel_size)
-    self.conv_2 = nn.Conv1d(filter_channels, out_channels, kernel_size)
-    self.drop = nn.Dropout(p_dropout)
-
-  def forward(self, x, x_mask):
-    x = self.conv_1(self.padding(x * x_mask))
-    if self.activation == "gelu":
-      x = x * torch.sigmoid(1.702 * x)
-    else:
-      x = torch.relu(x)
-    x = self.drop(x)
-    x = self.conv_2(self.padding(x * x_mask))
-    return x * x_mask
-  
-  def _causal_padding(self, x):
-    if self.kernel_size == 1:
-      return x
-    pad_l = self.kernel_size - 1
-    pad_r = 0
-    padding = [[0, 0], [0, 0], [pad_l, pad_r]]
-    x = F.pad(x, convert_pad_shape(padding))
-    return x
-
-  def _same_padding(self, x):
-    if self.kernel_size == 1:
-      return x
-    pad_l = (self.kernel_size - 1) // 2
-    pad_r = self.kernel_size // 2
-    padding = [[0, 0], [0, 0], [pad_l, pad_r]]
-    x = F.pad(x, convert_pad_shape(padding))
-    return x

models/synthesizer/models/tacotron.py

@@ -1,298 +0,0 @@
-import torch
-import torch.nn as nn
-from .sublayer.global_style_token import GlobalStyleToken
-from .sublayer.pre_net import PreNet
-from .sublayer.cbhg import CBHG
-from .sublayer.lsa import LSA
-from .base import Base
-from models.synthesizer.gst_hyperparameters import GSTHyperparameters as gst_hp
-from models.synthesizer.hparams import hparams
-
-class Encoder(nn.Module):
-    def __init__(self, num_chars, embed_dims=512, encoder_dims=256, K=5, num_highways=4, dropout=0.5):
-        """ Encoder for SV2TTS
-
-        Args:
-            num_chars (int): length of symbols
-            embed_dims (int, optional): embedding dim for input texts. Defaults to 512.
-            encoder_dims (int, optional): output dim for encoder. Defaults to 256.
-            K (int, optional): _description_. Defaults to 5.
-            num_highways (int, optional): _description_. Defaults to 4.
-            dropout (float, optional): _description_. Defaults to 0.5.
-        """             
-        super().__init__()
-        self.embedding = nn.Embedding(num_chars, embed_dims)
-        self.pre_net = PreNet(embed_dims, fc1_dims=encoder_dims, fc2_dims=encoder_dims,
-                              dropout=dropout)
-        self.cbhg = CBHG(K=K, in_channels=encoder_dims, channels=encoder_dims,
-                         proj_channels=[encoder_dims, encoder_dims],
-                         num_highways=num_highways)
-
-    def forward(self, x):
-        """forward pass for encoder
-
-        Args:
-            x (2D tensor with size `[batch_size, text_num_chars]`): input texts list
-
-        Returns:
-            3D tensor with size `[batch_size, text_num_chars, encoder_dims]`
-            
-        """
-        x = self.embedding(x) # return: [batch_size, text_num_chars, tts_embed_dims]
-        x = self.pre_net(x) # return: [batch_size, text_num_chars, encoder_dims]
-        x.transpose_(1, 2)  # return: [batch_size, encoder_dims, text_num_chars]
-        return self.cbhg(x) # return: [batch_size, text_num_chars, encoder_dims]
-
-class Decoder(nn.Module):
-    # Class variable because its value doesn't change between classes
-    # yet ought to be scoped by class because its a property of a Decoder
-    max_r = 20
-    def __init__(self, n_mels, input_dims, decoder_dims, lstm_dims,
-                 dropout, speaker_embedding_size):
-        super().__init__()
-        self.register_buffer("r", torch.tensor(1, dtype=torch.int))
-        self.n_mels = n_mels
-        self.prenet = PreNet(n_mels, fc1_dims=decoder_dims * 2, fc2_dims=decoder_dims * 2,
-                             dropout=dropout)
-        self.attn_net = LSA(decoder_dims)
-        if hparams.use_gst:
-            speaker_embedding_size += gst_hp.E
-        self.attn_rnn = nn.GRUCell(input_dims + decoder_dims * 2, decoder_dims)
-        self.rnn_input = nn.Linear(input_dims  + decoder_dims, lstm_dims)
-        self.res_rnn1 = nn.LSTMCell(lstm_dims, lstm_dims)
-        self.res_rnn2 = nn.LSTMCell(lstm_dims, lstm_dims)
-        self.mel_proj = nn.Linear(lstm_dims, n_mels * self.max_r, bias=False)
-        self.stop_proj = nn.Linear(input_dims + lstm_dims, 1)
-
-    def zoneout(self, prev, current, device, p=0.1):
-        mask = torch.zeros(prev.size(),device=device).bernoulli_(p)
-        return prev * mask + current * (1 - mask)
-
-    def forward(self, encoder_seq, encoder_seq_proj, prenet_in,
-                hidden_states, cell_states, context_vec, times, chars):
-        """_summary_
-
-        Args:
-            encoder_seq (3D tensor `[batch_size, text_num_chars, project_dim(default to 512)]`): _description_
-            encoder_seq_proj (3D tensor `[batch_size, text_num_chars, decoder_dims(default to 128)]`): _description_
-            prenet_in (2D tensor `[batch_size, n_mels]`): _description_
-            hidden_states (_type_): _description_
-            cell_states (_type_): _description_
-            context_vec (2D tensor `[batch_size, project_dim(default to 512)]`): _description_
-            times (int): the number of times runned
-            chars (2D tensor with size `[batch_size, text_num_chars]`): original texts list input
-
-        """
-        # Need this for reshaping mels
-        batch_size = encoder_seq.size(0)
-        device = encoder_seq.device
-        # Unpack the hidden and cell states
-        attn_hidden, rnn1_hidden, rnn2_hidden = hidden_states
-        rnn1_cell, rnn2_cell = cell_states
-
-        # PreNet for the Attention RNN
-        prenet_out = self.prenet(prenet_in) # return: `[batch_size, decoder_dims * 2(256)]`
-
-        # Compute the Attention RNN hidden state
-        attn_rnn_in = torch.cat([context_vec, prenet_out], dim=-1) # `[batch_size, project_dim + decoder_dims * 2 (768)]`
-        attn_hidden = self.attn_rnn(attn_rnn_in.squeeze(1), attn_hidden) #  `[batch_size, decoder_dims (128)]`
-
-        # Compute the attention scores
-        scores = self.attn_net(encoder_seq_proj, attn_hidden, times, chars)
-
-        # Dot product to create the context vector
-        context_vec = scores @ encoder_seq
-        context_vec = context_vec.squeeze(1)
-
-        # Concat Attention RNN output w. Context Vector & project
-        x = torch.cat([context_vec, attn_hidden], dim=1) # `[batch_size, project_dim + decoder_dims (630)]`
-        x = self.rnn_input(x) # `[batch_size, lstm_dims(1024)]`
-
-        # Compute first Residual RNN, training with fixed zoneout rate 0.1
-        rnn1_hidden_next, rnn1_cell = self.res_rnn1(x, (rnn1_hidden, rnn1_cell)) # `[batch_size, lstm_dims(1024)]`
-        if self.training:
-            rnn1_hidden = self.zoneout(rnn1_hidden, rnn1_hidden_next,device=device)
-        else:
-            rnn1_hidden = rnn1_hidden_next
-        x = x + rnn1_hidden
-
-        # Compute second Residual RNN
-        rnn2_hidden_next, rnn2_cell = self.res_rnn2(x, (rnn2_hidden, rnn2_cell)) # `[batch_size, lstm_dims(1024)]`
-        if self.training:
-            rnn2_hidden = self.zoneout(rnn2_hidden, rnn2_hidden_next, device=device)
-        else:
-            rnn2_hidden = rnn2_hidden_next
-        x = x + rnn2_hidden
-
-        # Project Mels
-        mels = self.mel_proj(x) # `[batch_size, 1600]`
-        mels = mels.view(batch_size, self.n_mels, self.max_r)[:, :, :self.r] # `[batch_size, n_mels, r]`
-        hidden_states = (attn_hidden, rnn1_hidden, rnn2_hidden)
-        cell_states = (rnn1_cell, rnn2_cell)
-
-        # Stop token prediction
-        s = torch.cat((x, context_vec), dim=1)
-        s = self.stop_proj(s)
-        stop_tokens = torch.sigmoid(s)
-
-        return mels, scores, hidden_states, cell_states, context_vec, stop_tokens
-
-class Tacotron(Base):
-    def __init__(self, embed_dims, num_chars, encoder_dims, decoder_dims, n_mels, 
-                 fft_bins, postnet_dims, encoder_K, lstm_dims, postnet_K, num_highways,
-                 dropout, stop_threshold, speaker_embedding_size):
-        super().__init__(stop_threshold)
-        self.n_mels = n_mels
-        self.lstm_dims = lstm_dims
-        self.encoder_dims = encoder_dims
-        self.decoder_dims = decoder_dims
-        self.speaker_embedding_size = speaker_embedding_size
-        self.encoder = Encoder(num_chars, embed_dims, encoder_dims,
-                               encoder_K, num_highways, dropout)
-        self.project_dims = encoder_dims + speaker_embedding_size
-        if hparams.use_gst: 
-            self.project_dims += gst_hp.E
-        self.encoder_proj = nn.Linear(self.project_dims, decoder_dims, bias=False)
-        if hparams.use_gst: 
-            self.gst = GlobalStyleToken(speaker_embedding_size)
-        self.decoder = Decoder(n_mels, self.project_dims, decoder_dims, lstm_dims,
-                               dropout, speaker_embedding_size)
-        self.postnet = CBHG(postnet_K, n_mels, postnet_dims,
-                            [postnet_dims, fft_bins], num_highways)
-        self.post_proj = nn.Linear(postnet_dims, fft_bins, bias=False)
-
-    @staticmethod
-    def _concat_speaker_embedding(outputs, speaker_embeddings):
-        speaker_embeddings_ = speaker_embeddings.expand(
-            outputs.size(0), outputs.size(1), -1)
-        outputs = torch.cat([outputs, speaker_embeddings_], dim=-1)
-        return outputs
-
-    @staticmethod
-    def _add_speaker_embedding(x, speaker_embedding):
-        """Add speaker embedding
-            This concats the speaker embedding for each char in the encoder output
-        Args:
-            x (3D tensor with size `[batch_size, text_num_chars, encoder_dims]`): the encoder output
-            speaker_embedding (2D tensor `[batch_size, speaker_embedding_size]`): the speaker embedding
-
-        Returns:
-            3D tensor with size `[batch_size, text_num_chars, encoder_dims+speaker_embedding_size]`
-        """        
-        # Save the dimensions as human-readable names
-        batch_size = x.size()[0]
-        text_num_chars = x.size()[1]
-
-        # Start by making a copy of each speaker embedding to match the input text length
-        # The output of this has size (batch_size, text_num_chars * speaker_embedding_size)
-        speaker_embedding_size = speaker_embedding.size()[1]
-        e = speaker_embedding.repeat_interleave(text_num_chars, dim=1)
-
-        # Reshape it and transpose
-        e = e.reshape(batch_size, speaker_embedding_size, text_num_chars)
-        e = e.transpose(1, 2)
-
-        # Concatenate the tiled speaker embedding with the encoder output
-        x = torch.cat((x, e), 2)
-        return x
-
-    def forward(self, texts, mels, speaker_embedding, steps=2000, style_idx=0, min_stop_token=5):
-        """Forward pass for Tacotron
-
-        Args:
-            texts (`[batch_size, text_num_chars]`): input texts list
-            mels (`[batch_size, varied_mel_lengths, steps]`): mels for comparison (training only)
-            speaker_embedding (`[batch_size, speaker_embedding_size(default to 256)]`): referring embedding.
-            steps (int, optional): . Defaults to 2000.
-            style_idx (int, optional): GST style selected. Defaults to 0.
-            min_stop_token (int, optional): decoder min_stop_token. Defaults to 5.
-        """
-        device = texts.device  # use same device as parameters
-
-        if self.training:
-            self.step += 1
-            batch_size, _, steps  = mels.size()
-        else:
-            batch_size, _  = texts.size()
-
-        # Initialise all hidden states and pack into tuple
-        attn_hidden = torch.zeros(batch_size, self.decoder_dims, device=device)
-        rnn1_hidden = torch.zeros(batch_size, self.lstm_dims, device=device)
-        rnn2_hidden = torch.zeros(batch_size, self.lstm_dims, device=device)
-        hidden_states = (attn_hidden, rnn1_hidden, rnn2_hidden)
-
-        # Initialise all lstm cell states and pack into tuple
-        rnn1_cell = torch.zeros(batch_size, self.lstm_dims, device=device)
-        rnn2_cell = torch.zeros(batch_size, self.lstm_dims, device=device)
-        cell_states = (rnn1_cell, rnn2_cell)
-
-        # <GO> Frame for start of decoder loop
-        go_frame = torch.zeros(batch_size, self.n_mels, device=device)
-
-        # SV2TTS: Run the encoder with the speaker embedding
-        # The projection avoids unnecessary matmuls in the decoder loop
-        encoder_seq = self.encoder(texts)
-        
-        encoder_seq = self._add_speaker_embedding(encoder_seq, speaker_embedding)
-
-        if hparams.use_gst and self.gst is not None:
-            if self.training:
-                style_embed = self.gst(speaker_embedding, speaker_embedding) # for training, speaker embedding can represent both style inputs and referenced
-                # style_embed = style_embed.expand_as(encoder_seq)
-                # encoder_seq = torch.cat((encoder_seq, style_embed), 2)
-            elif style_idx >= 0 and style_idx < 10:
-                query = torch.zeros(1, 1, self.gst.stl.attention.num_units)
-                if device.type == 'cuda':
-                    query = query.cuda()
-                gst_embed = torch.tanh(self.gst.stl.embed)
-                key = gst_embed[style_idx].unsqueeze(0).expand(1, -1, -1)
-                style_embed = self.gst.stl.attention(query, key)
-            else:
-                speaker_embedding_style = torch.zeros(speaker_embedding.size()[0], 1, self.speaker_embedding_size).to(device)
-                style_embed = self.gst(speaker_embedding_style, speaker_embedding)
-            encoder_seq = self._concat_speaker_embedding(encoder_seq, style_embed) # return: [batch_size, text_num_chars, project_dims]
-        
-        encoder_seq_proj = self.encoder_proj(encoder_seq) # return: [batch_size, text_num_chars, decoder_dims]
-
-        # Need a couple of lists for outputs
-        mel_outputs, attn_scores, stop_outputs = [], [], []
-
-        # Need an initial context vector
-        context_vec = torch.zeros(batch_size, self.project_dims, device=device)
-
-        # Run the decoder loop
-        for t in range(0, steps, self.r):
-            if self.training:
-                prenet_in = mels[:, :, t -1] if t > 0 else go_frame
-            else:
-                prenet_in = mel_outputs[-1][:, :, -1] if t > 0 else go_frame
-            mel_frames, scores, hidden_states, cell_states, context_vec, stop_tokens = \
-                self.decoder(encoder_seq, encoder_seq_proj, prenet_in,
-                             hidden_states, cell_states, context_vec, t, texts)
-            mel_outputs.append(mel_frames)
-            attn_scores.append(scores)
-            stop_outputs.extend([stop_tokens] * self.r)
-            if not self.training and (stop_tokens * 10 > min_stop_token).all() and t > 10: break
-
-        # Concat the mel outputs into sequence
-        mel_outputs = torch.cat(mel_outputs, dim=2)
-
-        # Post-Process for Linear Spectrograms
-        postnet_out = self.postnet(mel_outputs)
-        linear = self.post_proj(postnet_out)
-        linear = linear.transpose(1, 2)
-
-        # For easy visualisation
-        attn_scores = torch.cat(attn_scores, 1)
-        # attn_scores = attn_scores.cpu().data.numpy()
-        stop_outputs = torch.cat(stop_outputs, 1)
-
-        if self.training:
-            self.train()
-            
-        return mel_outputs, linear, attn_scores, stop_outputs
-
-    def generate(self, x, speaker_embedding, steps=2000, style_idx=0, min_stop_token=5):
-        self.eval()
-        mel_outputs, linear, attn_scores, _ =  self.forward(x, None, speaker_embedding, steps, style_idx, min_stop_token)
-        return mel_outputs, linear, attn_scores

