Merge branch 'main' into restruct-project

* add docker support

* 修复训练集解压问题

* 修复web.py启动问题

* 混合数据集训练参数

.dockerignore

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

.gitignore

@@ -13,10 +13,12 @@
 *.bbl
 *.bcf
 *.toc
-*.wav
 *.sh
-*/saved_models
-!vocoder/saved_models/pretrained/**
-!encoder/saved_models/pretrained.pt
+data/ckpt
+!data/ckpt/vocoder/pretrained/**
+!data/ckpt/encoder/pretrained.pt
 wavs
-log
+log
+!/docker-entrypoint.sh
+!/datasets_download/*.sh
+/datasets

.vscode/launch.json

@@ -15,7 +15,8 @@
       "name": "Python: Vocoder Preprocess",
       "type": "python",
       "request": "launch",
-      "program": "vocoder_preprocess.py",
+      "program": "control\\cli\\vocoder_preprocess.py",
+      "cwd": "${workspaceFolder}",
       "console": "integratedTerminal",
       "args": ["..\\audiodata"]
     },
@@ -23,7 +24,8 @@
       "name": "Python: Vocoder Train",
       "type": "python",
       "request": "launch",
-      "program": "vocoder_train.py",
+      "program": "control\\cli\\vocoder_train.py",
+      "cwd": "${workspaceFolder}",
       "console": "integratedTerminal",
       "args": ["dev", "..\\audiodata"]
     },
@@ -32,6 +34,7 @@
       "type": "python",
       "request": "launch",
       "program": "demo_toolbox.py",
+      "cwd": "${workspaceFolder}",
       "console": "integratedTerminal",
       "args": ["-d","..\\audiodata"]
     },
@@ -40,6 +43,7 @@
       "type": "python",
       "request": "launch",
       "program": "demo_toolbox.py",
+      "cwd": "${workspaceFolder}",
       "console": "integratedTerminal",
       "args": ["-d","..\\audiodata","-vc"]
     },
@@ -47,9 +51,9 @@
       "name": "Python: Synth Train",
       "type": "python",
       "request": "launch",
-      "program": "synthesizer_train.py",
+      "program": "train.py",
       "console": "integratedTerminal",
-      "args": ["my_run", "..\\"]
+      "args": ["--type", "synth", "..\\audiodata\\SV2TTS\\synthesizer"]
     },
     {
       "name": "Python: PPG Convert",
@@ -60,6 +64,6 @@
       "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\\"
       ]
-    },
+    }
   ]
 }

Dockerfile

@@ -0,0 +1,17 @@
+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

@@ -18,6 +18,20 @@
 
 🌍 **Webserver Ready** 可伺服你的训练结果，供远程调用
 
+### 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] 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 
+
+- [ ] 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
+
+
 ## 开始
 ### 1. 安装要求
 > 按照原始存储库测试您是否已准备好所有环境。
@@ -68,7 +82,7 @@
 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_mode\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_models\xxx*
 
 
 * 训练wavernn声码器:
@@ -78,19 +92,17 @@
 * 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：
 
-### 3.1 启动Web程序：
+### 3.1 启动Web程序（v2）：
 `python web.py`
 运行成功后在浏览器打开地址, 默认为 `http://localhost:8080`
-![123](https://user-images.githubusercontent.com/12797292/135494044-ae59181c-fe3a-406f-9c7d-d21d12fdb4cb.png)
-> 注：目前界面比较buggy, 
-> * 第一次点击`录制`要等待几秒浏览器正常启动录音，否则会有重音
-> * 录制结束不要再点`录制`而是`停止`
 > * 仅支持手动新录音（16khz）, 不支持超过4MB的录音，最佳长度在5~15秒
-> * 默认使用第一个找到的模型，有动手能力的可以看代码修改 `web\__init__.py`。
 
 ### 3.2 启动工具箱：
 `python demo_toolbox.py -d <datasets_root>`
@@ -101,12 +113,12 @@
 ### 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 -r requirements_vc.txt` 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 espnet` 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_mode\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_mode\xxx*
+  * 24K采样率专用的vocoder（hifigan）到 *vocoder\saved_models\xxx*
+  * 预训练的ppg特征encoder(ppg_extractor)到 *ppg_extractor\saved_models\xxx*
+  * 预训练的PPG2Mel到 *ppg2mel\saved_models\xxx*
 
 #### 4.1 使用数据集自己训练PPG2Mel模型 (可选)
 
@@ -115,7 +127,7 @@
 `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 11770k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 11700k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：
@@ -124,7 +136,7 @@
 
 #### 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">
 
@@ -135,35 +147,36 @@
 | --- | ----------- | ----- | --------------------- |
 | [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 |
 
-## 常見問題(FQ&A)
-#### 1.數據集哪裡下載?
+## 常见问题(FQ&A)
+#### 1.数据集在哪里下载?
 | 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/) |  |
-> 解壓 aidatatang_200zh 後，還需將 `aidatatang_200zh\corpus\train`下的檔案全選解壓縮
+> 解压 aidatatang_200zh 后，还需将 `aidatatang_200zh\corpus\train`下的文件全选解压缩
 
 #### 2.`<datasets_root>`是什麼意思?
-假如數據集路徑為 `D:\data\aidatatang_200zh`，那麼 `<datasets_root>`就是 `D:\data`
+假如数据集路径为 `D:\data\aidatatang_200zh`，那么 `<datasets_root>`就是 `D:\data`
 
-#### 3.訓練模型顯存不足
-訓練合成器時：將 `synthesizer/hparams.py`中的batch_size參數調小
+#### 3.训练模型显存不足
+训练合成器时：将 `synthesizer/hparams.py`中的batch_size参数调小
 ```
-//調整前
+//调整前
 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
-//調整後
+//调整后
 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),   #
@@ -172,15 +185,15 @@ tts_schedule = [(2,  1e-3,  20_000,  8),   # Progressive training schedule
                 (2,  1e-5, 640_000,  8)],  # lr = learning rate
 ```
 
-聲碼器-預處理數據集時：將 `synthesizer/hparams.py`中的batch_size參數調小
+声码器-预处理数据集时：将 `synthesizer/hparams.py`中的batch_size参数调小
 ```
-//調整前
+//调整前
 ### Data Preprocessing
         max_mel_frames = 900,
         rescale = True,
         rescaling_max = 0.9,
         synthesis_batch_size = 16,                  # For vocoder preprocessing and inference.
-//調整後
+//调整后
 ### Data Preprocessing
         max_mel_frames = 900,
         rescale = True,
@@ -188,16 +201,16 @@ tts_schedule = [(2,  1e-3,  20_000,  8),   # Progressive training schedule
         synthesis_batch_size = 8,                  # For vocoder preprocessing and inference.
 ```
 
-聲碼器-訓練聲碼器時：將 `vocoder/wavernn/hparams.py`中的batch_size參數調小
+声码器-训练声码器时：将 `vocoder/wavernn/hparams.py`中的batch_size参数调小
 ```
-//調整前
+//调整前
 # Training
 voc_batch_size = 100
 voc_lr = 1e-4
 voc_gen_at_checkpoint = 5
 voc_pad = 2
 
-//調整後
+//调整后
 # Training
 voc_batch_size = 6
 voc_lr = 1e-4
@@ -206,13 +219,13 @@ 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]).`
-請參照 issue [#37](https://github.com/babysor/MockingBird/issues/37)
+请参照 issue [#37](https://github.com/babysor/MockingBird/issues/37)
 
-#### 5.如何改善CPU、GPU佔用率?
-適情況調整batch_size參數來改善
+#### 5.如何改善CPU、GPU占用率?
+视情况调整batch_size参数来改善
 
-#### 6.發生 `頁面文件太小，無法完成操作`
-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
+#### 6.发生 `页面文件太小，无法完成操作`
+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
 
 #### 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

@@ -18,6 +18,14 @@
 
 ### [DEMO VIDEO](https://www.bilibili.com/video/BV17Q4y1B7mY/)
 
+### Ongoing Works(Helps Needed)
+* Major upgrade on GUI/Client and unifying web and toolbox
+[X] Init framework `./mkgui` and [tech design](https://vaj2fgg8yn.feishu.cn/docs/doccnvotLWylBub8VJIjKzoEaee)
+[X] Add demo part of Voice Cloning and Conversion
+[X] Add preprocessing and training for Voice Conversion
+[ ] Add preprocessing and training for Encoder/Synthesizer/Vocoder
+* Major upgrade on model backend based on ESPnet2(not yet started)
+
 ## Quick Start
 
 ### 1. Install Requirements
@@ -29,7 +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 sympols. 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 symbols. 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:
 
@@ -60,7 +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.bilibili.com/video/BV1uh411B7AD/ | 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
 
 #### 2.4 Train vocoder (Optional)
 > note: vocoder has little difference in effect, so you may not need to train a new one.
@@ -82,6 +90,11 @@ 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.
 
@@ -89,6 +102,7 @@ You can then try the toolbox:
 | --- | ----------- | ----- | --------------------- |
 | [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 encoder.params_model import model_embedding_size as speaker_embedding_size
+from models.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 synthesizer.inference import Synthesizer
-from encoder import inference as encoder
-from vocoder import inference as vocoder
+from models.synthesizer.inference import Synthesizer
+from models.encoder import inference as encoder
+from models.vocoder import inference as vocoder
 from pathlib import Path
 import numpy as np
 import soundfile as sf

control/cli/encoder_preprocess.py

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

control/cli/encoder_train.py

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

control/cli/ppg2mel_train.py

@@ -2,8 +2,8 @@ import sys
 import torch
 import argparse
 import numpy as np
-from utils.load_yaml import HpsYaml
-from ppg2mel.train.train_linglf02mel_seq2seq_oneshotvc import Solver
+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

control/cli/pre4ppg.py

@@ -1,7 +1,7 @@
 from pathlib import Path
 import argparse
 
-from ppg2mel.preprocess import preprocess_dataset
+from models.ppg2mel.preprocess import preprocess_dataset
 from pathlib import Path
 import argparse
 

control/cli/synthesizer_train.py

@@ -1,10 +1,9 @@
-from synthesizer.hparams import hparams
-from synthesizer.train import train
+from models.synthesizer.hparams import hparams
+from models.synthesizer.train import train
 from utils.argutils import print_args
 import argparse
 
-
-if __name__ == "__main__":
+def new_train():
     parser = argparse.ArgumentParser()
     parser.add_argument("run_id", type=str, help= \
         "Name for this model instance. If a model state from the same run ID was previously "
@@ -13,7 +12,7 @@ if __name__ == "__main__":
     parser.add_argument("syn_dir", type=str, default=argparse.SUPPRESS, help= \
         "Path to the synthesizer directory that contains the ground truth mel spectrograms, "
         "the wavs and the embeds.")
-    parser.add_argument("-m", "--models_dir", type=str, default="synthesizer/saved_models/", help=\
+    parser.add_argument("-m", "--models_dir", type=str, default=f"data/ckpt/synthesizer/", help=\
         "Path to the output directory that will contain the saved model weights and the logs.")
     parser.add_argument("-s", "--save_every", type=int, default=1000, help= \
         "Number of steps between updates of the model on the disk. Set to 0 to never save the "
@@ -28,10 +27,14 @@ if __name__ == "__main__":
     parser.add_argument("--hparams", default="",
                         help="Hyperparameter overrides as a comma-separated list of name=value "
 							 "pairs")
-    args = parser.parse_args()
+    args, _ = parser.parse_known_args()
     print_args(args, parser)
 
     args.hparams = hparams.parse(args.hparams)
 
     # Run the training
     train(**vars(args))
+
+
+if __name__ == "__main__":
+    new_train()

control/cli/train_ppg2mel.py