models/synthesizer/models/vits.py

@@ -1,524 +0,0 @@
-import math
-import torch
-from torch import nn
-from torch.nn import functional as F
-
-from .sublayer.vits_modules import *
-import monotonic_align
-
-from .base import Base
-from torch.nn import Conv1d, ConvTranspose1d, AvgPool1d, Conv2d
-from torch.nn.utils import weight_norm, remove_weight_norm, spectral_norm
-from utils.util import init_weights, get_padding, sequence_mask, rand_slice_segments, generate_path
-
-
-class StochasticDurationPredictor(nn.Module):
-  def __init__(self, in_channels, filter_channels, kernel_size, p_dropout, n_flows=4, gin_channels=0):
-    super().__init__()
-    filter_channels = in_channels # it needs to be removed from future version.
-    self.in_channels = in_channels
-    self.filter_channels = filter_channels
-    self.kernel_size = kernel_size
-    self.p_dropout = p_dropout
-    self.n_flows = n_flows
-    self.gin_channels = gin_channels
-
-    self.log_flow = Log()
-    self.flows = nn.ModuleList()
-    self.flows.append(ElementwiseAffine(2))
-    for i in range(n_flows):
-      self.flows.append(ConvFlow(2, filter_channels, kernel_size, n_layers=3))
-      self.flows.append(Flip())
-
-    self.post_pre = nn.Conv1d(1, filter_channels, 1)
-    self.post_proj = nn.Conv1d(filter_channels, filter_channels, 1)
-    self.post_convs = DDSConv(filter_channels, kernel_size, n_layers=3, p_dropout=p_dropout)
-    self.post_flows = nn.ModuleList()
-    self.post_flows.append(ElementwiseAffine(2))
-    for i in range(4):
-      self.post_flows.append(ConvFlow(2, filter_channels, kernel_size, n_layers=3))
-      self.post_flows.append(Flip())
-
-    self.pre = nn.Conv1d(in_channels, filter_channels, 1)
-    self.proj = nn.Conv1d(filter_channels, filter_channels, 1)
-    self.convs = DDSConv(filter_channels, kernel_size, n_layers=3, p_dropout=p_dropout)
-    if gin_channels != 0:
-      self.cond = nn.Conv1d(gin_channels, filter_channels, 1)
-
-  def forward(self, x, x_mask, w=None, g=None, reverse=False, noise_scale=1.0):
-    x = torch.detach(x)
-    x = self.pre(x)
-    if g is not None:
-      g = torch.detach(g)
-      x = x + self.cond(g)
-    x = self.convs(x, x_mask)
-    x = self.proj(x) * x_mask
-
-    if not reverse:
-      flows = self.flows
-      assert w is not None
-
-      logdet_tot_q = 0 
-      h_w = self.post_pre(w)
-      h_w = self.post_convs(h_w, x_mask)
-      h_w = self.post_proj(h_w) * x_mask
-      e_q = torch.randn(w.size(0), 2, w.size(2)).to(device=x.device, dtype=x.dtype) * x_mask
-      z_q = e_q
-      for flow in self.post_flows:
-        z_q, logdet_q = flow(z_q, x_mask, g=(x + h_w))
-        logdet_tot_q += logdet_q
-      z_u, z1 = torch.split(z_q, [1, 1], 1) 
-      u = torch.sigmoid(z_u) * x_mask
-      z0 = (w - u) * x_mask
-      logdet_tot_q += torch.sum((F.logsigmoid(z_u) + F.logsigmoid(-z_u)) * x_mask, [1,2])
-      logq = torch.sum(-0.5 * (math.log(2*math.pi) + (e_q**2)) * x_mask, [1,2]) - logdet_tot_q
-
-      logdet_tot = 0
-      z0, logdet = self.log_flow(z0, x_mask)
-      logdet_tot += logdet
-      z = torch.cat([z0, z1], 1)
-      for flow in flows:
-        z, logdet = flow(z, x_mask, g=x, reverse=reverse)
-        logdet_tot = logdet_tot + logdet
-      nll = torch.sum(0.5 * (math.log(2*math.pi) + (z**2)) * x_mask, [1,2]) - logdet_tot
-      return nll + logq # [b]
-    else:
-      flows = list(reversed(self.flows))
-      flows = flows[:-2] + [flows[-1]] # remove a useless vflow
-      z = torch.randn(x.size(0), 2, x.size(2)).to(device=x.device, dtype=x.dtype) * noise_scale
-      for flow in flows:
-        z = flow(z, x_mask, g=x, reverse=reverse)
-      z0, z1 = torch.split(z, [1, 1], 1)
-      logw = z0
-      return logw
-
-
-class DurationPredictor(nn.Module):
-  def __init__(self, in_channels, filter_channels, kernel_size, p_dropout, gin_channels=0):
-    super().__init__()
-
-    self.in_channels = in_channels
-    self.filter_channels = filter_channels
-    self.kernel_size = kernel_size
-    self.p_dropout = p_dropout
-    self.gin_channels = gin_channels
-
-    self.drop = nn.Dropout(p_dropout)
-    self.conv_1 = nn.Conv1d(in_channels, filter_channels, kernel_size, padding=kernel_size//2)
-    self.norm_1 = LayerNorm(filter_channels)
-    self.conv_2 = nn.Conv1d(filter_channels, filter_channels, kernel_size, padding=kernel_size//2)
-    self.norm_2 = LayerNorm(filter_channels)
-    self.proj = nn.Conv1d(filter_channels, 1, 1)
-
-    if gin_channels != 0:
-      self.cond = nn.Conv1d(gin_channels, in_channels, 1)
-
-  def forward(self, x, x_mask, g=None):
-    x = torch.detach(x)
-    if g is not None:
-      g = torch.detach(g)
-      x = x + self.cond(g)
-    x = self.conv_1(x * x_mask)
-    x = torch.relu(x)
-    x = self.norm_1(x)
-    x = self.drop(x)
-    x = self.conv_2(x * x_mask)
-    x = torch.relu(x)
-    x = self.norm_2(x)
-    x = self.drop(x)
-    x = self.proj(x * x_mask)
-    return x * x_mask
-
-
-class TextEncoder(nn.Module):
-  def __init__(self,
-      n_vocab,
-      out_channels,
-      hidden_channels,
-      filter_channels,
-      n_heads,
-      n_layers,
-      kernel_size,
-      p_dropout):
-    super().__init__()
-    self.n_vocab = n_vocab
-    self.out_channels = out_channels
-    self.hidden_channels = hidden_channels
-    self.filter_channels = filter_channels
-    self.n_heads = n_heads
-    self.n_layers = n_layers
-    self.kernel_size = kernel_size
-    self.p_dropout = p_dropout
-
-    self.emb = nn.Embedding(n_vocab, hidden_channels)
-    self.emo_proj = nn.Linear(1024, hidden_channels)
-
-    nn.init.normal_(self.emb.weight, 0.0, hidden_channels**-0.5)
-
-    self.encoder = Encoder(
-      hidden_channels,
-      filter_channels,
-      n_heads,
-      n_layers,
-      kernel_size,
-      p_dropout)
-    self.proj= nn.Conv1d(hidden_channels, out_channels * 2, 1)
-
-  def forward(self, x, x_lengths, emo):
-    x = self.emb(x) * math.sqrt(self.hidden_channels) # [b, t, h]
-    x = x + self.emo_proj(emo.unsqueeze(1))
-    x = torch.transpose(x, 1, -1) # [b, h, t]
-    x_mask = torch.unsqueeze(sequence_mask(x_lengths, x.size(2)), 1).to(x.dtype)
-
-    x = self.encoder(x * x_mask, x_mask)
-    stats = self.proj(x) * x_mask
-
-    m, logs = torch.split(stats, self.out_channels, dim=1)
-    return x, m, logs, x_mask
-
-
-class ResidualCouplingBlock(nn.Module):
-  def __init__(self,
-      channels,
-      hidden_channels,
-      kernel_size,
-      dilation_rate,
-      n_layers,
-      n_flows=4,
-      gin_channels=0):
-    super().__init__()
-    self.channels = channels
-    self.hidden_channels = hidden_channels
-    self.kernel_size = kernel_size
-    self.dilation_rate = dilation_rate
-    self.n_layers = n_layers
-    self.n_flows = n_flows
-    self.gin_channels = gin_channels
-
-    self.flows = nn.ModuleList()
-    for i in range(n_flows):
-      self.flows.append(ResidualCouplingLayer(channels, hidden_channels, kernel_size, dilation_rate, n_layers, gin_channels=gin_channels, mean_only=True))
-      self.flows.append(Flip())
-
-  def forward(self, x, x_mask, g=None, reverse=False):
-    if not reverse:
-      for flow in self.flows:
-        x, _ = flow(x, x_mask, g=g, reverse=reverse)
-    else:
-      for flow in reversed(self.flows):
-        x = flow(x, x_mask, g=g, reverse=reverse)
-    return x
-
-
-class PosteriorEncoder(nn.Module):
-  def __init__(self,
-      in_channels,
-      out_channels,
-      hidden_channels,
-      kernel_size,
-      dilation_rate,
-      n_layers,
-      gin_channels=0):
-    super().__init__()
-    self.in_channels = in_channels
-    self.out_channels = out_channels
-    self.hidden_channels = hidden_channels
-    self.kernel_size = kernel_size
-    self.dilation_rate = dilation_rate
-    self.n_layers = n_layers
-    self.gin_channels = gin_channels
-
-    self.pre = nn.Conv1d(in_channels, hidden_channels, 1)
-    self.enc = WN(hidden_channels, kernel_size, dilation_rate, n_layers, gin_channels=gin_channels)
-    self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
-
-  def forward(self, x, x_lengths, g=None):
-    x_mask = torch.unsqueeze(sequence_mask(x_lengths, x.size(2)), 1).to(x.dtype)
-    x = self.pre(x) * x_mask
-    x = self.enc(x, x_mask, g=g)
-    stats = self.proj(x) * x_mask
-    m, logs = torch.split(stats, self.out_channels, dim=1)
-    z = (m + torch.randn_like(m) * torch.exp(logs)) * x_mask
-    return z, m, logs, x_mask
-
-
-class Generator(torch.nn.Module):
-    def __init__(self, initial_channel, resblock, resblock_kernel_sizes, resblock_dilation_sizes, upsample_rates, upsample_initial_channel, upsample_kernel_sizes, gin_channels=0):
-        super(Generator, self).__init__()
-        self.num_kernels = len(resblock_kernel_sizes)
-        self.num_upsamples = len(upsample_rates)
-        self.conv_pre = Conv1d(initial_channel, upsample_initial_channel, 7, 1, padding=3)
-        resblock = ResBlock1 if resblock == '1' else ResBlock2
-
-        self.ups = nn.ModuleList()
-        for i, (u, k) in enumerate(zip(upsample_rates, upsample_kernel_sizes)):
-            self.ups.append(weight_norm(
-                ConvTranspose1d(upsample_initial_channel//(2**i), upsample_initial_channel//(2**(i+1)),
-                                k, u, padding=(k-u)//2)))
-
-        self.resblocks = nn.ModuleList()
-        for i in range(len(self.ups)):
-            ch = upsample_initial_channel//(2**(i+1))
-            for j, (k, d) in enumerate(zip(resblock_kernel_sizes, resblock_dilation_sizes)):
-                self.resblocks.append(resblock(ch, k, d))
-
-        self.conv_post = Conv1d(ch, 1, 7, 1, padding=3, bias=False)
-        self.ups.apply(init_weights)
-
-        if gin_channels != 0:
-            self.cond = nn.Conv1d(gin_channels, upsample_initial_channel, 1)
-
-    def forward(self, x, g=None):
-        x = self.conv_pre(x)
-        if g is not None:
-          x = x + self.cond(g)
-
-        for i in range(self.num_upsamples):
-            x = F.leaky_relu(x, LRELU_SLOPE)
-            x = self.ups[i](x)
-            xs = None
-            for j in range(self.num_kernels):
-                if xs is None:
-                    xs = self.resblocks[i*self.num_kernels+j](x)
-                else:
-                    xs += self.resblocks[i*self.num_kernels+j](x)
-            x = xs / self.num_kernels
-        x = F.leaky_relu(x)
-        x = self.conv_post(x)
-        x = torch.tanh(x)
-
-        return x
-
-    def remove_weight_norm(self):
-        print('Removing weight norm...')
-        for l in self.ups:
-            remove_weight_norm(l)
-        for l in self.resblocks:
-            l.remove_weight_norm()
-
-
-class DiscriminatorP(torch.nn.Module):
-    def __init__(self, period, kernel_size=5, stride=3, use_spectral_norm=False):
-        super(DiscriminatorP, self).__init__()
-        self.period = period
-        self.use_spectral_norm = use_spectral_norm
-        norm_f = weight_norm if use_spectral_norm == False else spectral_norm
-        self.convs = nn.ModuleList([
-            norm_f(Conv2d(1, 32, (kernel_size, 1), (stride, 1), padding=(get_padding(kernel_size, 1), 0))),
-            norm_f(Conv2d(32, 128, (kernel_size, 1), (stride, 1), padding=(get_padding(kernel_size, 1), 0))),
-            norm_f(Conv2d(128, 512, (kernel_size, 1), (stride, 1), padding=(get_padding(kernel_size, 1), 0))),
-            norm_f(Conv2d(512, 1024, (kernel_size, 1), (stride, 1), padding=(get_padding(kernel_size, 1), 0))),
-            norm_f(Conv2d(1024, 1024, (kernel_size, 1), 1, padding=(get_padding(kernel_size, 1), 0))),
-        ])
-        self.conv_post = norm_f(Conv2d(1024, 1, (3, 1), 1, padding=(1, 0)))
-
-    def forward(self, x):
-        fmap = []
-
-        # 1d to 2d
-        b, c, t = x.shape
-        if t % self.period != 0: # pad first
-            n_pad = self.period - (t % self.period)
-            x = F.pad(x, (0, n_pad), "reflect")
-            t = t + n_pad
-        x = x.view(b, c, t // self.period, self.period)
-
-        for l in self.convs:
-            x = l(x)
-            x = F.leaky_relu(x, LRELU_SLOPE)
-            fmap.append(x)
-        x = self.conv_post(x)
-        fmap.append(x)
-        x = torch.flatten(x, 1, -1)
-
-        return x, fmap
-
-
-class DiscriminatorS(torch.nn.Module):
-    def __init__(self, use_spectral_norm=False):
-        super(DiscriminatorS, self).__init__()
-        norm_f = weight_norm if use_spectral_norm == False else spectral_norm
-        self.convs = nn.ModuleList([
-            norm_f(Conv1d(1, 16, 15, 1, padding=7)),
-            norm_f(Conv1d(16, 64, 41, 4, groups=4, padding=20)),
-            norm_f(Conv1d(64, 256, 41, 4, groups=16, padding=20)),
-            norm_f(Conv1d(256, 1024, 41, 4, groups=64, padding=20)),
-            norm_f(Conv1d(1024, 1024, 41, 4, groups=256, padding=20)),
-            norm_f(Conv1d(1024, 1024, 5, 1, padding=2)),
-        ])
-        self.conv_post = norm_f(Conv1d(1024, 1, 3, 1, padding=1))
-
-    def forward(self, x):
-        fmap = []
-
-        for l in self.convs:
-            x = l(x)
-            x = F.leaky_relu(x, LRELU_SLOPE)
-            fmap.append(x)
-        x = self.conv_post(x)
-        fmap.append(x)
-        x = torch.flatten(x, 1, -1)
-
-        return x, fmap
-
-
-class MultiPeriodDiscriminator(torch.nn.Module):
-    def __init__(self, use_spectral_norm=False):
-        super(MultiPeriodDiscriminator, self).__init__()
-        periods = [2,3,5,7,11]
-
-        discs = [DiscriminatorS(use_spectral_norm=use_spectral_norm)]
-        discs = discs + [DiscriminatorP(i, use_spectral_norm=use_spectral_norm) for i in periods]
-        self.discriminators = nn.ModuleList(discs)
-
-    def forward(self, y, y_hat):
-        y_d_rs = []
-        y_d_gs = []
-        fmap_rs = []
-        fmap_gs = []
-        for i, d in enumerate(self.discriminators):
-            y_d_r, fmap_r = d(y)
-            y_d_g, fmap_g = d(y_hat)
-            y_d_rs.append(y_d_r)
-            y_d_gs.append(y_d_g)
-            fmap_rs.append(fmap_r)
-            fmap_gs.append(fmap_g)
-
-        return y_d_rs, y_d_gs, fmap_rs, fmap_gs
-
-class Vits(Base):
-  """
-  Synthesizer of Vits
-  """
-
-  def __init__(self, 
-    n_vocab,
-    spec_channels,
-    segment_size,
-    inter_channels,
-    hidden_channels,
-    filter_channels,
-    n_heads,
-    n_layers,
-    kernel_size,
-    p_dropout,
-    resblock, 
-    resblock_kernel_sizes, 
-    resblock_dilation_sizes, 
-    upsample_rates, 
-    upsample_initial_channel, 
-    upsample_kernel_sizes,
-    stop_threshold,
-    n_speakers=0,
-    gin_channels=0,
-    use_sdp=True,
-    **kwargs):
-
-    super().__init__(stop_threshold)
-    self.n_vocab = n_vocab
-    self.spec_channels = spec_channels
-    self.inter_channels = inter_channels
-    self.hidden_channels = hidden_channels
-    self.filter_channels = filter_channels
-    self.n_heads = n_heads
-    self.n_layers = n_layers
-    self.kernel_size = kernel_size
-    self.p_dropout = p_dropout
-    self.resblock = resblock
-    self.resblock_kernel_sizes = resblock_kernel_sizes
-    self.resblock_dilation_sizes = resblock_dilation_sizes
-    self.upsample_rates = upsample_rates
-    self.upsample_initial_channel = upsample_initial_channel
-    self.upsample_kernel_sizes = upsample_kernel_sizes
-    self.segment_size = segment_size
-    self.n_speakers = n_speakers
-    self.gin_channels = gin_channels
-
-    self.use_sdp = use_sdp
-
-    self.enc_p = TextEncoder(n_vocab,
-        inter_channels,
-        hidden_channels,
-        filter_channels,
-        n_heads,
-        n_layers,
-        kernel_size,
-        p_dropout)
-    self.dec = Generator(inter_channels, resblock, resblock_kernel_sizes, resblock_dilation_sizes, upsample_rates, upsample_initial_channel, upsample_kernel_sizes, gin_channels=gin_channels)
-    self.enc_q = PosteriorEncoder(spec_channels, inter_channels, hidden_channels, 5, 1, 16, gin_channels=gin_channels)
-    self.flow = ResidualCouplingBlock(inter_channels, hidden_channels, 5, 1, 4, gin_channels=gin_channels)
-
-    if use_sdp:
-      self.dp = StochasticDurationPredictor(hidden_channels, 192, 3, 0.5, 4, gin_channels=gin_channels)
-    else:
-      self.dp = DurationPredictor(hidden_channels, 256, 3, 0.5, gin_channels=gin_channels)
-
-    if n_speakers > 1:
-      self.emb_g = nn.Embedding(n_speakers, gin_channels)
-
-  def forward(self, x, x_lengths, y, y_lengths, sid=None, emo=None):
-
-    x, m_p, logs_p, x_mask = self.enc_p(x, x_lengths, emo)
-    if self.n_speakers > 0:
-      g = self.emb_g(sid).unsqueeze(-1) # [b, h, 1]
-    else:
-      g = None
-
-    z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g)
-    z_p = self.flow(z, y_mask, g=g)
-
-    with torch.no_grad():
-      # negative cross-entropy
-      s_p_sq_r = torch.exp(-2 * logs_p) # [b, d, t]
-      neg_cent1 = torch.sum(-0.5 * math.log(2 * math.pi) - logs_p, [1], keepdim=True) # [b, 1, t_s]
-      neg_cent2 = torch.matmul(-0.5 * (z_p ** 2).transpose(1, 2), s_p_sq_r) # [b, t_t, d] x [b, d, t_s] = [b, t_t, t_s]
-      neg_cent3 = torch.matmul(z_p.transpose(1, 2), (m_p * s_p_sq_r)) # [b, t_t, d] x [b, d, t_s] = [b, t_t, t_s]
-      neg_cent4 = torch.sum(-0.5 * (m_p ** 2) * s_p_sq_r, [1], keepdim=True) # [b, 1, t_s]
-      neg_cent = neg_cent1 + neg_cent2 + neg_cent3 + neg_cent4
-
-      attn_mask = torch.unsqueeze(x_mask, 2) * torch.unsqueeze(y_mask, -1)
-      attn = monotonic_align.maximum_path(neg_cent, attn_mask.squeeze(1)).unsqueeze(1).detach()
-
-    w = attn.sum(2)
-    if self.use_sdp:
-      l_length = self.dp(x, x_mask, w, g=g)
-      l_length = l_length / torch.sum(x_mask)
-    else:
-      logw_ = torch.log(w + 1e-6) * x_mask
-      logw = self.dp(x, x_mask, g=g)
-      l_length = torch.sum((logw - logw_)**2, [1,2]) / torch.sum(x_mask) # for averaging 
-
-    # expand prior
-    m_p = torch.matmul(attn.squeeze(1), m_p.transpose(1, 2)).transpose(1, 2)
-    logs_p = torch.matmul(attn.squeeze(1), logs_p.transpose(1, 2)).transpose(1, 2)
-
-    z_slice, ids_slice = rand_slice_segments(z, y_lengths, self.segment_size)
-    o = self.dec(z_slice, g=g)
-    return o, l_length, attn, ids_slice, x_mask, y_mask, (z, z_p, m_p, logs_p, m_q, logs_q)
-
-  def infer(self, x, x_lengths, sid=None, emo=None, noise_scale=1, length_scale=1, noise_scale_w=1., max_len=None):
-    x, m_p, logs_p, x_mask = self.enc_p(x, x_lengths,emo)
-    if self.n_speakers > 0:
-      g = self.emb_g(sid).unsqueeze(-1) # [b, h, 1]
-    else:
-      g = None
-
-    if self.use_sdp:
-      logw = self.dp(x, x_mask, g=g, reverse=True, noise_scale=noise_scale_w)
-    else:
-      logw = self.dp(x, x_mask, g=g)
-    w = torch.exp(logw) * x_mask * length_scale
-    w_ceil = torch.ceil(w)
-    y_lengths = torch.clamp_min(torch.sum(w_ceil, [1, 2]), 1).long()
-    y_mask = torch.unsqueeze(sequence_mask(y_lengths, None), 1).to(x_mask.dtype)
-    attn_mask = torch.unsqueeze(x_mask, 2) * torch.unsqueeze(y_mask, -1)
-    attn = generate_path(w_ceil, attn_mask)
-
-    m_p = torch.matmul(attn.squeeze(1), m_p.transpose(1, 2)).transpose(1, 2) # [b, t', t], [b, t, d] -> [b, d, t']
-    logs_p = torch.matmul(attn.squeeze(1), logs_p.transpose(1, 2)).transpose(1, 2) # [b, t', t], [b, t, d] -> [b, d, t']
-
-    z_p = m_p + torch.randn_like(m_p) * torch.exp(logs_p) * noise_scale
-    z = self.flow(z_p, y_mask, g=g, reverse=True)
-    o = self.dec((z * y_mask)[:,:,:max_len], g=g)
-    return o, attn, y_mask, (z, z_p, m_p, logs_p)
-