@@ -2,8 +2,8 @@ import sys
 import torch
 import argparse
 import numpy as np
-from utils.load_yaml import HpsYaml
-from ppg2mel.train.train_linglf02mel_seq2seq_oneshotvc import Solver
+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
@@ -18,7 +18,7 @@ def main():
     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='ckpt/', type=str,
+    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)
@@ -41,7 +41,6 @@ def main():
     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)

control/cli/vocoder_preprocess.py

@@ -1,5 +1,5 @@
-from synthesizer.synthesize import run_synthesis
-from synthesizer.hparams import hparams
+from models.synthesizer.synthesize import run_synthesis
+from models.synthesizer.hparams import hparams
 from utils.argutils import print_args
 import argparse
 import os

control/cli/vocoder_train.py

@@ -1,11 +1,13 @@
 from utils.argutils import print_args
-from vocoder.wavernn.train import train
-from vocoder.hifigan.train import train as train_hifigan
-from vocoder.hifigan.env import AttrDict
+from models.vocoder.wavernn.train import train
+from models.vocoder.hifigan.train import train as train_hifigan
+from models.vocoder.fregan.train import train as train_fregan
+from utils.util import AttrDict
 from pathlib import Path
 import argparse
 import json
-
+import torch
+import torch.multiprocessing as mp
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser(
@@ -61,11 +63,30 @@ if __name__ == "__main__":
     # Process the arguments
     if args.vocoder_type == "wavernn":
         # Run the training wavernn
+        delattr(args, 'vocoder_type')
+        delattr(args, 'config')
         train(**vars(args))
     elif args.vocoder_type == "hifigan":
         with open(args.config) as f:
             json_config = json.load(f)
         h = AttrDict(json_config)
-        train_hifigan(0, args, h)
+        if h.num_gpus > 1:
+            h.num_gpus = torch.cuda.device_count()
+            h.batch_size = int(h.batch_size / h.num_gpus)
+            print('Batch size per GPU :', h.batch_size)
+            mp.spawn(train_hifigan, nprocs=h.num_gpus, args=(args, h,))
+        else:
+            train_hifigan(0, args, h)
+    elif args.vocoder_type == "fregan":
+        with open('vocoder/fregan/config.json') as f:
+            json_config = json.load(f)
+        h = AttrDict(json_config)
+        if h.num_gpus > 1:
+            h.num_gpus = torch.cuda.device_count()
+            h.batch_size = int(h.batch_size / h.num_gpus)
+            print('Batch size per GPU :', h.batch_size)
+            mp.spawn(train_fregan, nprocs=h.num_gpus, args=(args, h,))
+        else:
+            train_fregan(0, args, h)
 
         

control/mkgui/app.py

@@ -1,24 +1,27 @@
-from asyncio.windows_events import NULL
 from pydantic import BaseModel, Field
 import os
 from pathlib import Path
 from enum import Enum
-from encoder import inference as encoder
+from models.encoder import inference as encoder
 import librosa
 from scipy.io.wavfile import write
 import re
 import numpy as np
-from opyrator.components.types import FileContent
-from vocoder.hifigan import inference as gan_vocoder
-from synthesizer.inference import Synthesizer
+from control.mkgui.base.components.types import FileContent
+from models.vocoder.hifigan import inference as gan_vocoder
+from models.synthesizer.inference import Synthesizer
+from typing import Any, Tuple
+import matplotlib.pyplot as plt
 
 # Constants
-AUDIO_SAMPLES_DIR = 'samples\\'
-SYN_MODELS_DIRT = "synthesizer\\saved_models"
-ENC_MODELS_DIRT = "encoder\\saved_models"
-VOC_MODELS_DIRT = "vocoder\\saved_models"
-TEMP_SOURCE_AUDIO = "wavs/temp_source.wav"
-TEMP_RESULT_AUDIO = "wavs/temp_result.wav"
+AUDIO_SAMPLES_DIR = f"data{os.sep}samples{os.sep}"
+SYN_MODELS_DIRT = f"data{os.sep}ckpt{os.sep}synthesizer"
+ENC_MODELS_DIRT = f"data{os.sep}ckpt{os.sep}encoder"
+VOC_MODELS_DIRT = f"data{os.sep}ckpt{os.sep}vocoder"
+TEMP_SOURCE_AUDIO = f"wavs{os.sep}temp_source.wav"
+TEMP_RESULT_AUDIO = f"wavs{os.sep}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):
@@ -27,20 +30,33 @@ if os.path.isdir(AUDIO_SAMPLES_DIR):
 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！")
+
 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.")
+
 if os.path.isdir(VOC_MODELS_DIRT):    
     vocoders =  Enum('vocoders', list((file.name, file) for file in Path(VOC_MODELS_DIRT).glob("**/*gan*.pt")))
     print("Loaded vocoders models: " + str(len(synthesizers)))
+else:
+    raise Exception(f"Model folder {VOC_MODELS_DIRT} doesn't exist.")
 
 
 class Input(BaseModel):
+    message: str = Field(
+        ..., 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="输入语音（本地wav）",
+        ..., alias="选择语音（本地wav）",
         description="选择本地语音文件."
     )
-    upload_audio_file: FileContent = Field(..., alias="或上传语音",
+    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(
         ..., alias="编码模型", 
@@ -48,37 +64,53 @@ class Input(BaseModel):
     )
     synthesizer: synthesizers = Field(
         ..., alias="合成模型", 
-        description="选择语音编码模型文件."
+        description="选择语音合成模型文件."
     )
     vocoder: vocoders = Field(
-        ..., alias="语音编码模型", 
-        description="选择语音编码模型文件(目前只支持HifiGan类型)."
-    )
-    message: str = Field(
-        ..., example="欢迎使用工具箱, 现已支持中文输入！", alias="输出文本内容"
+        ..., alias="语音解码模型", 
+        description="选择语音解码模型文件(目前只支持HifiGan类型)."
     )
 
+class AudioEntity(BaseModel):
+    content: bytes
+    mel: Any
+
 class Output(BaseModel):
-    result_file: FileContent = Field(
-        ...,
-        mime_type="audio/wav",
-        description="输出音频",
-    )
-    source_file: FileContent = Field(
-        ...,
-        mime_type="audio/wav",
-        description="原始音频.",
-    )
+    __root__: Tuple[AudioEntity, AudioEntity]
 
-def mocking_bird(input: Input) -> Output:
-    """欢迎使用MockingBird Web 2"""
+    def render_output_ui(self, streamlit_app, input) -> None:  # type: ignore
+        """Custom output UI.
+        If this method is implmeneted, it will be used instead of the default Output UI renderer.
+        """
+        src, result = self.__root__
+        
+        streamlit_app.subheader("Synthesized Audio")
+        streamlit_app.audio(result.content, format="audio/wav")
+
+        fig, ax = plt.subplots()
+        ax.imshow(src.mel, aspect="equal", interpolation="none")
+        ax.set_title("mel spectrogram(Source Audio)")
+        streamlit_app.pyplot(fig)
+        fig, ax = plt.subplots()
+        ax.imshow(result.mel, aspect="equal", interpolation="none")
+        ax.set_title("mel spectrogram(Result Audio)")
+        streamlit_app.pyplot(fig)
+
+
+def synthesize(input: Input) -> Output:
+    """synthesize(合成)"""
     # load models
     encoder.load_model(Path(input.encoder.value))
     current_synt = Synthesizer(Path(input.synthesizer.value))
     gan_vocoder.load_model(Path(input.vocoder.value))
 
     # load file
-    if input.upload_audio_file != NULL:
+    if input.record_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)
@@ -87,6 +119,8 @@ def mocking_bird(input: Input) -> Output:
         wav, sample_rate  = librosa.load(input.local_audio_file.value)
         write(TEMP_SOURCE_AUDIO, sample_rate, wav) #Make sure we get the correct wav
 
+    source_spec = Synthesizer.make_spectrogram(wav)
+
     # preprocess
     encoder_wav = encoder.preprocess_wav(wav, sample_rate)
     embed, _, _ = encoder.embed_utterance(encoder_wav, return_partials=True)
@@ -114,4 +148,4 @@ def mocking_bird(input: Input) -> Output:
         source_file = f.read()
     with open(TEMP_RESULT_AUDIO, "rb") as f:
         result_file = f.read()
-    return Output(source_file=source_file, result_file=result_file)
+    return Output(__root__=(AudioEntity(content=source_file, mel=source_spec), AudioEntity(content=result_file, mel=spec)))

control/mkgui/app_vc.py

@@ -0,0 +1,166 @@
+import os
+from enum import Enum
+from pathlib import Path
+from typing import Any, Tuple
+
+import librosa
+import matplotlib.pyplot as plt
+import torch
+from pydantic import BaseModel, Field
+from scipy.io.wavfile import write
+
+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}'
+EXT_MODELS_DIRT = f'data{os.sep}ckpt{os.sep}ppg_extractor'
+CONV_MODELS_DIRT = f'data{os.sep}ckpt{os.sep}ppg2mel'
+VOC_MODELS_DIRT = f'data{os.sep}ckpt{os.sep}vocoder'
+TEMP_SOURCE_AUDIO = f'wavs{os.sep}temp_source.wav'
+TEMP_TARGET_AUDIO = f'wavs{os.sep}temp_target.wav'
+TEMP_RESULT_AUDIO = f'wavs{os.sep}temp_result.wav'
+
+# Load local sample audio as options TODO: load dataset 
+if os.path.isdir(AUDIO_SAMPLES_DIR):
+    audio_input_selection = Enum('samples', list((file.name, file) for file in Path(AUDIO_SAMPLES_DIR).glob("*.wav")))
+# Pre-Load models
+if os.path.isdir(EXT_MODELS_DIRT):    
+    extractors =  Enum('extractors', list((file.name, file) for file in Path(EXT_MODELS_DIRT).glob("**/*.pt")))
+    print("Loaded extractor models: " + str(len(extractors)))
+else:
+    raise Exception(f"Model folder {EXT_MODELS_DIRT} doesn't exist.")
+
+if os.path.isdir(CONV_MODELS_DIRT):    
+    convertors =  Enum('convertors', list((file.name, file) for file in Path(CONV_MODELS_DIRT).glob("**/*.pth")))
+    print("Loaded convertor models: " + str(len(convertors)))
+else:
+    raise Exception(f"Model folder {CONV_MODELS_DIRT} doesn't exist.")
+
+if os.path.isdir(VOC_MODELS_DIRT):    
+    vocoders =  Enum('vocoders', list((file.name, file) for file in Path(VOC_MODELS_DIRT).glob("**/*gan*.pt")))
+    print("Loaded vocoders models: " + str(len(vocoders)))
+else:
+    raise Exception(f"Model folder {VOC_MODELS_DIRT} doesn't exist.")
+
+class Input(BaseModel):
+    local_audio_file: audio_input_selection = Field(
+        ..., alias="输入语音（本地wav）",
+        description="选择本地语音文件."
+    )
+    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")
+    local_audio_file_target: audio_input_selection = Field(
+        ..., alias="目标语音（本地wav）",
+        description="选择本地语音文件."
+    )
+    upload_audio_file_target: FileContent = Field(default=None, alias="或上传目标语音",
+        description="拖拽或点击上传.", mime_type="audio/wav")
+    extractor: extractors = Field(
+        ..., alias="编码模型", 
+        description="选择语音编码模型文件."
+    )
+    convertor: convertors = Field(
+        ..., alias="转换模型", 
+        description="选择语音转换模型文件."
+    )
+    vocoder: vocoders = Field(
+        ..., alias="语音解码模型", 
+        description="选择语音解码模型文件(目前只支持HifiGan类型)."
+    )
+
+class AudioEntity(BaseModel):
+    content: bytes
+    mel: Any
+
+class Output(BaseModel):
+    __root__: Tuple[AudioEntity, AudioEntity, AudioEntity]
+
+    def render_output_ui(self, streamlit_app, input) -> None:  # type: ignore
+        """Custom output UI.
+        If this method is implmeneted, it will be used instead of the default Output UI renderer.
+        """
+        src, target, result = self.__root__
+        
+        streamlit_app.subheader("Synthesized Audio")
+        streamlit_app.audio(result.content, format="audio/wav")
+
+        fig, ax = plt.subplots()
+        ax.imshow(src.mel, aspect="equal", interpolation="none")
+        ax.set_title("mel spectrogram(Source Audio)")
+        streamlit_app.pyplot(fig)
+        fig, ax = plt.subplots()
+        ax.imshow(target.mel, aspect="equal", interpolation="none")
+        ax.set_title("mel spectrogram(Target Audio)")