models/synthesizer/models/wav2emo.py

@@ -1,50 +0,0 @@
-import torch
-import torch.nn as nn
-from transformers.models.wav2vec2.modeling_wav2vec2 import (
-    Wav2Vec2Model,
-    Wav2Vec2PreTrainedModel,
-)
-
-
-class RegressionHead(nn.Module):
-    r"""Classification head."""
-
-    def __init__(self, config):
-        super().__init__()
-
-        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
-        self.dropout = nn.Dropout(config.final_dropout)
-        self.out_proj = nn.Linear(config.hidden_size, config.num_labels)
-
-    def forward(self, features, **kwargs):
-        x = features
-        x = self.dropout(x)
-        x = self.dense(x)
-        x = torch.tanh(x)
-        x = self.dropout(x)
-        x = self.out_proj(x)
-
-        return x
-
-
-class EmotionExtractorModel(Wav2Vec2PreTrainedModel):
-    r"""Speech emotion classifier."""
-
-    def __init__(self, config):
-        super().__init__(config)
-
-        self.config = config
-        self.wav2vec2 = Wav2Vec2Model(config)
-        self.classifier = RegressionHead(config)
-        self.init_weights()
-
-    def forward(
-            self,
-            input_values,
-    ):
-        outputs = self.wav2vec2(input_values)
-        hidden_states = outputs[0]
-        hidden_states = torch.mean(hidden_states, dim=1)
-        logits = self.classifier(hidden_states)
-
-        return hidden_states, logits

models/synthesizer/train_vits.py

@@ -1,389 +0,0 @@
-import os
-from loguru import logger
-import torch
-import glob
-from torch.nn import functional as F
-from torch.utils.data import DataLoader
-from torch.utils.tensorboard import SummaryWriter
-import torch.distributed as dist
-from torch.nn.parallel import DistributedDataParallel as DDP
-from torch.cuda.amp import autocast, GradScaler
-from utils.audio_utils import mel_spectrogram, spec_to_mel
-from utils.loss import feature_loss, generator_loss, discriminator_loss, kl_loss
-from utils.util import slice_segments, clip_grad_value_
-from models.synthesizer.vits_dataset import (
-    VitsDataset,
-    VitsDatasetCollate,
-    DistributedBucketSampler
-)
-from models.synthesizer.models.vits import (
-    Vits,
-    MultiPeriodDiscriminator,
-)
-from models.synthesizer.utils.symbols import symbols
-from models.synthesizer.utils.plot import plot_spectrogram_to_numpy, plot_alignment_to_numpy
-from pathlib import Path
-from utils.hparams import HParams
-import torch.multiprocessing as mp
-import argparse
-
-# torch.backends.cudnn.benchmark = True
-global_step = 0
-
-
-def new_train():
-    """Assume Single Node Multi GPUs Training Only"""
-    assert torch.cuda.is_available(), "CPU training is not allowed."
-
-    parser = argparse.ArgumentParser()
-    parser.add_argument("--syn_dir", type=str, default="../audiodata/SV2TTS/synthesizer", help= \
-        "Path to the synthesizer directory that contains the ground truth mel spectrograms, "
-        "the wavs, the emos and the embeds.")
-    parser.add_argument("-m", "--model_dir", type=str, default="data/ckpt/synthesizer/vits", help=\
-        "Path to the output directory that will contain the saved model weights and the logs.")
-    parser.add_argument('--ckptG', type=str, required=False,
-                      help='original VITS G checkpoint path')
-    parser.add_argument('--ckptD', type=str, required=False,
-                      help='original VITS D checkpoint path')
-    args, _ = parser.parse_known_args()
-
-    datasets_root = Path(args.syn_dir)
-    hparams= HParams(
-        model_dir = args.model_dir,
-    )
-    hparams.loadJson(Path(hparams.model_dir).joinpath("config.json"))
-    hparams.data["training_files"] = str(datasets_root.joinpath("train.txt"))
-    hparams.data["validation_files"] = str(datasets_root.joinpath("train.txt"))
-    hparams.data["datasets_root"] = str(datasets_root)
-    hparams.ckptG = args.ckptG
-    hparams.ckptD = args.ckptD
-    n_gpus = torch.cuda.device_count()
-    # for spawn
-    os.environ['MASTER_ADDR'] = 'localhost'
-    os.environ['MASTER_PORT'] = '8899'
-    # mp.spawn(run, nprocs=n_gpus, args=(n_gpus, hparams))
-    run(0, 1, hparams)
-
-
-def load_checkpoint(checkpoint_path, model, optimizer=None, is_old=False):
-  assert os.path.isfile(checkpoint_path)
-  checkpoint_dict = torch.load(checkpoint_path, map_location='cpu')
-  iteration = checkpoint_dict['iteration']
-  learning_rate = checkpoint_dict['learning_rate']
-  if optimizer is not None:
-    if not is_old:
-      optimizer.load_state_dict(checkpoint_dict['optimizer'])
-    else:
-      new_opt_dict = optimizer.state_dict()
-      new_opt_dict_params = new_opt_dict['param_groups'][0]['params']
-      new_opt_dict['param_groups'] = checkpoint_dict['optimizer']['param_groups']
-      new_opt_dict['param_groups'][0]['params'] = new_opt_dict_params
-      optimizer.load_state_dict(new_opt_dict)
-  saved_state_dict = checkpoint_dict['model']
-  if hasattr(model, 'module'):
-    state_dict = model.module.state_dict()
-  else:
-    state_dict = model.state_dict()
-  new_state_dict= {}
-  for k, v in state_dict.items():
-    try:
-      new_state_dict[k] = saved_state_dict[k]
-    except:
-      logger.info("%s is not in the checkpoint" % k)
-      new_state_dict[k] = v
-  if hasattr(model, 'module'):
-    model.module.load_state_dict(new_state_dict, strict=False)
-  else:
-    model.load_state_dict(new_state_dict, strict=False)
-  logger.info("Loaded checkpoint '{}' (iteration {})" .format(
-    checkpoint_path, iteration))
-  return model, optimizer, learning_rate, iteration
-
-
-def save_checkpoint(model, optimizer, learning_rate, iteration, checkpoint_path):
-  logger.info("Saving model and optimizer state at iteration {} to {}".format(
-    iteration, checkpoint_path))
-  if hasattr(model, 'module'):
-    state_dict = model.module.state_dict()
-  else:
-    state_dict = model.state_dict()
-  torch.save({'model': state_dict,
-              'iteration': iteration,
-              'optimizer': optimizer.state_dict(),
-              'learning_rate': learning_rate}, checkpoint_path)
-
-def latest_checkpoint_path(dir_path, regex="G_*.pth"):
-  f_list = glob.glob(os.path.join(dir_path, regex))
-  f_list.sort(key=lambda f: int("".join(filter(str.isdigit, f))))
-  x = f_list[-1]
-  print(x)
-  return x
-
-def run(rank, n_gpus, hps):
-    global global_step
-    if rank == 0:
-        logger.info(hps)
-        writer = SummaryWriter(log_dir=hps.model_dir)
-        writer_eval = SummaryWriter(log_dir=os.path.join(hps.model_dir, "eval"))
-
-    dist.init_process_group(backend='gloo', init_method='env://', world_size=n_gpus, rank=rank)
-    torch.manual_seed(hps.train.seed)
-    torch.cuda.set_device(rank)
-    train_dataset = VitsDataset(hps.data.training_files, hps.data)
-    train_sampler = DistributedBucketSampler(
-        train_dataset,
-        hps.train.batch_size,
-        [32, 300, 400, 500, 600, 700, 800, 900, 1000],
-        num_replicas=n_gpus,
-        rank=rank,
-        shuffle=True)
-    collate_fn = VitsDatasetCollate()
-    train_loader = DataLoader(train_dataset, num_workers=8, shuffle=False, pin_memory=True,
-                              collate_fn=collate_fn, batch_sampler=train_sampler)
-    if rank == 0:
-        eval_dataset = VitsDataset(hps.data.validation_files, hps.data)
-        eval_loader = DataLoader(eval_dataset, num_workers=8, shuffle=False,
-                                 batch_size=hps.train.batch_size, pin_memory=True,
-                                 drop_last=False, collate_fn=collate_fn)
-
-    net_g = Vits(
-        len(symbols),
-        hps.data.filter_length // 2 + 1,
-        hps.train.segment_size // hps.data.hop_length,
-        n_speakers=hps.data.n_speakers,
-        **hps.model).cuda(rank)
-    net_d = MultiPeriodDiscriminator(hps.model.use_spectral_norm).cuda(rank)
-    optim_g = torch.optim.AdamW(
-        net_g.parameters(),
-        hps.train.learning_rate,
-        betas=hps.train.betas,
-        eps=hps.train.eps)
-    optim_d = torch.optim.AdamW(
-        net_d.parameters(),
-        hps.train.learning_rate,
-        betas=hps.train.betas,
-        eps=hps.train.eps)
-    net_g = DDP(net_g, device_ids=[rank])
-    net_d = DDP(net_d, device_ids=[rank])
-    ckptG = hps.ckptG
-    ckptD = hps.ckptD
-    try:
-        if ckptG is not None:
-            _, _, _, epoch_str = load_checkpoint(ckptG, net_g, optim_g, is_old=True)
-            print("加载原版VITS模型G记录点成功")
-        else:
-            _, _, _, epoch_str = load_checkpoint(latest_checkpoint_path(hps.model_dir, "G_*.pth"), net_g,
-                                                   optim_g)
-        if ckptD is not None:
-            _, _, _, epoch_str = load_checkpoint(ckptG, net_g, optim_g, is_old=True)
-            print("加载原版VITS模型D记录点成功")
-        else:
-            _, _, _, epoch_str = load_checkpoint(latest_checkpoint_path(hps.model_dir, "D_*.pth"), net_d,
-                                                   optim_d)
-        global_step = (epoch_str - 1) * len(train_loader)
-    except:
-        epoch_str = 1
-        global_step = 0
-    if ckptG is not None or ckptD is not None:
-        epoch_str = 1
-        global_step = 0
-    scheduler_g = torch.optim.lr_scheduler.ExponentialLR(optim_g, gamma=hps.train.lr_decay, last_epoch=epoch_str - 2)
-    scheduler_d = torch.optim.lr_scheduler.ExponentialLR(optim_d, gamma=hps.train.lr_decay, last_epoch=epoch_str - 2)
-
-    scaler = GradScaler(enabled=hps.train.fp16_run)
-
-    for epoch in range(epoch_str, hps.train.epochs + 1):
-        if rank == 0:
-            train_and_evaluate(rank, epoch, hps, [net_g, net_d], [optim_g, optim_d], [scheduler_g, scheduler_d], scaler,
-                               [train_loader, eval_loader], logger, [writer, writer_eval])
-        else:
-            train_and_evaluate(rank, epoch, hps, [net_g, net_d], [optim_g, optim_d], [scheduler_g, scheduler_d], scaler,
-                               [train_loader, None], None, None)
-        scheduler_g.step()
-        scheduler_d.step()
-
-
-def train_and_evaluate(rank, epoch, hps, nets, optims, schedulers, scaler, loaders, logger, writers):
-    net_g, net_d = nets
-    optim_g, optim_d = optims
-    scheduler_g, scheduler_d = schedulers
-    train_loader, eval_loader = loaders
-    if writers is not None:
-        writer, writer_eval = writers
-    train_loader.batch_sampler.set_epoch(epoch)
-    global global_step
-
-    net_g.train()
-    net_d.train()
-    for batch_idx, (x, x_lengths, spec, spec_lengths, y, y_lengths, speakers, emo) in enumerate(train_loader):
-        logger.info(f'====> Step: 1 {batch_idx}')
-        x, x_lengths = x.cuda(rank, non_blocking=True), x_lengths.cuda(rank, non_blocking=True)
-        spec, spec_lengths = spec.cuda(rank, non_blocking=True), spec_lengths.cuda(rank, non_blocking=True)
-        y, y_lengths = y.cuda(rank, non_blocking=True), y_lengths.cuda(rank, non_blocking=True)
-        speakers = speakers.cuda(rank, non_blocking=True)
-        emo = emo.cuda(rank, non_blocking=True)
-
-        with autocast(enabled=hps.train.fp16_run):
-            y_hat, l_length, attn, ids_slice, x_mask, z_mask, \
-            (z, z_p, m_p, logs_p, m_q, logs_q) = net_g(x, x_lengths, spec, spec_lengths, speakers, emo)
-
-            mel = spec_to_mel(
-                spec,
-                hps.data.filter_length,
-                hps.data.n_mel_channels,
-                hps.data.sampling_rate,
-                hps.data.mel_fmin,
-                hps.data.mel_fmax)
-            y_mel = slice_segments(mel, ids_slice, hps.train.segment_size // hps.data.hop_length)
-            y_hat_mel = mel_spectrogram(
-                y_hat.squeeze(1),
-                hps.data.filter_length,
-                hps.data.n_mel_channels,
-                hps.data.sampling_rate,
-                hps.data.hop_length,
-                hps.data.win_length,
-                hps.data.mel_fmin,
-                hps.data.mel_fmax
-            )
-
-            y = slice_segments(y, ids_slice * hps.data.hop_length, hps.train.segment_size)  # slice
-
-            # Discriminator
-            y_d_hat_r, y_d_hat_g, _, _ = net_d(y, y_hat.detach())
-            with autocast(enabled=False):
-                loss_disc, losses_disc_r, losses_disc_g = discriminator_loss(y_d_hat_r, y_d_hat_g)
-                loss_disc_all = loss_disc
-        optim_d.zero_grad()
-        scaler.scale(loss_disc_all).backward()
-        scaler.unscale_(optim_d)
-        grad_norm_d = clip_grad_value_(net_d.parameters(), None)
-        scaler.step(optim_d)
-        logger.info(f'====> Step: 2 {batch_idx}')
-
-        with autocast(enabled=hps.train.fp16_run):
-            # Generator
-            y_d_hat_r, y_d_hat_g, fmap_r, fmap_g = net_d(y, y_hat)
-            with autocast(enabled=False):
-                loss_dur = torch.sum(l_length.float())
-                loss_mel = F.l1_loss(y_mel, y_hat_mel) * hps.train.c_mel
-                loss_kl = kl_loss(z_p, logs_q, m_p, logs_p, z_mask) * hps.train.c_kl
-
-                loss_fm = feature_loss(fmap_r, fmap_g)
-                loss_gen, losses_gen = generator_loss(y_d_hat_g)
-                loss_gen_all = loss_gen + loss_fm + loss_mel + loss_dur + loss_kl
-        optim_g.zero_grad()
-        scaler.scale(loss_gen_all.float()).backward()
-        scaler.unscale_(optim_g)
-        grad_norm_g = clip_grad_value_(net_g.parameters(), None)
-        scaler.step(optim_g)
-        scaler.update()
-        # logger.info(f'====> Step: 3 {batch_idx}')
-        if rank == 0:
-            if global_step % hps.train.log_interval == 0:
-                lr = optim_g.param_groups[0]['lr']
-                losses = [loss_disc, loss_gen, loss_fm, loss_mel, loss_dur, loss_kl]
-                logger.info('Train Epoch: {} [{:.0f}%]'.format(
-                    epoch,
-                    100. * batch_idx / len(train_loader)))
-                logger.info([x.item() for x in losses] + [global_step, lr])
-
-                scalar_dict = {"loss/g/total": loss_gen_all, "loss/d/total": loss_disc_all, "learning_rate": lr,
-                               "grad_norm_d": grad_norm_d, "grad_norm_g": grad_norm_g}
-                scalar_dict.update(
-                    {"loss/g/fm": loss_fm, "loss/g/mel": loss_mel, "loss/g/dur": loss_dur, "loss/g/kl": loss_kl})
-
-                scalar_dict.update({"loss/g/{}".format(i): v for i, v in enumerate(losses_gen)})
-                scalar_dict.update({"loss/d_r/{}".format(i): v for i, v in enumerate(losses_disc_r)})
-                scalar_dict.update({"loss/d_g/{}".format(i): v for i, v in enumerate(losses_disc_g)})
-                image_dict = {
-                    "slice/mel_org": plot_spectrogram_to_numpy(y_mel[0].data.cpu().numpy()),
-                    "slice/mel_gen": plot_spectrogram_to_numpy(y_hat_mel[0].data.cpu().numpy()),
-                    "all/mel": plot_spectrogram_to_numpy(mel[0].data.cpu().numpy()),
-                    "all/attn": plot_alignment_to_numpy(attn[0, 0].data.cpu().numpy())
-                }
-                summarize(
-                    writer=writer,
-                    global_step=global_step,
-                    images=image_dict,
-                    scalars=scalar_dict)
-
-            if global_step % hps.train.eval_interval == 0:
-                evaluate(hps, net_g, eval_loader, writer_eval)
-                save_checkpoint(net_g, optim_g, hps.train.learning_rate, epoch,
-                                      os.path.join(hps.model_dir, "G_{}.pth".format(global_step)))
-                save_checkpoint(net_d, optim_d, hps.train.learning_rate, epoch,
-                                      os.path.join(hps.model_dir, "D_{}.pth".format(global_step)))
-        global_step += 1
-
-    if rank == 0:
-        logger.info('====> Epoch: {}'.format(epoch))
-
-
-def evaluate(hps, generator, eval_loader, writer_eval):
-    generator.eval()
-    with torch.no_grad():
-        for batch_idx, (x, x_lengths, spec, spec_lengths, y, y_lengths, speakers, emo) in enumerate(eval_loader):
-            x, x_lengths = x.cuda(0), x_lengths.cuda(0)
-            spec, spec_lengths = spec.cuda(0), spec_lengths.cuda(0)
-            y, y_lengths = y.cuda(0), y_lengths.cuda(0)
-            speakers = speakers.cuda(0)
-            emo = emo.cuda(0)
-            # remove else
-            x = x[:1]
-            x_lengths = x_lengths[:1]
-            spec = spec[:1]
-            spec_lengths = spec_lengths[:1]
-            y = y[:1]
-            y_lengths = y_lengths[:1]
-            speakers = speakers[:1]
-            emo = emo[:1]
-            break
-        y_hat, attn, mask, *_ = generator.module.infer(x, x_lengths, speakers, emo, max_len=1000)
-        y_hat_lengths = mask.sum([1, 2]).long() * hps.data.hop_length
-
-        mel = spec_to_mel(
-            spec,
-            hps.data.filter_length,
-            hps.data.n_mel_channels,
-            hps.data.sampling_rate,
-            hps.data.mel_fmin,
-            hps.data.mel_fmax)
-        y_hat_mel = mel_spectrogram(
-            y_hat.squeeze(1).float(),
-            hps.data.filter_length,
-            hps.data.n_mel_channels,
-            hps.data.sampling_rate,
-            hps.data.hop_length,
-            hps.data.win_length,
-            hps.data.mel_fmin,
-            hps.data.mel_fmax
-        )
-    image_dict = {
-        "gen/mel": plot_spectrogram_to_numpy(y_hat_mel[0].cpu().numpy())
-    }
-    audio_dict = {
-        "gen/audio": y_hat[0, :, :y_hat_lengths[0]]
-    }
-    if global_step == 0:
-        image_dict.update({"gt/mel": plot_spectrogram_to_numpy(mel[0].cpu().numpy())})
-        audio_dict.update({"gt/audio": y[0, :, :y_lengths[0]]})
-
-    summarize(
-        writer=writer_eval,
-        global_step=global_step,
-        images=image_dict,
-        audios=audio_dict,
-        audio_sampling_rate=hps.data.sampling_rate
-    )
-    generator.train()
-
-def summarize(writer, global_step, scalars={}, histograms={}, images={}, audios={}, audio_sampling_rate=22050):
-    for k, v in scalars.items():
-        writer.add_scalar(k, v, global_step)
-    for k, v in histograms.items():
-        writer.add_histogram(k, v, global_step)
-    for k, v in images.items():
-        writer.add_image(k, v, global_step, dataformats='HWC')
-    for k, v in audios.items():
-        writer.add_audio(k, v, global_step, audio_sampling_rate)
-