+        streamlit_app.pyplot(fig)
+        fig, ax = plt.subplots()
+        ax.imshow(result.mel, aspect="equal", interpolation="none")
+        ax.set_title("mel spectrogram(Result Audio)")
+        streamlit_app.pyplot(fig)
+
+def convert(input: Input) -> Output:
+    """convert(转换)"""
+    # load models
+    extractor = Extractor.load_model(Path(input.extractor.value))
+    convertor = Convertor.load_model(Path(input.convertor.value))
+    # current_synt = Synthesizer(Path(input.synthesizer.value))
+    gan_vocoder.load_model(Path(input.vocoder.value))
+
+    # load file
+    if 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)
+        src_wav, sample_rate = librosa.load(TEMP_SOURCE_AUDIO)
+    else:
+        src_wav, sample_rate  = librosa.load(input.local_audio_file.value)
+        write(TEMP_SOURCE_AUDIO, sample_rate, src_wav) #Make sure we get the correct wav
+
+    if input.upload_audio_file_target != None:
+        with open(TEMP_TARGET_AUDIO, "w+b") as f:
+            f.write(input.upload_audio_file_target.as_bytes())
+            f.seek(0)
+        ref_wav, _ = librosa.load(TEMP_TARGET_AUDIO)
+    else:
+        ref_wav, _  = librosa.load(input.local_audio_file_target.value)
+        write(TEMP_TARGET_AUDIO, sample_rate, ref_wav) #Make sure we get the correct wav
+
+    ppg = extractor.extract_from_wav(src_wav)
+    # Import necessary dependency of Voice Conversion
+    from utils.f0_utils import (compute_f0, compute_mean_std, f02lf0,
+                                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"))
+    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))
+    ppg = ppg[:, :min_len]
+    lf0_uv = lf0_uv[:min_len]
+    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+    _, mel_pred, att_ws = convertor.inference(
+        ppg,
+        logf0_uv=torch.from_numpy(lf0_uv).unsqueeze(0).float().to(device),
+        spembs=torch.from_numpy(embed).unsqueeze(0).to(device),
+    )
+    mel_pred= mel_pred.transpose(0, 1)
+    breaks = [mel_pred.shape[1]]
+    mel_pred= mel_pred.detach().cpu().numpy()
+
+    # synthesize and vocode
+    wav, sample_rate = gan_vocoder.infer_waveform(mel_pred)
+
+    # write and output 
+    write(TEMP_RESULT_AUDIO, sample_rate, wav) #Make sure we get the correct wav
+    with open(TEMP_SOURCE_AUDIO, "rb") as f:
+        source_file = f.read()
+    with open(TEMP_TARGET_AUDIO, "rb") as f:
+        target_file = f.read()
+    with open(TEMP_RESULT_AUDIO, "rb") as f:
+        result_file = f.read()
+    
+
+    return Output(__root__=(AudioEntity(content=source_file, mel=Synthesizer.make_spectrogram(src_wav)), AudioEntity(content=target_file, mel=Synthesizer.make_spectrogram(ref_wav)), AudioEntity(content=result_file, mel=Synthesizer.make_spectrogram(wav))))

control/mkgui/base/__init__.py

@@ -0,0 +1,2 @@
+
+from .core import Opyrator

control/mkgui/base/api/__init__.py

@@ -0,0 +1 @@
+from .fastapi_app import create_api

control/mkgui/base/api/fastapi_utils.py

@@ -0,0 +1,102 @@
+"""Collection of utilities for FastAPI apps."""
+
+import inspect
+from typing import Any, Type
+
+from fastapi import FastAPI, Form
+from pydantic import BaseModel
+
+
+def as_form(cls: Type[BaseModel]) -> Any:
+    """Adds an as_form class method to decorated models.
+
+    The as_form class method can be used with FastAPI endpoints
+    """
+    new_params = [
+        inspect.Parameter(
+            field.alias,
+            inspect.Parameter.POSITIONAL_ONLY,
+            default=(Form(field.default) if not field.required else Form(...)),
+        )
+        for field in cls.__fields__.values()
+    ]
+
+    async def _as_form(**data):  # type: ignore
+        return cls(**data)
+
+    sig = inspect.signature(_as_form)
+    sig = sig.replace(parameters=new_params)
+    _as_form.__signature__ = sig  # type: ignore
+    setattr(cls, "as_form", _as_form)
+    return cls
+
+
+def patch_fastapi(app: FastAPI) -> None:
+    """Patch function to allow relative url resolution.
+
+    This patch is required to make fastapi fully functional with a relative url path.
+    This code snippet can be copy-pasted to any Fastapi application.
+    """
+    from fastapi.openapi.docs import get_redoc_html, get_swagger_ui_html
+    from starlette.requests import Request
+    from starlette.responses import HTMLResponse
+
+    async def redoc_ui_html(req: Request) -> HTMLResponse:
+        assert app.openapi_url is not None
+        redoc_ui = get_redoc_html(
+            openapi_url="./" + app.openapi_url.lstrip("/"),
+            title=app.title + " - Redoc UI",
+        )
+
+        return HTMLResponse(redoc_ui.body.decode("utf-8"))
+
+    async def swagger_ui_html(req: Request) -> HTMLResponse:
+        assert app.openapi_url is not None
+        swagger_ui = get_swagger_ui_html(
+            openapi_url="./" + app.openapi_url.lstrip("/"),
+            title=app.title + " - Swagger UI",
+            oauth2_redirect_url=app.swagger_ui_oauth2_redirect_url,
+        )
+
+        # insert request interceptor to have all request run on relativ path
+        request_interceptor = (
+            "requestInterceptor: (e)  => {"
+            "\n\t\t\tvar url = window.location.origin + window.location.pathname"
+            '\n\t\t\turl = url.substring( 0, url.lastIndexOf( "/" ) + 1);'
+            "\n\t\t\turl = e.url.replace(/http(s)?:\/\/[^/]*\//i, url);"  # noqa: W605
+            "\n\t\t\te.contextUrl = url"
+            "\n\t\t\te.url = url"
+            "\n\t\t\treturn e;}"
+        )
+
+        return HTMLResponse(
+            swagger_ui.body.decode("utf-8").replace(
+                "dom_id: '#swagger-ui',",
+                "dom_id: '#swagger-ui',\n\t\t" + request_interceptor + ",",
+            )
+        )
+
+    # remove old docs route and add our patched route
+    routes_new = []
+    for app_route in app.routes:
+        if app_route.path == "/docs":  # type: ignore
+            continue
+
+        if app_route.path == "/redoc":  # type: ignore
+            continue
+
+        routes_new.append(app_route)
+
+    app.router.routes = routes_new
+
+    assert app.docs_url is not None
+    app.add_route(app.docs_url, swagger_ui_html, include_in_schema=False)
+    assert app.redoc_url is not None
+    app.add_route(app.redoc_url, redoc_ui_html, include_in_schema=False)
+
+    # Make graphql realtive
+    from starlette import graphql
+
+    graphql.GRAPHIQL = graphql.GRAPHIQL.replace(
+        "({{REQUEST_PATH}}", '("." + {{REQUEST_PATH}}'
+    )

control/mkgui/base/components/outputs.py

@@ -0,0 +1,43 @@
+from typing import List
+
+from pydantic import BaseModel
+
+
+class ScoredLabel(BaseModel):
+    label: str
+    score: float
+
+
+class ClassificationOutput(BaseModel):
+    __root__: List[ScoredLabel]
+
+    def __iter__(self):  # type: ignore
+        return iter(self.__root__)
+
+    def __getitem__(self, item):  # type: ignore
+        return self.__root__[item]
+
+    def render_output_ui(self, streamlit) -> None:  # type: ignore
+        import plotly.express as px
+
+        sorted_predictions = sorted(
+            [prediction.dict() for prediction in self.__root__],
+            key=lambda k: k["score"],
+        )
+
+        num_labels = len(sorted_predictions)
+        if len(sorted_predictions) > 10:
+            num_labels = streamlit.slider(
+                "Maximum labels to show: ",
+                min_value=1,
+                max_value=len(sorted_predictions),
+                value=len(sorted_predictions),
+            )
+        fig = px.bar(
+            sorted_predictions[len(sorted_predictions) - num_labels :],
+            x="score",
+            y="label",
+            orientation="h",
+        )
+        streamlit.plotly_chart(fig, use_container_width=True)
+        # fig.show()

control/mkgui/base/components/types.py

@@ -0,0 +1,46 @@
+import base64
+from typing import Any, Dict, overload
+
+
+class FileContent(str):
+    def as_bytes(self) -> bytes:
+        return base64.b64decode(self, validate=True)
+
+    def as_str(self) -> str:
+        return self.as_bytes().decode()
+
+    @classmethod
+    def __modify_schema__(cls, field_schema: Dict[str, Any]) -> None:
+        field_schema.update(format="byte")
+
+    @classmethod
+    def __get_validators__(cls) -> Any:  # type: ignore
+        yield cls.validate
+
+    @classmethod
+    def validate(cls, value: Any) -> "FileContent":
+        if isinstance(value, FileContent):
+            return value
+        elif isinstance(value, str):
+            return FileContent(value)
+        elif isinstance(value, (bytes, bytearray, memoryview)):
+            return FileContent(base64.b64encode(value).decode())
+        else:
+            raise Exception("Wrong type")
+
+# # 暂时无法使用，因为浏览器中没有考虑选择文件夹
+# class DirectoryContent(FileContent):
+#     @classmethod
+#     def __modify_schema__(cls, field_schema: Dict[str, Any]) -> None:
+#         field_schema.update(format="path")
+
+#     @classmethod
+#     def validate(cls, value: Any) -> "DirectoryContent":
+#         if isinstance(value, DirectoryContent):
+#             return value
+#         elif isinstance(value, str):
+#             return DirectoryContent(value)
+#         elif isinstance(value, (bytes, bytearray, memoryview)):
+#             return DirectoryContent(base64.b64encode(value).decode())
+#         else:
+#             raise Exception("Wrong type")

control/mkgui/base/core.py

@@ -0,0 +1,203 @@
+import importlib
+import inspect
+import re
+from typing import Any, Callable, Type, Union, get_type_hints
+
+from pydantic import BaseModel, parse_raw_as
+from pydantic.tools import parse_obj_as
+
+
+def name_to_title(name: str) -> str:
+    """Converts a camelCase or snake_case name to title case."""
+    # If camelCase -> convert to snake case
+    name = re.sub("(.)([A-Z][a-z]+)", r"\1_\2", name)
+    name = re.sub("([a-z0-9])([A-Z])", r"\1_\2", name).lower()
+    # Convert to title case
+    return name.replace("_", " ").strip().title()
+
+
+def is_compatible_type(type: Type) -> bool:
+    """Returns `True` if the type is opyrator-compatible."""
+    try:
+        if issubclass(type, BaseModel):
+            return True
+    except Exception:
+        pass
+
+    try:
+        # valid list type
+        if type.__origin__ is list and issubclass(type.__args__[0], BaseModel):
+            return True
+    except Exception:
+        pass
+
+    return False
+
+
+def get_input_type(func: Callable) -> Type:
+    """Returns the input type of a given function (callable).
+
+    Args:
+        func: The function for which to get the input type.
+
+    Raises:
+        ValueError: If the function does not have a valid input type annotation.
+    """
+    type_hints = get_type_hints(func)
+
+    if "input" not in type_hints:
+        raise ValueError(
+            "The callable MUST have a parameter with the name `input` with typing annotation. "
+            "For example: `def my_opyrator(input: InputModel) -> OutputModel:`."
+        )
+
+    input_type = type_hints["input"]
+
+    if not is_compatible_type(input_type):
+        raise ValueError(
+            "The `input` parameter MUST be a subclass of the Pydantic BaseModel or a list of Pydantic models."
+        )
+
+    # TODO: return warning if more than one input parameters
+
+    return input_type
+
+