models/synthesizer/vits_dataset.py

@@ -1,280 +0,0 @@
-import os
-import random
-import numpy as np
-import torch
-import torch.utils.data
-
-from utils.audio_utils import spectrogram, load_wav
-from utils.util import intersperse
-from models.synthesizer.utils.text import text_to_sequence
-
-
-"""Multi speaker version"""
-class VitsDataset(torch.utils.data.Dataset):
-    """
-        1) loads audio, speaker_id, text pairs
-        2) normalizes text and converts them to sequences of integers
-        3) computes spectrograms from audio files.
-    """
-    def __init__(self, audio_file_path, hparams):
-        with open(audio_file_path, encoding='utf-8') as f:
-            self.audio_metadata = [line.strip().split('|') for line in f]
-        self.text_cleaners = hparams.text_cleaners
-        self.max_wav_value = hparams.max_wav_value
-        self.sampling_rate = hparams.sampling_rate
-        self.filter_length  = hparams.filter_length
-        self.hop_length     = hparams.hop_length
-        self.win_length     = hparams.win_length
-        self.sampling_rate  = hparams.sampling_rate
-
-        self.cleaned_text = getattr(hparams, "cleaned_text", False)
-
-        self.add_blank = hparams.add_blank
-        self.datasets_root = hparams.datasets_root
-
-        self.min_text_len = getattr(hparams, "min_text_len", 1)
-        self.max_text_len = getattr(hparams, "max_text_len", 190)
-
-        random.seed(1234)
-        random.shuffle(self.audio_metadata)
-        self._filter()
-
-    def _filter(self):
-        """
-        Filter text & store spec lengths
-        """
-        # Store spectrogram lengths for Bucketing
-        # wav_length ~= file_size / (wav_channels * Bytes per dim) = file_size / (1 * 2)
-        # spec_length = wav_length // hop_length
-
-        audio_metadata_new = []
-        lengths = []
-        
-        # for audiopath, sid, text in self.audio_metadata:
-        sid = 0
-        spk_to_sid = {}
-        for wav_fpath, mel_fpath, embed_path, wav_length, mel_frames, text in self.audio_metadata:
-            if self.min_text_len <= len(text) and len(text) <= self.max_text_len:
-                # TODO: for magic data only
-                speaker_name = wav_fpath.split("_")[1]
-                if speaker_name not in spk_to_sid:
-                    sid += 1
-                    spk_to_sid[speaker_name] = sid
-                
-                audio_metadata_new.append([wav_fpath, mel_fpath, embed_path, wav_length, mel_frames, text, spk_to_sid[speaker_name]])
-                lengths.append(os.path.getsize(f'{self.datasets_root}{os.sep}audio{os.sep}{wav_fpath}') // (2 * self.hop_length))
-        print("found sid:%d", sid)
-        self.audio_metadata = audio_metadata_new
-        self.lengths = lengths
-
-    def get_audio_text_speaker_pair(self, audio_metadata):
-        # separate filename, speaker_id and text
-        wav_fpath, text, sid = audio_metadata[0], audio_metadata[5], audio_metadata[6]
-        text = self.get_text(text)
-        
-        spec, wav = self.get_audio(f'{self.datasets_root}{os.sep}audio{os.sep}{wav_fpath}')
-        sid = self.get_sid(sid)
-        emo = torch.FloatTensor(np.load(f'{self.datasets_root}{os.sep}emo{os.sep}{wav_fpath.replace("audio", "emo")}'))
-        return (text, spec, wav, sid, emo)
-
-    def get_audio(self, filename):
-        # audio, sampling_rate = load_wav(filename)
-
-        # if sampling_rate != self.sampling_rate:
-        #     raise ValueError("{} {} SR doesn't match target {} SR".format(
-        #         sampling_rate, self.sampling_rate))
-        # audio = torch.load(filename)
-        audio = torch.FloatTensor(np.load(filename).astype(np.float32)) 
-        audio = audio.unsqueeze(0)
-        # audio_norm = audio / self.max_wav_value
-        # audio_norm = audio_norm.unsqueeze(0)
-        # spec_filename = filename.replace(".wav", ".spec.pt")
-        # if os.path.exists(spec_filename):
-        #     spec = torch.load(spec_filename)
-        # else:
-        # spec = spectrogram(audio, self.filter_length,
-        #     self.sampling_rate, self.hop_length, self.win_length,
-        #     center=False)
-        # spec = torch.squeeze(spec, 0)
-        # torch.save(spec, spec_filename)
-        spec = spectrogram(audio, self.filter_length, self.hop_length, self.win_length,
-            center=False)
-        spec = torch.squeeze(spec, 0)
-        return spec, audio
-
-    def get_text(self, text):
-        if self.cleaned_text:
-            text_norm = text_to_sequence(text, self.text_cleaners)
-        if self.add_blank:
-            text_norm = intersperse(text_norm, 0)
-        text_norm = torch.LongTensor(text_norm)
-        return text_norm
-
-    def get_sid(self, sid):
-        sid = torch.LongTensor([int(sid)])
-        return sid
-
-    def __getitem__(self, index):
-        return self.get_audio_text_speaker_pair(self.audio_metadata[index])
-
-    def __len__(self):
-        return len(self.audio_metadata)
-
-
-class VitsDatasetCollate():
-    """ Zero-pads model inputs and targets
-    """
-    def __init__(self, return_ids=False):
-        self.return_ids = return_ids
-
-    def __call__(self, batch):
-        """Collate's training batch from normalized text, audio and speaker identities
-        PARAMS
-        ------
-        batch: [text_normalized, spec_normalized, wav_normalized, sid]
-        """
-        # Right zero-pad all one-hot text sequences to max input length
-        _, ids_sorted_decreasing = torch.sort(
-            torch.LongTensor([x[1].size(1) for x in batch]),
-            dim=0, descending=True)
-
-        max_text_len = max([len(x[0]) for x in batch])
-        max_spec_len = max([x[1].size(1) for x in batch])
-        max_wav_len = max([x[2].size(1) for x in batch])
-
-        text_lengths = torch.LongTensor(len(batch))
-        spec_lengths = torch.LongTensor(len(batch))
-        wav_lengths = torch.LongTensor(len(batch))
-        sid = torch.LongTensor(len(batch))
-
-        text_padded = torch.LongTensor(len(batch), max_text_len)
-        spec_padded = torch.FloatTensor(len(batch), batch[0][1].size(0), max_spec_len)
-        wav_padded = torch.FloatTensor(len(batch), 1, max_wav_len)
-        emo = torch.FloatTensor(len(batch), 1024)
-
-        text_padded.zero_()
-        spec_padded.zero_()
-        wav_padded.zero_()
-        emo.zero_()
-
-        for i in range(len(ids_sorted_decreasing)):
-            row = batch[ids_sorted_decreasing[i]]
-
-            text = row[0]
-            text_padded[i, :text.size(0)] = text
-            text_lengths[i] = text.size(0)
-
-            spec = row[1]
-            spec_padded[i, :, :spec.size(1)] = spec
-            spec_lengths[i] = spec.size(1)
-
-            wav = row[2]
-            wav_padded[i, :, :wav.size(1)] = wav
-            wav_lengths[i] = wav.size(1)
-
-            sid[i] = row[3]
-
-            emo[i, :] = row[4]
-
-        if self.return_ids:
-            return text_padded, text_lengths, spec_padded, spec_lengths, wav_padded, wav_lengths, sid, ids_sorted_decreasing
-        return text_padded, text_lengths, spec_padded, spec_lengths, wav_padded, wav_lengths, sid, emo
-
-
-class DistributedBucketSampler(torch.utils.data.distributed.DistributedSampler):
-    """
-    Maintain similar input lengths in a batch.
-    Length groups are specified by boundaries.
-    Ex) boundaries = [b1, b2, b3] -> any batch is included either {x | b1 < length(x) <=b2} or {x | b2 < length(x) <= b3}.
-  
-    It removes samples which are not included in the boundaries.
-    Ex) boundaries = [b1, b2, b3] -> any x s.t. length(x) <= b1 or length(x) > b3 are discarded.
-    """
-    def __init__(self, dataset, batch_size, boundaries, num_replicas=None, rank=None, shuffle=True):
-        super().__init__(dataset, num_replicas=num_replicas, rank=rank, shuffle=shuffle)
-        self.lengths = dataset.lengths
-        self.batch_size = batch_size
-        self.boundaries = boundaries
-  
-        self.buckets, self.num_samples_per_bucket = self._create_buckets()
-        self.total_size = sum(self.num_samples_per_bucket)
-        self.num_samples = self.total_size // self.num_replicas
-  
-    def _create_buckets(self):
-        buckets = [[] for _ in range(len(self.boundaries) - 1)]
-        for i in range(len(self.lengths)):
-            length = self.lengths[i]
-            idx_bucket = self._bisect(length)
-            if idx_bucket != -1:
-                buckets[idx_bucket].append(i)
-  
-        for i in range(len(buckets) - 1, 0, -1):
-            if len(buckets[i]) == 0:
-                buckets.pop(i)
-                self.boundaries.pop(i+1)
-  
-        num_samples_per_bucket = []
-        for i in range(len(buckets)):
-            len_bucket = len(buckets[i])
-            total_batch_size = self.num_replicas * self.batch_size
-            rem = (total_batch_size - (len_bucket % total_batch_size)) % total_batch_size
-            num_samples_per_bucket.append(len_bucket + rem)
-        return buckets, num_samples_per_bucket
-  
-    def __iter__(self):
-      # deterministically shuffle based on epoch
-      g = torch.Generator()
-      g.manual_seed(self.epoch)
-  
-      indices = []
-      if self.shuffle:
-          for bucket in self.buckets:
-              indices.append(torch.randperm(len(bucket), generator=g).tolist())
-      else:
-          for bucket in self.buckets:
-              indices.append(list(range(len(bucket))))
-  
-      batches = []
-      for i in range(len(self.buckets)):
-          bucket = self.buckets[i]
-          len_bucket = len(bucket)
-          ids_bucket = indices[i]
-          num_samples_bucket = self.num_samples_per_bucket[i]
-  
-          # add extra samples to make it evenly divisible
-          rem = num_samples_bucket - len_bucket
-          ids_bucket = ids_bucket + ids_bucket * (rem // len_bucket) + ids_bucket[:(rem % len_bucket)]
-  
-          # subsample
-          ids_bucket = ids_bucket[self.rank::self.num_replicas]
-  
-          # batching
-          for j in range(len(ids_bucket) // self.batch_size):
-              batch = [bucket[idx] for idx in ids_bucket[j*self.batch_size:(j+1)*self.batch_size]]
-              batches.append(batch)
-  
-      if self.shuffle:
-          batch_ids = torch.randperm(len(batches), generator=g).tolist()
-          batches = [batches[i] for i in batch_ids]
-      self.batches = batches
-  
-      assert len(self.batches) * self.batch_size == self.num_samples
-      return iter(self.batches)
-  
-    def _bisect(self, x, lo=0, hi=None):
-      if hi is None:
-          hi = len(self.boundaries) - 1
-  
-      if hi > lo:
-          mid = (hi + lo) // 2
-          if self.boundaries[mid] < x and x <= self.boundaries[mid+1]:
-              return mid
-          elif x <= self.boundaries[mid]:
-              return self._bisect(x, lo, mid)
-          else:
-              return self._bisect(x, mid + 1, hi)
-      else:
-          return -1
-
-    def __len__(self):
-        return self.num_samples // self.batch_size

models/vocoder/__init__.py

@@ -1 +0,0 @@
-#

models/vocoder/fregan/.gitignore

@@ -1,129 +0,0 @@
-# Byte-compiled / optimized / DLL files
-__pycache__/
-*.py[cod]
-*$py.class
-
-# C extensions
-*.so
-
-# Distribution / packaging
-.Python
-build/
-develop-eggs/
-dist/
-downloads/
-eggs/
-.eggs/
-lib/
-lib64/
-parts/
-sdist/
-var/
-wheels/
-pip-wheel-metadata/
-share/python-wheels/
-*.egg-info/
-.installed.cfg
-*.egg
-MANIFEST
-
-# PyInstaller
-#  Usually these files are written by a python script from a template
-#  before PyInstaller builds the exe, so as to inject date/other infos into it.
-*.manifest
-*.spec
-
-# Installer logs
-pip-log.txt
-pip-delete-this-directory.txt
-
-# Unit test / coverage reports
-htmlcov/
-.tox/
-.nox/
-.coverage
-.coverage.*
-.cache
-nosetests.xml
-coverage.xml
-*.cover
-*.py,cover
-.hypothesis/
-.pytest_cache/
-
-# Translations
-*.mo
-*.pot
-
-# Django stuff:
-*.log
-local_settings.py
-db.sqlite3
-db.sqlite3-journal
-
-# Flask stuff:
-instance/
-.webassets-cache
-
-# Scrapy stuff:
-.scrapy
-
-# Sphinx documentation
-docs/_build/
-
-# PyBuilder
-target/
-
-# Jupyter Notebook
-.ipynb_checkpoints
-
-# IPython
-profile_default/
-ipython_config.py
-
-# pyenv
-.python-version
-
-# pipenv
-#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
-#   However, in case of collaboration, if having platform-specific dependencies or dependencies
-#   having no cross-platform support, pipenv may install dependencies that don't work, or not
-#   install all needed dependencies.
-#Pipfile.lock
-
-# PEP 582; used by e.g. github.com/David-OConnor/pyflow
-__pypackages__/
-
-# Celery stuff
-celerybeat-schedule
-celerybeat.pid
-
-# SageMath parsed files
-*.sage.py
-
-# Environments
-.env
-.venv
-env/
-venv/
-ENV/
-env.bak/
-venv.bak/
-
-# Spyder project settings
-.spyderproject
-.spyproject
-
-# Rope project settings
-.ropeproject
-
-# mkdocs documentation
-/site
-
-# mypy
-.mypy_cache/
-.dmypy.json
-dmypy.json
-
-# Pyre type checker
-.pyre/

models/vocoder/fregan/LICENSE

@@ -1,21 +0,0 @@
-MIT License
-
-Copyright (c) 2021 Rishikesh (ऋषिकेश)
-
-Permission is hereby granted, free of charge, to any person obtaining a copy
-of this software and associated documentation files (the "Software"), to deal
-in the Software without restriction, including without limitation the rights
-to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
-copies of the Software, and to permit persons to whom the Software is
-furnished to do so, subject to the following conditions:
-
-The above copyright notice and this permission notice shall be included in all
-copies or substantial portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
-IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
-FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
-AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
-LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
-OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
-SOFTWARE.

models/vocoder/fregan/__init__.py

@@ -1 +0,0 @@
-#

models/vocoder/fregan/config.json

@@ -1,42 +0,0 @@
-{
-    "resblock": "1",
-    "num_gpus": 0,
-    "batch_size": 16,
-    "learning_rate": 0.0002,
-    "adam_b1": 0.8,
-    "adam_b2": 0.99,
-    "lr_decay": 0.999,
-    "seed": 1234,
-    "disc_start_step":0,
-
-
-    "upsample_rates": [5,5,2,2,2],
-    "upsample_kernel_sizes": [10,10,4,4,4],
-    "upsample_initial_channel": 512,
-    "resblock_kernel_sizes": [3,7,11],
-    "resblock_dilation_sizes": [[1, 3, 5, 7], [1,3,5,7], [1,3,5,7]],
-
-    "segment_size": 6400,
-    "num_mels": 80,
-    "num_freq": 1025,
-    "n_fft": 1024,
-    "hop_size": 200,
-    "win_size": 800,
-
-    "sampling_rate": 16000,
-
-    "fmin": 0,
-    "fmax": 7600,
-    "fmax_for_loss": null,
-
-    "num_workers": 4,
-
-    "dist_config": {
-        "dist_backend": "nccl",
-        "dist_url": "tcp://localhost:54321",
-        "world_size": 1
-    }
-
-
-
-}

models/vocoder/fregan/discriminator.py

@@ -1,303 +0,0 @@
-import torch
-import torch.nn.functional as F
-import torch.nn as nn
-from torch.nn import Conv1d, AvgPool1d, Conv2d
-from torch.nn.utils import weight_norm, spectral_norm
-from models.vocoder.fregan.utils import get_padding
-from models.vocoder.fregan.stft_loss import stft
-from models.vocoder.fregan.dwt import DWT_1D
-LRELU_SLOPE = 0.1
-
-
-
-class SpecDiscriminator(nn.Module):
-    """docstring for Discriminator."""
-
-    def __init__(self, fft_size=1024, shift_size=120, win_length=600, window="hann_window", use_spectral_norm=False):
-        super(SpecDiscriminator, self).__init__()
-        norm_f = weight_norm if use_spectral_norm == False else spectral_norm
-        self.fft_size = fft_size
-        self.shift_size = shift_size
-        self.win_length = win_length
-        self.window = getattr(torch, window)(win_length)
-        self.discriminators = nn.ModuleList([
-            norm_f(nn.Conv2d(1, 32, kernel_size=(3, 9), padding=(1, 4))),
-            norm_f(nn.Conv2d(32, 32, kernel_size=(3, 9), stride=(1,2), padding=(1, 4))),
-            norm_f(nn.Conv2d(32, 32, kernel_size=(3, 9), stride=(1,2), padding=(1, 4))),
-            norm_f(nn.Conv2d(32, 32, kernel_size=(3, 9), stride=(1,2), padding=(1, 4))),
-            norm_f(nn.Conv2d(32, 32, kernel_size=(3, 3), stride=(1,1), padding=(1, 1))),
-        ])
-
-        self.out = norm_f(nn.Conv2d(32, 1, 3, 1, 1))
-
-    def forward(self, y):
-
-        fmap = []
-        with torch.no_grad():
-            y = y.squeeze(1)
-            y = stft(y, self.fft_size, self.shift_size, self.win_length, self.window.to(y.get_device()))
-        y = y.unsqueeze(1)
-        for i, d in enumerate(self.discriminators):
-            y = d(y)
-            y = F.leaky_relu(y, LRELU_SLOPE)
-            fmap.append(y)
-
-        y = self.out(y)
-        fmap.append(y)
-
-        return torch.flatten(y, 1, -1), fmap
-
-class MultiResSpecDiscriminator(torch.nn.Module):
-
-    def __init__(self,
-                 fft_sizes=[1024, 2048, 512],
-                 hop_sizes=[120, 240, 50],
-                 win_lengths=[600, 1200, 240],
-                 window="hann_window"):
-
-        super(MultiResSpecDiscriminator, self).__init__()
-        self.discriminators = nn.ModuleList([
-            SpecDiscriminator(fft_sizes[0], hop_sizes[0], win_lengths[0], window),
-            SpecDiscriminator(fft_sizes[1], hop_sizes[1], win_lengths[1], window),
-            SpecDiscriminator(fft_sizes[2], hop_sizes[2], win_lengths[2], window)
-            ])
-
-    def forward(self, y, y_hat):
-        y_d_rs = []
-        y_d_gs = []
-        fmap_rs = []
-        fmap_gs = []
-        for i, d in enumerate(self.discriminators):
-            y_d_r, fmap_r = d(y)
-            y_d_g, fmap_g = d(y_hat)
-            y_d_rs.append(y_d_r)
-            fmap_rs.append(fmap_r)
-            y_d_gs.append(y_d_g)
-            fmap_gs.append(fmap_g)
-
-        return y_d_rs, y_d_gs, fmap_rs, fmap_gs
-
-
-class DiscriminatorP(torch.nn.Module):
-    def __init__(self, period, kernel_size=5, stride=3, use_spectral_norm=False):
-        super(DiscriminatorP, self).__init__()
-        self.period = period
-        norm_f = weight_norm if use_spectral_norm == False else spectral_norm
-        self.dwt1d = DWT_1D()
-        self.dwt_conv1 = norm_f(Conv1d(2, 1, 1))
-        self.dwt_proj1 = norm_f(Conv2d(1, 32, (kernel_size, 1), (stride, 1), padding=(get_padding(5, 1), 0)))
-        self.dwt_conv2 = norm_f(Conv1d(4, 1, 1))
-        self.dwt_proj2 = norm_f(Conv2d(1, 128, (kernel_size, 1), (stride, 1), padding=(get_padding(5, 1), 0)))
-        self.dwt_conv3 = norm_f(Conv1d(8, 1, 1))
-        self.dwt_proj3 = norm_f(Conv2d(1, 512, (kernel_size, 1), (stride, 1), padding=(get_padding(5, 1), 0)))
-        self.convs = nn.ModuleList([
-            norm_f(Conv2d(1, 32, (kernel_size, 1), (stride, 1), padding=(get_padding(5, 1), 0))),
-            norm_f(Conv2d(32, 128, (kernel_size, 1), (stride, 1), padding=(get_padding(5, 1), 0))),
-            norm_f(Conv2d(128, 512, (kernel_size, 1), (stride, 1), padding=(get_padding(5, 1), 0))),
-            norm_f(Conv2d(512, 1024, (kernel_size, 1), (stride, 1), padding=(get_padding(5, 1), 0))),
-            norm_f(Conv2d(1024, 1024, (kernel_size, 1), 1, padding=(2, 0))),
-        ])
-        self.conv_post = norm_f(Conv2d(1024, 1, (3, 1), 1, padding=(1, 0)))
-
-    def forward(self, x):
-        fmap = []
-
-        # DWT 1
-        x_d1_high1, x_d1_low1 = self.dwt1d(x)
-        x_d1 = self.dwt_conv1(torch.cat([x_d1_high1, x_d1_low1], dim=1))
-        # 1d to 2d
-        b, c, t = x_d1.shape
-        if t % self.period != 0:  # pad first
-            n_pad = self.period - (t % self.period)
-            x_d1 = F.pad(x_d1, (0, n_pad), "reflect")
-            t = t + n_pad
-        x_d1 = x_d1.view(b, c, t // self.period, self.period)
-
-        x_d1 = self.dwt_proj1(x_d1)
-
-        # DWT 2
-        x_d2_high1, x_d2_low1 = self.dwt1d(x_d1_high1)
-        x_d2_high2, x_d2_low2 = self.dwt1d(x_d1_low1)
-        x_d2 = self.dwt_conv2(torch.cat([x_d2_high1, x_d2_low1, x_d2_high2, x_d2_low2], dim=1))
-        # 1d to 2d
-        b, c, t = x_d2.shape
-        if t % self.period != 0:  # pad first
-            n_pad = self.period - (t % self.period)
-            x_d2 = F.pad(x_d2, (0, n_pad), "reflect")
-            t = t + n_pad
-        x_d2 = x_d2.view(b, c, t // self.period, self.period)
-
-        x_d2 = self.dwt_proj2(x_d2)
-
-        # DWT 3
-
-        x_d3_high1, x_d3_low1 = self.dwt1d(x_d2_high1)
-        x_d3_high2, x_d3_low2 = self.dwt1d(x_d2_low1)
-        x_d3_high3, x_d3_low3 = self.dwt1d(x_d2_high2)
-        x_d3_high4, x_d3_low4 = self.dwt1d(x_d2_low2)
-        x_d3 = self.dwt_conv3(
-            torch.cat([x_d3_high1, x_d3_low1, x_d3_high2, x_d3_low2, x_d3_high3, x_d3_low3, x_d3_high4, x_d3_low4],
-                      dim=1))
-        # 1d to 2d
-        b, c, t = x_d3.shape
-        if t % self.period != 0:  # pad first
-            n_pad = self.period - (t % self.period)
-            x_d3 = F.pad(x_d3, (0, n_pad), "reflect")
-            t = t + n_pad
-        x_d3 = x_d3.view(b, c, t // self.period, self.period)
-
-        x_d3 = self.dwt_proj3(x_d3)
-
-        # 1d to 2d
-        b, c, t = x.shape
-        if t % self.period != 0:  # pad first
-            n_pad = self.period - (t % self.period)
-            x = F.pad(x, (0, n_pad), "reflect")
-            t = t + n_pad
-        x = x.view(b, c, t // self.period, self.period)
-        i = 0
-        for l in self.convs:
-            x = l(x)
-            x = F.leaky_relu(x, LRELU_SLOPE)
-
-            fmap.append(x)
-            if i == 0:
-                x = torch.cat([x, x_d1], dim=2)
-            elif i == 1:
-                x = torch.cat([x, x_d2], dim=2)
-            elif i == 2:
-                x = torch.cat([x, x_d3], dim=2)
-            else:
-                x = x
-            i = i + 1
-        x = self.conv_post(x)
-        fmap.append(x)
-        x = torch.flatten(x, 1, -1)
-
-        return x, fmap
-
-
-class ResWiseMultiPeriodDiscriminator(torch.nn.Module):
-    def __init__(self):
-        super(ResWiseMultiPeriodDiscriminator, self).__init__()
-        self.discriminators = nn.ModuleList([
-            DiscriminatorP(2),
-            DiscriminatorP(3),
-            DiscriminatorP(5),
-            DiscriminatorP(7),
-            DiscriminatorP(11),
-        ])
-
-    def forward(self, y, y_hat):
-        y_d_rs = []
-        y_d_gs = []
-        fmap_rs = []
-        fmap_gs = []
-        for i, d in enumerate(self.discriminators):
-            y_d_r, fmap_r = d(y)
-            y_d_g, fmap_g = d(y_hat)
-            y_d_rs.append(y_d_r)
-            fmap_rs.append(fmap_r)
-            y_d_gs.append(y_d_g)
-            fmap_gs.append(fmap_g)
-
-        return y_d_rs, y_d_gs, fmap_rs, fmap_gs
-
-
-class DiscriminatorS(torch.nn.Module):
-    def __init__(self, use_spectral_norm=False):
-        super(DiscriminatorS, self).__init__()
-        norm_f = weight_norm if use_spectral_norm == False else spectral_norm
-        self.dwt1d = DWT_1D()
-        self.dwt_conv1 = norm_f(Conv1d(2, 128, 15, 1, padding=7))
-        self.dwt_conv2 = norm_f(Conv1d(4, 128, 41, 2, padding=20))
-        self.convs = nn.ModuleList([
-            norm_f(Conv1d(1, 128, 15, 1, padding=7)),
-            norm_f(Conv1d(128, 128, 41, 2, groups=4, padding=20)),
-            norm_f(Conv1d(128, 256, 41, 2, groups=16, padding=20)),
-            norm_f(Conv1d(256, 512, 41, 4, groups=16, padding=20)),
-            norm_f(Conv1d(512, 1024, 41, 4, groups=16, padding=20)),
-            norm_f(Conv1d(1024, 1024, 41, 1, groups=16, padding=20)),
-            norm_f(Conv1d(1024, 1024, 5, 1, padding=2)),
-        ])
-        self.conv_post = norm_f(Conv1d(1024, 1, 3, 1, padding=1))
-
-    def forward(self, x):
-        fmap = []
-
-        # DWT 1
-        x_d1_high1, x_d1_low1 = self.dwt1d(x)
-        x_d1 = self.dwt_conv1(torch.cat([x_d1_high1, x_d1_low1], dim=1))
-
-        # DWT 2
-        x_d2_high1, x_d2_low1 = self.dwt1d(x_d1_high1)
-        x_d2_high2, x_d2_low2 = self.dwt1d(x_d1_low1)
-        x_d2 = self.dwt_conv2(torch.cat([x_d2_high1, x_d2_low1, x_d2_high2, x_d2_low2], dim=1))
-
-        i = 0
-        for l in self.convs:
-            x = l(x)
-            x = F.leaky_relu(x, LRELU_SLOPE)
-            fmap.append(x)
-            if i == 0:
-                x = torch.cat([x, x_d1], dim=2)
-            if i == 1:
-                x = torch.cat([x, x_d2], dim=2)
-            i = i + 1
-        x = self.conv_post(x)
-        fmap.append(x)
-        x = torch.flatten(x, 1, -1)
-
-        return x, fmap
-
-
-class ResWiseMultiScaleDiscriminator(torch.nn.Module):
-    def __init__(self, use_spectral_norm=False):
-        super(ResWiseMultiScaleDiscriminator, self).__init__()
-        norm_f = weight_norm if use_spectral_norm == False else spectral_norm
-        self.dwt1d = DWT_1D()
-        self.dwt_conv1 = norm_f(Conv1d(2, 1, 1))
-        self.dwt_conv2 = norm_f(Conv1d(4, 1, 1))
-        self.discriminators = nn.ModuleList([
-            DiscriminatorS(use_spectral_norm=True),
-            DiscriminatorS(),
-            DiscriminatorS(),
-        ])
-
-    def forward(self, y, y_hat):
-        y_d_rs = []
-        y_d_gs = []
-        fmap_rs = []
-        fmap_gs = []
-        # DWT 1
-        y_hi, y_lo = self.dwt1d(y)
-        y_1 = self.dwt_conv1(torch.cat([y_hi, y_lo], dim=1))
-        x_d1_high1, x_d1_low1 = self.dwt1d(y_hat)
-        y_hat_1 = self.dwt_conv1(torch.cat([x_d1_high1, x_d1_low1], dim=1))
-
-        # DWT 2
-        x_d2_high1, x_d2_low1 = self.dwt1d(y_hi)
-        x_d2_high2, x_d2_low2 = self.dwt1d(y_lo)
-        y_2 = self.dwt_conv2(torch.cat([x_d2_high1, x_d2_low1, x_d2_high2, x_d2_low2], dim=1))
-
-        x_d2_high1, x_d2_low1 = self.dwt1d(x_d1_high1)
-        x_d2_high2, x_d2_low2 = self.dwt1d(x_d1_low1)
-        y_hat_2 = self.dwt_conv2(torch.cat([x_d2_high1, x_d2_low1, x_d2_high2, x_d2_low2], dim=1))
-
-        for i, d in enumerate(self.discriminators):
-
-            if i == 1:
-                y = y_1
-                y_hat = y_hat_1
-            if i == 2:
-                y = y_2
-                y_hat = y_hat_2
-
-            y_d_r, fmap_r = d(y)
-            y_d_g, fmap_g = d(y_hat)
-            y_d_rs.append(y_d_r)
-            fmap_rs.append(fmap_r)
-            y_d_gs.append(y_d_g)
-            fmap_gs.append(fmap_g)
-
-        return y_d_rs, y_d_gs, fmap_rs, fmap_gs

models/vocoder/fregan/dwt.py

@@ -1,76 +0,0 @@
-# Copyright (c) 2019, Adobe Inc. All rights reserved.
-#
-# This work is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike
-# 4.0 International Public License. To view a copy of this license, visit
-# https://creativecommons.org/licenses/by-nc-sa/4.0/legalcode.
-
-# DWT code borrow from https://github.com/LiQiufu/WaveSNet/blob/12cb9d24208c3d26917bf953618c30f0c6b0f03d/DWT_IDWT/DWT_IDWT_layer.py
-
-
-import pywt
-import torch
-import torch.nn as nn
-import torch.nn.functional as F
-
-__all__ = ['DWT_1D']
-Pad_Mode = ['constant', 'reflect', 'replicate', 'circular']
-
-
-class DWT_1D(nn.Module):
-    def __init__(self, pad_type='reflect', wavename='haar',
-                 stride=2, in_channels=1, out_channels=None, groups=None,
-                 kernel_size=None, trainable=False):
-
-        super(DWT_1D, self).__init__()
-        self.trainable = trainable
-        self.kernel_size = kernel_size
-        if not self.trainable:
-            assert self.kernel_size == None
-        self.in_channels = in_channels
-        self.out_channels = self.in_channels if out_channels == None else out_channels
-        self.groups = self.in_channels if groups == None else groups
-        assert isinstance(self.groups, int) and self.in_channels % self.groups == 0
-        self.stride = stride
-        assert self.stride == 2
-        self.wavename = wavename
-        self.pad_type = pad_type
-        assert self.pad_type in Pad_Mode
-        self.get_filters()
-        self.initialization()
-
-    def get_filters(self):
-        wavelet = pywt.Wavelet(self.wavename)
-        band_low = torch.tensor(wavelet.rec_lo)
-        band_high = torch.tensor(wavelet.rec_hi)
-        length_band = band_low.size()[0]
-        self.kernel_size = length_band if self.kernel_size == None else self.kernel_size
-        assert self.kernel_size >= length_band
-        a = (self.kernel_size - length_band) // 2
-        b = - (self.kernel_size - length_band - a)
-        b = None if b == 0 else b
-        self.filt_low = torch.zeros(self.kernel_size)
-        self.filt_high = torch.zeros(self.kernel_size)
-        self.filt_low[a:b] = band_low
-        self.filt_high[a:b] = band_high
-
-    def initialization(self):
-        self.filter_low = self.filt_low[None, None, :].repeat((self.out_channels, self.in_channels // self.groups, 1))
-        self.filter_high = self.filt_high[None, None, :].repeat((self.out_channels, self.in_channels // self.groups, 1))
-        if torch.cuda.is_available():
-            self.filter_low = self.filter_low.cuda()
-            self.filter_high = self.filter_high.cuda()
-        if self.trainable:
-            self.filter_low = nn.Parameter(self.filter_low)
-            self.filter_high = nn.Parameter(self.filter_high)
-        if self.kernel_size % 2 == 0:
-            self.pad_sizes = [self.kernel_size // 2 - 1, self.kernel_size // 2 - 1]
-        else:
-            self.pad_sizes = [self.kernel_size // 2, self.kernel_size // 2]
-
-    def forward(self, input):
-        assert isinstance(input, torch.Tensor)
-        assert len(input.size()) == 3
-        assert input.size()[1] == self.in_channels
-        input = F.pad(input, pad=self.pad_sizes, mode=self.pad_type)
-        return F.conv1d(input, self.filter_low.to(input.device), stride=self.stride, groups=self.groups), \
-               F.conv1d(input, self.filter_high.to(input.device), stride=self.stride, groups=self.groups)

models/vocoder/fregan/generator.py

@@ -1,210 +0,0 @@
-import torch
-import torch.nn.functional as F
-import torch.nn as nn
-from torch.nn import Conv1d, ConvTranspose1d, AvgPool1d, Conv2d
-from torch.nn.utils import weight_norm, remove_weight_norm, spectral_norm
-from utils.util import init_weights, get_padding
-
-LRELU_SLOPE = 0.1
-
-
-class ResBlock1(torch.nn.Module):
-    def __init__(self, h, channels, kernel_size=3, dilation=(1, 3, 5, 7)):
-        super(ResBlock1, self).__init__()
-        self.h = h
-        self.convs1 = nn.ModuleList([
-            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=dilation[0],
-                               padding=get_padding(kernel_size, dilation[0]))),
-            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=dilation[1],
-                               padding=get_padding(kernel_size, dilation[1]))),
-            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=dilation[2],
-                               padding=get_padding(kernel_size, dilation[2]))),
-            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=dilation[3],
-                               padding=get_padding(kernel_size, dilation[3])))
-        ])
-        self.convs1.apply(init_weights)
-
-        self.convs2 = nn.ModuleList([
-            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=1,
-                               padding=get_padding(kernel_size, 1))),
-            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=1,
-                               padding=get_padding(kernel_size, 1))),
-            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=1,
-                               padding=get_padding(kernel_size, 1))),
-            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=1,
-                               padding=get_padding(kernel_size, 1)))
-        ])
-        self.convs2.apply(init_weights)
-
-    def forward(self, x):
-        for c1, c2 in zip(self.convs1, self.convs2):
-            xt = F.leaky_relu(x, LRELU_SLOPE)
-            xt = c1(xt)
-            xt = F.leaky_relu(xt, LRELU_SLOPE)
-            xt = c2(xt)
-            x = xt + x
-        return x
-
-    def remove_weight_norm(self):
-        for l in self.convs1:
-            remove_weight_norm(l)
-        for l in self.convs2:
-            remove_weight_norm(l)
-
-
-class ResBlock2(torch.nn.Module):
-    def __init__(self, h, channels, kernel_size=3, dilation=(1, 3)):
-        super(ResBlock2, self).__init__()
-        self.h = h
-        self.convs = nn.ModuleList([
-            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=dilation[0],
-                               padding=get_padding(kernel_size, dilation[0]))),
-            weight_norm(Conv1d(channels, channels, kernel_size, 1, dilation=dilation[1],
-                               padding=get_padding(kernel_size, dilation[1])))
-        ])
-        self.convs.apply(init_weights)
-
-    def forward(self, x):
-        for c in self.convs:
-            xt = F.leaky_relu(x, LRELU_SLOPE)
-            xt = c(xt)
-            x = xt + x
-        return x
-
-    def remove_weight_norm(self):
-        for l in self.convs:
-            remove_weight_norm(l)
-
-
-class FreGAN(torch.nn.Module):
-    def __init__(self, h, top_k=4):
-        super(FreGAN, self).__init__()
-        self.h = h
-
-        self.num_kernels = len(h.resblock_kernel_sizes)
-        self.num_upsamples = len(h.upsample_rates)
-        self.upsample_rates = h.upsample_rates
-        self.up_kernels = h.upsample_kernel_sizes
-        self.cond_level = self.num_upsamples - top_k
-        self.conv_pre = weight_norm(Conv1d(80, h.upsample_initial_channel, 7, 1, padding=3))
-        resblock = ResBlock1 if h.resblock == '1' else ResBlock2
-
-        self.ups = nn.ModuleList()
-        self.cond_up = nn.ModuleList()
-        self.res_output = nn.ModuleList()
-        upsample_ = 1
-        kr = 80
-
-        for i, (u, k) in enumerate(zip(self.upsample_rates, self.up_kernels)):
-#            self.ups.append(weight_norm(
- #               ConvTranspose1d(h.upsample_initial_channel // (2 ** i), h.upsample_initial_channel // (2 ** (i + 1)),
- #                               k, u, padding=(k - u) // 2)))
-            self.ups.append(weight_norm(ConvTranspose1d(h.upsample_initial_channel//(2**i),
-                            h.upsample_initial_channel//(2**(i+1)),
-                            k, u, padding=(u//2 + u%2), output_padding=u%2)))
-
-            if i > (self.num_upsamples - top_k):
-                self.res_output.append(
-                    nn.Sequential(
-                        nn.Upsample(scale_factor=u, mode='nearest'),
-                        weight_norm(nn.Conv1d(h.upsample_initial_channel // (2 ** i),
-                                              h.upsample_initial_channel // (2 ** (i + 1)), 1))
-                    )
-                )
-            if i >= (self.num_upsamples - top_k):
-                self.cond_up.append(
-                    weight_norm(
-                        ConvTranspose1d(kr, h.upsample_initial_channel // (2 ** i),
-                                        self.up_kernels[i - 1], self.upsample_rates[i - 1],
-                                        padding=(self.upsample_rates[i-1]//2+self.upsample_rates[i-1]%2), output_padding=self.upsample_rates[i-1]%2))
-                )
-                kr = h.upsample_initial_channel // (2 ** i)
-
-            upsample_ *= u
-
-        self.resblocks = nn.ModuleList()
-        for i in range(len(self.ups)):
-            ch = h.upsample_initial_channel // (2 ** (i + 1))
-            for j, (k, d) in enumerate(zip(h.resblock_kernel_sizes, h.resblock_dilation_sizes)):
-                self.resblocks.append(resblock(h, ch, k, d))
-
-        self.conv_post = weight_norm(Conv1d(ch, 1, 7, 1, padding=3))
-        self.ups.apply(init_weights)
-        self.conv_post.apply(init_weights)
-        self.cond_up.apply(init_weights)
-        self.res_output.apply(init_weights)
-
-    def forward(self, x):
-        mel = x
-        x = self.conv_pre(x)
-        output = None
-        for i in range(self.num_upsamples):
-            if i >= self.cond_level:
-                mel = self.cond_up[i - self.cond_level](mel)
-                x += mel
-            if i > self.cond_level:
-                if output is None:
-                    output = self.res_output[i - self.cond_level - 1](x)
-                else:
-                    output = self.res_output[i - self.cond_level - 1](output)
-            x = F.leaky_relu(x, LRELU_SLOPE)
-            x = self.ups[i](x)
-            xs = None
-            for j in range(self.num_kernels):
-                if xs is None:
-                    xs = self.resblocks[i * self.num_kernels + j](x)
-                else:
-                    xs += self.resblocks[i * self.num_kernels + j](x)
-            x = xs / self.num_kernels
-            if output is not None:
-                output = output + x
-
-        x = F.leaky_relu(output)
-        x = self.conv_post(x)
-        x = torch.tanh(x)
-
-        return x
-
-    def remove_weight_norm(self):
-        print('Removing weight norm...')
-        for l in self.ups:
-            remove_weight_norm(l)
-        for l in self.resblocks:
-            l.remove_weight_norm()
-        for l in self.cond_up:
-            remove_weight_norm(l)
-        for l in self.res_output:
-            remove_weight_norm(l[1])
-        remove_weight_norm(self.conv_pre)
-        remove_weight_norm(self.conv_post)
-
-
-'''
-    to run this, fix 
-    from . import ResStack
-    into
-    from res_stack import ResStack
-'''
-if __name__ == '__main__':
-    '''
-    torch.Size([3, 80, 10])
-    torch.Size([3, 1, 2000])
-    4527362
-    '''
-    with open('config.json') as f:
-        data = f.read()
-    from utils import AttrDict
-    import json
-    json_config = json.loads(data)
-    h = AttrDict(json_config)
-    model = FreGAN(h)
-
-    c = torch.randn(3, 80, 10)  # (B, channels, T).
-    print(c.shape)
-
-    y = model(c) # (B, 1, T ** prod(upsample_scales)
-    print(y.shape)
-    assert y.shape == torch.Size([3, 1, 2560])  # For normal melgan torch.Size([3, 1, 2560])
-
-    pytorch_total_params = sum(p.numel() for p in model.parameters() if p.requires_grad)
-    print(pytorch_total_params)

models/vocoder/fregan/inference.py

@@ -1,74 +0,0 @@
-from __future__ import absolute_import, division, print_function, unicode_literals
-
-import os
-import json
-import torch
-from utils.util import AttrDict
-from models.vocoder.fregan.generator import FreGAN
-
-generator = None       # type: FreGAN
-output_sample_rate = None
-_device = None
-
-
-def load_checkpoint(filepath, device):
-    assert os.path.isfile(filepath)
-    print("Loading '{}'".format(filepath))
-    checkpoint_dict = torch.load(filepath, map_location=device)
-    print("Complete.")
-    return checkpoint_dict
-
-
-def load_model(weights_fpath, config_fpath=None, verbose=True):
-    global generator, _device, output_sample_rate
-
-    if verbose:
-        print("Building fregan")
-
-    if config_fpath == None:
-        model_config_fpaths = list(weights_fpath.parent.rglob("*.json"))
-        if len(model_config_fpaths) > 0:
-            config_fpath = model_config_fpaths[0]
-        else:
-            config_fpath = "./vocoder/fregan/config.json"
-    with open(config_fpath) as f:
-        data = f.read()
-    json_config = json.loads(data)
-    h = AttrDict(json_config)
-    output_sample_rate = h.sampling_rate
-    torch.manual_seed(h.seed)
-
-    if torch.cuda.is_available():
-        # _model = _model.cuda()
-        _device = torch.device('cuda')
-    else:
-        _device = torch.device('cpu')
-
-    generator = FreGAN(h).to(_device)
-    state_dict_g = load_checkpoint(
-        weights_fpath, _device
-    )
-    generator.load_state_dict(state_dict_g['generator'])
-    generator.eval()
-    generator.remove_weight_norm()
-
-
-def is_loaded():
-    return generator is not None
-
-
-def infer_waveform(mel, progress_callback=None):
-
-    if generator is None:
-        raise Exception("Please load fre-gan in memory before using it")
-
-    mel = torch.FloatTensor(mel).to(_device)
-    mel = mel.unsqueeze(0)
-
-    with torch.no_grad():
-        y_g_hat = generator(mel)
-        audio = y_g_hat.squeeze()
-    audio = audio.cpu().numpy()
-
-    return audio, output_sample_rate
-

models/vocoder/fregan/meldataset.py

@@ -1,120 +0,0 @@
-import math
-import os
-import random
-import torch
-import torch.utils.data
-import numpy as np
-from librosa.util import normalize
-from scipy.io.wavfile import read
-from utils.audio_utils import mel_spectrogram
-
-MAX_WAV_VALUE = 32768.0
-
-
-def load_wav(full_path):
-    sampling_rate, data = read(full_path)
-    return data, sampling_rate
-
-
-def get_dataset_filelist(a):
-    #with open(a.input_training_file, 'r', encoding='utf-8') as fi:
-    #    training_files = [os.path.join(a.input_wavs_dir, x.split('|')[0] + '.wav')
-    #                      for x in fi.read().split('\n') if len(x) > 0]
-
-    #with open(a.input_validation_file, 'r', encoding='utf-8') as fi:
-    #   validation_files = [os.path.join(a.input_wavs_dir, x.split('|')[0] + '.wav')
-    #                        for x in fi.read().split('\n') if len(x) > 0]
-    files = os.listdir(a.input_wavs_dir)
-    random.shuffle(files)
-    files = [os.path.join(a.input_wavs_dir, f) for f in files]
-    training_files = files[: -int(len(files) * 0.05)]
-    validation_files = files[-int(len(files) * 0.05):]
-    return training_files, validation_files
-
-
-class MelDataset(torch.utils.data.Dataset):
-    def __init__(self, training_files, segment_size, n_fft, num_mels,
-                 hop_size, win_size, sampling_rate,  fmin, fmax, split=True, shuffle=True, n_cache_reuse=1,
-                 device=None, fmax_loss=None, fine_tuning=False, base_mels_path=None):
-        self.audio_files = training_files
-        random.seed(1234)
-        if shuffle:
-            random.shuffle(self.audio_files)
-        self.segment_size = segment_size
-        self.sampling_rate = sampling_rate
-        self.split = split
-        self.n_fft = n_fft
-        self.num_mels = num_mels
-        self.hop_size = hop_size
-        self.win_size = win_size
-        self.fmin = fmin
-        self.fmax = fmax
-        self.fmax_loss = fmax_loss
-        self.cached_wav = None
-        self.n_cache_reuse = n_cache_reuse
-        self._cache_ref_count = 0
-        self.device = device
-        self.fine_tuning = fine_tuning
-        self.base_mels_path = base_mels_path
-
-    def __getitem__(self, index):
-        filename = self.audio_files[index]
-        if self._cache_ref_count == 0:
-            #audio, sampling_rate = load_wav(filename)
-            #audio = audio / MAX_WAV_VALUE
-            audio = np.load(filename)
-            if not self.fine_tuning:
-                audio = normalize(audio) * 0.95
-            self.cached_wav = audio
-            #if sampling_rate != self.sampling_rate:
-            #    raise ValueError("{} SR doesn't match target {} SR".format(
-            #        sampling_rate, self.sampling_rate))
-            self._cache_ref_count = self.n_cache_reuse
-        else:
-            audio = self.cached_wav
-            self._cache_ref_count -= 1
-
-        audio = torch.FloatTensor(audio)
-        audio = audio.unsqueeze(0)
-
-        if not self.fine_tuning:
-            if self.split:
-                if audio.size(1) >= self.segment_size:
-                    max_audio_start = audio.size(1) - self.segment_size
-                    audio_start = random.randint(0, max_audio_start)
-                    audio = audio[:, audio_start:audio_start+self.segment_size]
-                else:
-                    audio = torch.nn.functional.pad(audio, (0, self.segment_size - audio.size(1)), 'constant')
-
-            mel = mel_spectrogram(audio, self.n_fft, self.num_mels,
-                                  self.sampling_rate, self.hop_size, self.win_size, self.fmin, self.fmax,
-                                  center=False)
-        else:
-            mel_path = os.path.join(self.base_mels_path, "mel" + "-" + filename.split("/")[-1].split("-")[-1])
-            mel = np.load(mel_path).T
-            #mel = np.load(
-            #    os.path.join(self.base_mels_path, os.path.splitext(os.path.split(filename)[-1])[0] + '.npy'))
-            mel = torch.from_numpy(mel)
-
-            if len(mel.shape) < 3:
-                mel = mel.unsqueeze(0)
-
-            if self.split:
-                frames_per_seg = math.ceil(self.segment_size / self.hop_size)
-
-                if audio.size(1) >= self.segment_size:
-                    mel_start = random.randint(0, mel.size(2) - frames_per_seg - 1)
-                    mel = mel[:, :, mel_start:mel_start + frames_per_seg]
-                    audio = audio[:, mel_start * self.hop_size:(mel_start + frames_per_seg) * self.hop_size]
-                else:
-                    mel = torch.nn.functional.pad(mel, (0, frames_per_seg - mel.size(2)), 'constant')
-                    audio = torch.nn.functional.pad(audio, (0, self.segment_size - audio.size(1)), 'constant')
-
-        mel_loss = mel_spectrogram(audio, self.n_fft, self.num_mels,
-                                   self.sampling_rate, self.hop_size, self.win_size, self.fmin, self.fmax_loss,
-                                   center=False)
-
-        return (mel.squeeze(), audio.squeeze(0), filename, mel_loss.squeeze())
-
-    def __len__(self):
-        return len(self.audio_files)

models/vocoder/fregan/modules.py

@@ -1,201 +0,0 @@
-import torch
-import torch.nn.functional as F
-
-class KernelPredictor(torch.nn.Module):
-    ''' Kernel predictor for the location-variable convolutions
-    '''
-
-    def __init__(self,
-                 cond_channels,
-                 conv_in_channels,
-                 conv_out_channels,
-                 conv_layers,
-                 conv_kernel_size=3,
-                 kpnet_hidden_channels=64,
-                 kpnet_conv_size=3,
-                 kpnet_dropout=0.0,
-                 kpnet_nonlinear_activation="LeakyReLU",
-                 kpnet_nonlinear_activation_params={"negative_slope": 0.1}
-                 ):
-        '''
-        Args:
-            cond_channels (int): number of channel for the conditioning sequence,
-            conv_in_channels (int): number of channel for the input sequence,
-            conv_out_channels (int): number of channel for the output sequence,
-            conv_layers (int):
-            kpnet_
-        '''
-        super().__init__()
-
-        self.conv_in_channels = conv_in_channels
-        self.conv_out_channels = conv_out_channels
-        self.conv_kernel_size = conv_kernel_size
-        self.conv_layers = conv_layers
-
-        l_w = conv_in_channels * conv_out_channels * conv_kernel_size * conv_layers
-        l_b = conv_out_channels * conv_layers
-
-        padding = (kpnet_conv_size - 1) // 2
-        self.input_conv = torch.nn.Sequential(
-            torch.nn.Conv1d(cond_channels, kpnet_hidden_channels, 5, padding=(5 - 1) // 2, bias=True),
-            getattr(torch.nn, kpnet_nonlinear_activation)(**kpnet_nonlinear_activation_params),
-        )
-
-        self.residual_conv = torch.nn.Sequential(
-            torch.nn.Dropout(kpnet_dropout),
-            torch.nn.Conv1d(kpnet_hidden_channels, kpnet_hidden_channels, kpnet_conv_size, padding=padding, bias=True),
-            getattr(torch.nn, kpnet_nonlinear_activation)(**kpnet_nonlinear_activation_params),
-            torch.nn.Conv1d(kpnet_hidden_channels, kpnet_hidden_channels, kpnet_conv_size, padding=padding, bias=True),
-            getattr(torch.nn, kpnet_nonlinear_activation)(**kpnet_nonlinear_activation_params),
-            torch.nn.Dropout(kpnet_dropout),
-            torch.nn.Conv1d(kpnet_hidden_channels, kpnet_hidden_channels, kpnet_conv_size, padding=padding, bias=True),
-            getattr(torch.nn, kpnet_nonlinear_activation)(**kpnet_nonlinear_activation_params),
-            torch.nn.Conv1d(kpnet_hidden_channels, kpnet_hidden_channels, kpnet_conv_size, padding=padding, bias=True),
-            getattr(torch.nn, kpnet_nonlinear_activation)(**kpnet_nonlinear_activation_params),
-            torch.nn.Dropout(kpnet_dropout),
-            torch.nn.Conv1d(kpnet_hidden_channels, kpnet_hidden_channels, kpnet_conv_size, padding=padding, bias=True),
-            getattr(torch.nn, kpnet_nonlinear_activation)(**kpnet_nonlinear_activation_params),
-            torch.nn.Conv1d(kpnet_hidden_channels, kpnet_hidden_channels, kpnet_conv_size, padding=padding, bias=True),
-            getattr(torch.nn, kpnet_nonlinear_activation)(**kpnet_nonlinear_activation_params),
-        )
-
-        self.kernel_conv = torch.nn.Conv1d(kpnet_hidden_channels, l_w, kpnet_conv_size,
-                                           padding=padding, bias=True)
-        self.bias_conv = torch.nn.Conv1d(kpnet_hidden_channels, l_b, kpnet_conv_size, padding=padding,
-                                         bias=True)
-
-    def forward(self, c):
-        '''
-        Args:
-            c (Tensor): the conditioning sequence (batch, cond_channels, cond_length)
-        Returns:
-        '''
-        batch, cond_channels, cond_length = c.shape
-
-        c = self.input_conv(c)
-        c = c + self.residual_conv(c)
-        k = self.kernel_conv(c)
-        b = self.bias_conv(c)
-
-        kernels = k.contiguous().view(batch,
-                                      self.conv_layers,
-                                      self.conv_in_channels,
-                                      self.conv_out_channels,
-                                      self.conv_kernel_size,
-                                      cond_length)
-        bias = b.contiguous().view(batch,
-                                   self.conv_layers,
-                                   self.conv_out_channels,
-                                   cond_length)
-        return kernels, bias
-
-
-class LVCBlock(torch.nn.Module):
-    ''' the location-variable convolutions
-    '''
-
-    def __init__(self,
-                 in_channels,
-                 cond_channels,
-                 upsample_ratio,
-                 conv_layers=4,
-                 conv_kernel_size=3,
-                 cond_hop_length=256,
-                 kpnet_hidden_channels=64,
-                 kpnet_conv_size=3,
-                 kpnet_dropout=0.0
-                 ):
-        super().__init__()
-
-        self.cond_hop_length = cond_hop_length
-        self.conv_layers = conv_layers
-        self.conv_kernel_size = conv_kernel_size
-        self.convs = torch.nn.ModuleList()
-
-        self.upsample = torch.nn.ConvTranspose1d(in_channels, in_channels,
-                                    kernel_size=upsample_ratio*2, stride=upsample_ratio,
-                                    padding=upsample_ratio // 2 + upsample_ratio % 2,
-                                    output_padding=upsample_ratio % 2)
-
-
-        self.kernel_predictor = KernelPredictor(
-            cond_channels=cond_channels,
-            conv_in_channels=in_channels,
-            conv_out_channels=2 * in_channels,
-            conv_layers=conv_layers,
-            conv_kernel_size=conv_kernel_size,
-            kpnet_hidden_channels=kpnet_hidden_channels,
-            kpnet_conv_size=kpnet_conv_size,
-            kpnet_dropout=kpnet_dropout
-        )
-
-
-        for i in range(conv_layers):
-            padding = (3 ** i) * int((conv_kernel_size - 1) / 2)
-            conv = torch.nn.Conv1d(in_channels, in_channels, kernel_size=conv_kernel_size, padding=padding, dilation=3 ** i)
-
-            self.convs.append(conv)
-
-
-    def forward(self, x, c):
-        ''' forward propagation of the location-variable convolutions.
-        Args:
-            x (Tensor): the input sequence (batch, in_channels, in_length)
-            c (Tensor): the conditioning sequence (batch, cond_channels, cond_length)
-
-        Returns:
-            Tensor: the output sequence (batch, in_channels, in_length)
-        '''
-        batch, in_channels, in_length = x.shape
-
-
-        kernels, bias = self.kernel_predictor(c)
-
-        x = F.leaky_relu(x, 0.2)
-        x = self.upsample(x)
-
-        for i in range(self.conv_layers):
-            y = F.leaky_relu(x, 0.2)
-            y = self.convs[i](y)
-            y = F.leaky_relu(y, 0.2)
-
-            k = kernels[:, i, :, :, :, :]
-            b = bias[:, i, :, :]
-            y = self.location_variable_convolution(y, k, b, 1, self.cond_hop_length)
-            x = x + torch.sigmoid(y[:, :in_channels, :]) * torch.tanh(y[:, in_channels:, :])
-        return x
-
-    def location_variable_convolution(self, x, kernel, bias, dilation, hop_size):
-        ''' perform location-variable convolution operation on the input sequence (x) using the local convolution kernl.
-        Time: 414 μs ± 309 ns per loop (mean ± std. dev. of 7 runs, 1000 loops each), test on NVIDIA V100.
-        Args:
-            x (Tensor): the input sequence (batch, in_channels, in_length).
-            kernel (Tensor): the local convolution kernel (batch, in_channel, out_channels, kernel_size, kernel_length)
-            bias (Tensor): the bias for the local convolution (batch, out_channels, kernel_length)
-            dilation (int): the dilation of convolution.
-            hop_size (int): the hop_size of the conditioning sequence.
-        Returns:
-            (Tensor): the output sequence after performing local convolution. (batch, out_channels, in_length).
-        '''
-        batch, in_channels, in_length = x.shape
-        batch, in_channels, out_channels, kernel_size, kernel_length = kernel.shape
-
-
-        assert in_length == (kernel_length * hop_size), "length of (x, kernel) is not matched"
-
-        padding = dilation * int((kernel_size - 1) / 2)
-        x = F.pad(x, (padding, padding), 'constant', 0)  # (batch, in_channels, in_length + 2*padding)
-        x = x.unfold(2, hop_size + 2 * padding, hop_size)  # (batch, in_channels, kernel_length, hop_size + 2*padding)
-
-        if hop_size < dilation:
-            x = F.pad(x, (0, dilation), 'constant', 0)
-        x = x.unfold(3, dilation,
-                     dilation)  # (batch, in_channels, kernel_length, (hop_size + 2*padding)/dilation, dilation)
-        x = x[:, :, :, :, :hop_size]
-        x = x.transpose(3, 4)  # (batch, in_channels, kernel_length, dilation, (hop_size + 2*padding)/dilation)
-        x = x.unfold(4, kernel_size, 1)  # (batch, in_channels, kernel_length, dilation, _, kernel_size)
-
-        o = torch.einsum('bildsk,biokl->bolsd', x, kernel)
-        o = o + bias.unsqueeze(-1).unsqueeze(-1)
-        o = o.contiguous().view(batch, out_channels, -1)
-        return o

models/vocoder/fregan/stft_loss.py

@@ -1,136 +0,0 @@
-# -*- coding: utf-8 -*-
-
-# Copyright 2019 Tomoki Hayashi
-#  MIT License (https://opensource.org/licenses/MIT)
-
-"""STFT-based Loss modules."""
-
-import torch
-import torch.nn.functional as F
-
-
-def stft(x, fft_size, hop_size, win_length, window):
-    """Perform STFT and convert to magnitude spectrogram.
-    Args:
-        x (Tensor): Input signal tensor (B, T).
-        fft_size (int): FFT size.
-        hop_size (int): Hop size.
-        win_length (int): Window length.
-        window (str): Window function type.
-    Returns:
-        Tensor: Magnitude spectrogram (B, #frames, fft_size // 2 + 1).
-    """
-    x_stft = torch.stft(x, fft_size, hop_size, win_length, window)
-    real = x_stft[..., 0]
-    imag = x_stft[..., 1]
-
-    # NOTE(kan-bayashi): clamp is needed to avoid nan or inf
-    return torch.sqrt(torch.clamp(real ** 2 + imag ** 2, min=1e-7)).transpose(2, 1)
-
-
-class SpectralConvergengeLoss(torch.nn.Module):
-    """Spectral convergence loss module."""
-
-    def __init__(self):
-        """Initilize spectral convergence loss module."""
-        super(SpectralConvergengeLoss, self).__init__()
-
-    def forward(self, x_mag, y_mag):
-        """Calculate forward propagation.
-        Args:
-            x_mag (Tensor): Magnitude spectrogram of predicted signal (B, #frames, #freq_bins).
-            y_mag (Tensor): Magnitude spectrogram of groundtruth signal (B, #frames, #freq_bins).
-        Returns:
-            Tensor: Spectral convergence loss value.
-        """
-        return torch.norm(y_mag - x_mag, p="fro") / torch.norm(y_mag, p="fro")
-
-
-class LogSTFTMagnitudeLoss(torch.nn.Module):
-    """Log STFT magnitude loss module."""
-
-    def __init__(self):
-        """Initilize los STFT magnitude loss module."""
-        super(LogSTFTMagnitudeLoss, self).__init__()
-
-    def forward(self, x_mag, y_mag):
-        """Calculate forward propagation.
-        Args:
-            x_mag (Tensor): Magnitude spectrogram of predicted signal (B, #frames, #freq_bins).
-            y_mag (Tensor): Magnitude spectrogram of groundtruth signal (B, #frames, #freq_bins).
-        Returns:
-            Tensor: Log STFT magnitude loss value.
-        """
-        return F.l1_loss(torch.log(y_mag), torch.log(x_mag))
-
-
-class STFTLoss(torch.nn.Module):
-    """STFT loss module."""
-
-    def __init__(self, fft_size=1024, shift_size=120, win_length=600, window="hann_window"):
-        """Initialize STFT loss module."""
-        super(STFTLoss, self).__init__()
-        self.fft_size = fft_size
-        self.shift_size = shift_size
-        self.win_length = win_length
-        self.window = getattr(torch, window)(win_length)
-        self.spectral_convergenge_loss = SpectralConvergengeLoss()
-        self.log_stft_magnitude_loss = LogSTFTMagnitudeLoss()
-
-    def forward(self, x, y):
-        """Calculate forward propagation.
-        Args:
-            x (Tensor): Predicted signal (B, T).
-            y (Tensor): Groundtruth signal (B, T).
-        Returns:
-            Tensor: Spectral convergence loss value.
-            Tensor: Log STFT magnitude loss value.
-        """
-        x_mag = stft(x, self.fft_size, self.shift_size, self.win_length, self.window.to(x.get_device()))
-        y_mag = stft(y, self.fft_size, self.shift_size, self.win_length, self.window.to(x.get_device()))
-        sc_loss = self.spectral_convergenge_loss(x_mag, y_mag)
-        mag_loss = self.log_stft_magnitude_loss(x_mag, y_mag)
-
-        return sc_loss, mag_loss
-
-
-class MultiResolutionSTFTLoss(torch.nn.Module):
-    """Multi resolution STFT loss module."""
-
-    def __init__(self,
-                 fft_sizes=[1024, 2048, 512],
-                 hop_sizes=[120, 240, 50],
-                 win_lengths=[600, 1200, 240],
-                 window="hann_window"):
-        """Initialize Multi resolution STFT loss module.
-        Args:
-            fft_sizes (list): List of FFT sizes.
-            hop_sizes (list): List of hop sizes.
-            win_lengths (list): List of window lengths.
-            window (str): Window function type.
-        """
-        super(MultiResolutionSTFTLoss, self).__init__()
-        assert len(fft_sizes) == len(hop_sizes) == len(win_lengths)
-        self.stft_losses = torch.nn.ModuleList()
-        for fs, ss, wl in zip(fft_sizes, hop_sizes, win_lengths):
-            self.stft_losses += [STFTLoss(fs, ss, wl, window)]
-
-    def forward(self, x, y):
-        """Calculate forward propagation.
-        Args:
-            x (Tensor): Predicted signal (B, T).
-            y (Tensor): Groundtruth signal (B, T).
-        Returns:
-            Tensor: Multi resolution spectral convergence loss value.
-            Tensor: Multi resolution log STFT magnitude loss value.
-        """
-        sc_loss = 0.0
-        mag_loss = 0.0
-        for f in self.stft_losses:
-            sc_l, mag_l = f(x, y)
-            sc_loss += sc_l
-            mag_loss += mag_l
-        sc_loss /= len(self.stft_losses)
-        mag_loss /= len(self.stft_losses)
-
-        return sc_loss, mag_loss

models/vocoder/fregan/train.py

@@ -1,246 +0,0 @@
-import warnings
-
-warnings.simplefilter(action='ignore', category=FutureWarning)
-import itertools
-import os
-import time
-import torch
-import torch.nn.functional as F
-from torch.utils.tensorboard import SummaryWriter
-from torch.utils.data import DistributedSampler, DataLoader
-from torch.distributed import init_process_group
-from torch.nn.parallel import DistributedDataParallel
-from models.vocoder.fregan.meldataset import MelDataset, mel_spectrogram, get_dataset_filelist
-from models.vocoder.fregan.generator import FreGAN
-from models.vocoder.fregan.discriminator import ResWiseMultiPeriodDiscriminator, ResWiseMultiScaleDiscriminator
-from utils.loss import feature_loss, generator_loss, discriminator_loss
-from models.vocoder.fregan.utils import plot_spectrogram, scan_checkpoint, load_checkpoint, save_checkpoint
-
-
-torch.backends.cudnn.benchmark = True
-
-
-def train(rank, a, h):
-
-    a.checkpoint_path = a.models_dir.joinpath(a.run_id+'_fregan')
-    a.checkpoint_path.mkdir(exist_ok=True)
-    a.training_epochs = 3100
-    a.stdout_interval = 5
-    a.checkpoint_interval = a.backup_every
-    a.summary_interval = 5000
-    a.validation_interval = 1000
-    a.fine_tuning = True
-
-    a.input_wavs_dir = a.syn_dir.joinpath("audio")
-    a.input_mels_dir = a.syn_dir.joinpath("mels")
-
-    if h.num_gpus > 1:
-        init_process_group(backend=h.dist_config['dist_backend'], init_method=h.dist_config['dist_url'],
-                           world_size=h.dist_config['world_size'] * h.num_gpus, rank=rank)
-
-    torch.cuda.manual_seed(h.seed)
-    device = torch.device('cuda:{:d}'.format(rank))
-
-    generator = FreGAN(h).to(device)
-    mpd = ResWiseMultiPeriodDiscriminator().to(device)
-    msd = ResWiseMultiScaleDiscriminator().to(device)
-
-    if rank == 0:
-        print(generator)
-        os.makedirs(a.checkpoint_path, exist_ok=True)
-        print("checkpoints directory : ", a.checkpoint_path)
-
-    if os.path.isdir(a.checkpoint_path):
-        cp_g = scan_checkpoint(a.checkpoint_path, 'g_fregan_')
-        cp_do = scan_checkpoint(a.checkpoint_path, 'do_fregan_')
-
-    steps = 0
-    if cp_g is None or cp_do is None:
-        state_dict_do = None
-        last_epoch = -1
-    else:
-        state_dict_g = load_checkpoint(cp_g, device)
-        state_dict_do = load_checkpoint(cp_do, device)
-        generator.load_state_dict(state_dict_g['generator'])
-        mpd.load_state_dict(state_dict_do['mpd'])
-        msd.load_state_dict(state_dict_do['msd'])
-        steps = state_dict_do['steps'] + 1
-        last_epoch = state_dict_do['epoch']
-
-    if h.num_gpus > 1:
-        generator = DistributedDataParallel(generator, device_ids=[rank]).to(device)
-        mpd = DistributedDataParallel(mpd, device_ids=[rank]).to(device)
-        msd = DistributedDataParallel(msd, device_ids=[rank]).to(device)
-
-    optim_g = torch.optim.AdamW(generator.parameters(), h.learning_rate, betas=[h.adam_b1, h.adam_b2])
-    optim_d = torch.optim.AdamW(itertools.chain(msd.parameters(), mpd.parameters()),
-                                h.learning_rate, betas=[h.adam_b1, h.adam_b2])
-
-    if state_dict_do is not None:
-        optim_g.load_state_dict(state_dict_do['optim_g'])
-        optim_d.load_state_dict(state_dict_do['optim_d'])
-
-    scheduler_g = torch.optim.lr_scheduler.ExponentialLR(optim_g, gamma=h.lr_decay, last_epoch=last_epoch)
-    scheduler_d = torch.optim.lr_scheduler.ExponentialLR(optim_d, gamma=h.lr_decay, last_epoch=last_epoch)
-
-    training_filelist, validation_filelist = get_dataset_filelist(a)
-
-    trainset = MelDataset(training_filelist, h.segment_size, h.n_fft, h.num_mels,
-                          h.hop_size, h.win_size, h.sampling_rate, h.fmin, h.fmax, n_cache_reuse=0,
-                          shuffle=False if h.num_gpus > 1 else True, fmax_loss=h.fmax_for_loss, device=device,
-                          fine_tuning=a.fine_tuning, base_mels_path=a.input_mels_dir)
-
-    train_sampler = DistributedSampler(trainset) if h.num_gpus > 1 else None
-
-    train_loader = DataLoader(trainset, num_workers=h.num_workers, shuffle=False,
-                              sampler=train_sampler,
-                              batch_size=h.batch_size,
-                              pin_memory=True,
-                              drop_last=True)
-
-    if rank == 0:
-        validset = MelDataset(validation_filelist, h.segment_size, h.n_fft, h.num_mels,
-                              h.hop_size, h.win_size, h.sampling_rate, h.fmin, h.fmax, False, False, n_cache_reuse=0,
-                              fmax_loss=h.fmax_for_loss, device=device, fine_tuning=a.fine_tuning,
-                              base_mels_path=a.input_mels_dir)
-        validation_loader = DataLoader(validset, num_workers=1, shuffle=False,
-                                       sampler=None,
-                                       batch_size=1,
-                                       pin_memory=True,
-                                       drop_last=True)
-
-        sw = SummaryWriter(os.path.join(a.checkpoint_path, 'logs'))
-
-    generator.train()
-    mpd.train()
-    msd.train()
-    for epoch in range(max(0, last_epoch), a.training_epochs):
-        if rank == 0:
-            start = time.time()
-            print("Epoch: {}".format(epoch + 1))
-
-        if h.num_gpus > 1:
-            train_sampler.set_epoch(epoch)
-
-        for i, batch in enumerate(train_loader):
-            if rank == 0:
-                start_b = time.time()
-            x, y, _, y_mel = batch
-            x = torch.autograd.Variable(x.to(device, non_blocking=True))
-            y = torch.autograd.Variable(y.to(device, non_blocking=True))
-            y_mel = torch.autograd.Variable(y_mel.to(device, non_blocking=True))
-            y = y.unsqueeze(1)
-            y_g_hat = generator(x)
-            y_g_hat_mel = mel_spectrogram(y_g_hat.squeeze(1), h.n_fft, h.num_mels, h.sampling_rate, h.hop_size,
-                                          h.win_size,
-                                          h.fmin, h.fmax_for_loss)
-
-            if steps > h.disc_start_step:
-                optim_d.zero_grad()
-
-                # MPD
-                y_df_hat_r, y_df_hat_g, _, _ = mpd(y, y_g_hat.detach())
-                loss_disc_f, losses_disc_f_r, losses_disc_f_g = discriminator_loss(y_df_hat_r, y_df_hat_g)
-
-                # MSD
-                y_ds_hat_r, y_ds_hat_g, _, _ = msd(y, y_g_hat.detach())
-                loss_disc_s, losses_disc_s_r, losses_disc_s_g = discriminator_loss(y_ds_hat_r, y_ds_hat_g)
-
-                loss_disc_all = loss_disc_s + loss_disc_f
-
-                loss_disc_all.backward()
-                optim_d.step()
-
-            # Generator
-            optim_g.zero_grad()
-
-
-            # L1 Mel-Spectrogram Loss
-            loss_mel = F.l1_loss(y_mel, y_g_hat_mel) * 45
-
-            # sc_loss, mag_loss = stft_loss(y_g_hat[:, :, :y.size(2)].squeeze(1), y.squeeze(1))
-            # loss_mel = h.lambda_aux * (sc_loss + mag_loss)  # STFT Loss
-
-            if steps > h.disc_start_step:
-                y_df_hat_r, y_df_hat_g, fmap_f_r, fmap_f_g = mpd(y, y_g_hat)
-                y_ds_hat_r, y_ds_hat_g, fmap_s_r, fmap_s_g = msd(y, y_g_hat)
-                loss_fm_f = feature_loss(fmap_f_r, fmap_f_g)
-                loss_fm_s = feature_loss(fmap_s_r, fmap_s_g)
-                loss_gen_f, losses_gen_f = generator_loss(y_df_hat_g)
-                loss_gen_s, losses_gen_s = generator_loss(y_ds_hat_g)
-                loss_gen_all = loss_gen_s + loss_gen_f + (2 * (loss_fm_s + loss_fm_f)) + loss_mel
-            else:
-                loss_gen_all = loss_mel
-
-            loss_gen_all.backward()
-            optim_g.step()
-
-            if rank == 0:
-                # STDOUT logging
-                if steps % a.stdout_interval == 0:
-                    with torch.no_grad():
-                        mel_error = F.l1_loss(y_mel, y_g_hat_mel).item()
-
-                    print('Steps : {:d}, Gen Loss Total : {:4.3f}, Mel-Spec. Error : {:4.3f}, s/b : {:4.3f}'.
-                          format(steps, loss_gen_all, mel_error, time.time() - start_b))
-
-                # checkpointing
-                if steps % a.checkpoint_interval == 0 and steps != 0:
-                    checkpoint_path = "{}/g_fregan_{:08d}.pt".format(a.checkpoint_path, steps)
-                    save_checkpoint(checkpoint_path,
-                                    {'generator': (generator.module if h.num_gpus > 1 else generator).state_dict()})
-                    checkpoint_path = "{}/do_fregan_{:08d}.pt".format(a.checkpoint_path, steps)
-                    save_checkpoint(checkpoint_path,
-                                    {'mpd': (mpd.module if h.num_gpus > 1
-                                             else mpd).state_dict(),
-                                     'msd': (msd.module if h.num_gpus > 1
-                                             else msd).state_dict(),
-                                     'optim_g': optim_g.state_dict(), 'optim_d': optim_d.state_dict(), 'steps': steps,
-                                     'epoch': epoch})
-
-                # Tensorboard summary logging
-                if steps % a.summary_interval == 0:
-                    sw.add_scalar("training/gen_loss_total", loss_gen_all, steps)
-                    sw.add_scalar("training/mel_spec_error", mel_error, steps)
-
-                # Validation
-                if steps % a.validation_interval == 0:  # and steps != 0:
-                    generator.eval()
-                    torch.cuda.empty_cache()
-                    val_err_tot = 0
-                    with torch.no_grad():
-                        for j, batch in enumerate(validation_loader):
-                            x, y, _, y_mel = batch
-                            y_g_hat = generator(x.to(device))
-                            y_mel = torch.autograd.Variable(y_mel.to(device, non_blocking=True))
-                            y_g_hat_mel = mel_spectrogram(y_g_hat.squeeze(1), h.n_fft, h.num_mels, h.sampling_rate,
-                                                          h.hop_size, h.win_size,
-                                                          h.fmin, h.fmax_for_loss)
-                            #val_err_tot += F.l1_loss(y_mel, y_g_hat_mel).item()
-
-                            if j <= 4:
-                                if steps == 0:
-                                    sw.add_audio('gt/y_{}'.format(j), y[0], steps, h.sampling_rate)
-                                    sw.add_figure('gt/y_spec_{}'.format(j), plot_spectrogram(x[0]), steps)
-
-                                sw.add_audio('generated/y_hat_{}'.format(j), y_g_hat[0], steps, h.sampling_rate)
-                                y_hat_spec = mel_spectrogram(y_g_hat.squeeze(1), h.n_fft, h.num_mels,
-                                                             h.sampling_rate, h.hop_size, h.win_size,
-                                                             h.fmin, h.fmax)
-                                sw.add_figure('generated/y_hat_spec_{}'.format(j),
-                                              plot_spectrogram(y_hat_spec.squeeze(0).cpu().numpy()), steps)
-
-                        val_err = val_err_tot / (j + 1)
-                        sw.add_scalar("validation/mel_spec_error", val_err, steps)
-
-                    generator.train()
-
-            steps += 1
-
-        scheduler_g.step()
-        scheduler_d.step()
-
-        if rank == 0:
-            print('Time taken for epoch {} is {} sec\n'.format(epoch + 1, int(time.time() - start)))
-
-

models/vocoder/fregan/utils.py

@@ -1,56 +0,0 @@
-import glob
-import os
-import matplotlib
-import torch
-from torch.nn.utils import weight_norm
-matplotlib.use("Agg")
-import matplotlib.pylab as plt
-import shutil
-
-
-def build_env(config, config_name, path):
-    t_path = os.path.join(path, config_name)
-    if config != t_path:
-        os.makedirs(path, exist_ok=True)
-        shutil.copyfile(config, os.path.join(path, config_name))
-
-
-def plot_spectrogram(spectrogram):
-    fig, ax = plt.subplots(figsize=(10, 2))
-    im = ax.imshow(spectrogram, aspect="auto", origin="lower",
-                   interpolation='none')
-    plt.colorbar(im, ax=ax)
-
-    fig.canvas.draw()
-    plt.close()
-
-    return fig
-
-
-def apply_weight_norm(m):
-    classname = m.__class__.__name__
-    if classname.find("Conv") != -1:
-        weight_norm(m)
-
-
-
-def load_checkpoint(filepath, device):
-    assert os.path.isfile(filepath)
-    print("Loading '{}'".format(filepath))
-    checkpoint_dict = torch.load(filepath, map_location=device)
-    print("Complete.")
-    return checkpoint_dict
-
-
-def save_checkpoint(filepath, obj):
-    print("Saving checkpoint to {}".format(filepath))
-    torch.save(obj, filepath)
-    print("Complete.")
-
-
-def scan_checkpoint(cp_dir, prefix):
-    pattern = os.path.join(cp_dir, prefix + '????????.pt')
-    cp_list = glob.glob(pattern)
-    if len(cp_list) == 0:
-        return None
-    return sorted(cp_list)[-1]

models/vocoder/hifigan/__init__.py

@@ -1 +0,0 @@
-#

ppg2mel/__init__.py

@@ -15,7 +15,7 @@ from .rnn_decoder_mol import Decoder
 from .utils.cnn_postnet import Postnet
 from .utils.vc_utils import get_mask_from_lengths
 
-from utils.hparams import HpsYaml
+from utils.load_yaml import HpsYaml
 
 class MelDecoderMOLv2(AbsMelDecoder):
     """Use an encoder to preprocess ppg."""

ppg2mel/preprocess.py

@@ -7,10 +7,10 @@ from pathlib import Path
 import soundfile
 import resampy
 
-from models.ppg_extractor import load_model
+from ppg_extractor import load_model
 import encoder.inference as Encoder
-from models.encoder.audio import preprocess_wav
+from encoder.audio import preprocess_wav
-from models.encoder import audio
+from encoder import audio
 from utils.f0_utils import compute_f0
 
 from torch.multiprocessing import Pool, cpu_count

ppg2mel/train.py

@@ -2,8 +2,8 @@ import sys
 import torch
 import argparse
 import numpy as np
-from utils.hparams import HpsYaml
+from utils.load_yaml import HpsYaml
-from models.ppg2mel.train.train_linglf02mel_seq2seq_oneshotvc import Solver
+from ppg2mel.train.train_linglf02mel_seq2seq_oneshotvc import Solver
 
 # For reproducibility, comment these may speed up training
 torch.backends.cudnn.deterministic = True

ppg2mel/train/solver.py

@@ -8,6 +8,7 @@ from torch.utils.tensorboard import SummaryWriter
 
 from .option import default_hparas
 from utils.util import human_format, Timer
+from utils.load_yaml import HpsYaml
 
 
 class BaseSolver():

ppg2mel/train/train_linglf02mel_seq2seq_oneshotvc.py

@@ -14,7 +14,7 @@ from utils.data_load import OneshotVcDataset, MultiSpkVcCollate
 from .loss import MaskedMSELoss
 from .optim import Optimizer
 from utils.util import human_format
-from models.ppg2mel import MelDecoderMOLv2
+from ppg2mel import MelDecoderMOLv2
 
 
 class Solver(BaseSolver):