+def get_output_type(func: Callable) -> Type:
+    """Returns the output type of a given function (callable).
+
+    Args:
+        func: The function for which to get the output type.
+
+    Raises:
+        ValueError: If the function does not have a valid output type annotation.
+    """
+    type_hints = get_type_hints(func)
+    if "return" not in type_hints:
+        raise ValueError(
+            "The return type of the callable MUST be annotated with type hints."
+            "For example: `def my_opyrator(input: InputModel) -> OutputModel:`."
+        )
+
+    output_type = type_hints["return"]
+
+    if not is_compatible_type(output_type):
+        raise ValueError(
+            "The return value MUST be a subclass of the Pydantic BaseModel or a list of Pydantic models."
+        )
+
+    return output_type
+
+
+def get_callable(import_string: str) -> Callable:
+    """Import a callable from an string."""
+    callable_seperator = ":"
+    if callable_seperator not in import_string:
+        # Use dot as seperator
+        callable_seperator = "."
+
+    if callable_seperator not in import_string:
+        raise ValueError("The callable path MUST specify the function. ")
+
+    mod_name, callable_name = import_string.rsplit(callable_seperator, 1)
+    mod = importlib.import_module(mod_name)
+    return getattr(mod, callable_name)
+
+
+class Opyrator:
+    def __init__(self, func: Union[Callable, str]) -> None:
+        if isinstance(func, str):
+            # Try to load the function from a string notion
+            self.function = get_callable(func)
+        else:
+            self.function = func
+
+        self._action = "Execute"
+        self._input_type = None
+        self._output_type = None
+
+        if not callable(self.function):
+            raise ValueError("The provided function parameters is not a callable.")
+
+        if inspect.isclass(self.function):
+            raise ValueError(
+                "The provided callable is an uninitialized Class. This is not allowed."
+            )
+
+        if inspect.isfunction(self.function):
+            # The provided callable is a function
+            self._input_type = get_input_type(self.function)
+            self._output_type = get_output_type(self.function)
+
+            try:
+                # Get name
+                self._name = name_to_title(self.function.__name__)
+            except Exception:
+                pass
+
+            try:
+                # Get description from function
+                doc_string = inspect.getdoc(self.function)
+                if doc_string:
+                    self._action = doc_string
+            except Exception:
+                pass
+        elif hasattr(self.function, "__call__"):
+            # The provided callable is a function
+            self._input_type = get_input_type(self.function.__call__)  # type: ignore
+            self._output_type = get_output_type(self.function.__call__)  # type: ignore
+
+            try:
+                # Get name
+                self._name = name_to_title(type(self.function).__name__)
+            except Exception:
+                pass
+
+            try:
+                # Get action from
+                doc_string = inspect.getdoc(self.function.__call__)  # type: ignore
+                if doc_string:
+                    self._action = doc_string
+
+                if (
+                    not self._action
+                    or self._action == "Call"
+                ):
+                    # Get docstring from class instead of __call__ function
+                    doc_string = inspect.getdoc(self.function)
+                    if doc_string:
+                        self._action = doc_string
+            except Exception:
+                pass
+        else:
+            raise ValueError("Unknown callable type.")
+
+    @property
+    def name(self) -> str:
+        return self._name
+
+    @property
+    def action(self) -> str:
+        return self._action
+
+    @property
+    def input_type(self) -> Any:
+        return self._input_type
+
+    @property
+    def output_type(self) -> Any:
+        return self._output_type
+
+    def __call__(self, input: Any, **kwargs: Any) -> Any:
+
+        input_obj = input
+
+        if isinstance(input, str):
+            # Allow json input
+            input_obj = parse_raw_as(self.input_type, input)
+
+        if isinstance(input, dict):
+            # Allow dict input
+            input_obj = parse_obj_as(self.input_type, input)
+
+        return self.function(input_obj, **kwargs)

control/mkgui/base/ui/__init__.py

@@ -0,0 +1 @@
+from .streamlit_ui import render_streamlit_ui

control/mkgui/base/ui/schema_utils.py

@@ -0,0 +1,135 @@
+from typing import Dict
+
+
+def resolve_reference(reference: str, references: Dict) -> Dict:
+    return references[reference.split("/")[-1]]
+
+
+def get_single_reference_item(property: Dict, references: Dict) -> Dict:
+    # Ref can either be directly in the properties or the first element of allOf
+    reference = property.get("$ref")
+    if reference is None:
+        reference = property["allOf"][0]["$ref"]
+    return resolve_reference(reference, references)
+
+
+def is_single_string_property(property: Dict) -> bool:
+    return property.get("type") == "string"
+
+
+def is_single_datetime_property(property: Dict) -> bool:
+    if property.get("type") != "string":
+        return False
+    return property.get("format") in ["date-time", "time", "date"]
+
+
+def is_single_boolean_property(property: Dict) -> bool:
+    return property.get("type") == "boolean"
+
+
+def is_single_number_property(property: Dict) -> bool:
+    return property.get("type") in ["integer", "number"]
+
+
+def is_single_file_property(property: Dict) -> bool:
+    if property.get("type") != "string":
+        return False
+    # 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":
+        return False
+    return property.get("format") == "path"
+
+def is_multi_enum_property(property: Dict, references: Dict) -> bool:
+    if property.get("type") != "array":
+        return False
+
+    if property.get("uniqueItems") is not True:
+        # Only relevant if it is a set or other datastructures with unique items
+        return False
+
+    try:
+        _ = resolve_reference(property["items"]["$ref"], references)["enum"]
+        return True
+    except Exception:
+        return False
+
+
+def is_single_enum_property(property: Dict, references: Dict) -> bool:
+    try:
+        _ = get_single_reference_item(property, references)["enum"]
+        return True
+    except Exception:
+        return False
+
+
+def is_single_dict_property(property: Dict) -> bool:
+    if property.get("type") != "object":
+        return False
+    return "additionalProperties" in property
+
+
+def is_single_reference(property: Dict) -> bool:
+    if property.get("type") is not None:
+        return False
+
+    return bool(property.get("$ref"))
+
+
+def is_multi_file_property(property: Dict) -> bool:
+    if property.get("type") != "array":
+        return False
+
+    if property.get("items") is None:
+        return False
+
+    try:
+        # TODO: binary
+        return property["items"]["format"] == "byte"
+    except Exception:
+        return False
+
+
+def is_single_object(property: Dict, references: Dict) -> bool:
+    try:
+        object_reference = get_single_reference_item(property, references)
+        if object_reference["type"] != "object":
+            return False
+        return "properties" in object_reference
+    except Exception:
+        return False
+
+
+def is_property_list(property: Dict) -> bool:
+    if property.get("type") != "array":
+        return False
+
+    if property.get("items") is None:
+        return False
+
+    try:
+        return property["items"]["type"] in ["string", "number", "integer"]
+    except Exception:
+        return False
+
+
+def is_object_list_property(property: Dict, references: Dict) -> bool:
+    if property.get("type") != "array":
+        return False
+
+    try:
+        object_reference = resolve_reference(property["items"]["$ref"], references)
+        if object_reference["type"] != "object":
+            return False
+        return "properties" in object_reference
+    except Exception:
+        return False

control/mkgui/base/ui/streamlit_ui.py

@@ -0,0 +1,933 @@
+import datetime
+import inspect
+import mimetypes
+import sys
+from os import getcwd, unlink, path
+from platform import system
+from tempfile import NamedTemporaryFile
+from typing import Any, Callable, Dict, List, Type
+from PIL import Image
+
+import pandas as pd
+import streamlit as st
+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 control.mkgui.base.core import name_to_title
+from . import schema_utils
+from .streamlit_utils import CUSTOM_STREAMLIT_CSS
+
+STREAMLIT_RUNNER_SNIPPET = """
+from control.mkgui.base.ui import render_streamlit_ui
+
+import streamlit as st
+
+# TODO: Make it configurable
+# Page config can only be setup once
+st.set_page_config(
+    page_title="MockingBird", 
+    page_icon="🧊",
+    layout="wide")
+
+render_streamlit_ui()
+"""
+
+# with st.spinner("Loading MockingBird GUI. Please wait..."):
+#     opyrator = Opyrator("{opyrator_path}")
+
+
+def launch_ui(port: int = 8501) -> None:
+    with NamedTemporaryFile(
+        suffix=".py", mode="w", encoding="utf-8", delete=False
+    ) as f:
+        f.write(STREAMLIT_RUNNER_SNIPPET)
+        f.seek(0)
+
+        import subprocess
+
+        python_path = f'PYTHONPATH="$PYTHONPATH:{getcwd()}"'
+        if system() == "Windows":
+            python_path = f"set PYTHONPATH=%PYTHONPATH%;{getcwd()} &&"
+            subprocess.run(
+                f"""set STREAMLIT_GLOBAL_SHOW_WARNING_ON_DIRECT_EXECUTION=false""",
+                shell=True,
+            )
+
+        subprocess.run(
+            f"""{python_path} "{sys.executable}" -m streamlit run --server.port={port} --server.headless=True --runner.magicEnabled=False --server.maxUploadSize=50  --browser.gatherUsageStats=False {f.name}""",
+            shell=True,
+        )
+
+        f.close()
+        unlink(f.name)
+
+
+def function_has_named_arg(func: Callable, parameter: str) -> bool:
+    try:
+        sig = inspect.signature(func)
+        for param in sig.parameters.values():
+            if param.name == "input":
+                return True
+    except Exception:
+        return False
+    return False
+
+
+def has_output_ui_renderer(data_item: BaseModel) -> bool:
+    return hasattr(data_item, "render_output_ui")
+
+
+def has_input_ui_renderer(input_class: Type[BaseModel]) -> bool:
+    return hasattr(input_class, "render_input_ui")
+
+
+def is_compatible_audio(mime_type: str) -> bool:
+    return mime_type in ["audio/mpeg", "audio/ogg", "audio/wav"]
+
+
+def is_compatible_image(mime_type: str) -> bool:
+    return mime_type in ["image/png", "image/jpeg"]
+
+
+def is_compatible_video(mime_type: str) -> bool:
+    return mime_type in ["video/mp4"]
+
+
+class InputUI:
+    def __init__(self, session_state, input_class: Type[BaseModel]):
+        self._session_state = session_state
+        self._input_class = input_class
+
+        self._schema_properties = input_class.schema(by_alias=True).get(
+            "properties", {}
+        )
+        self._schema_references = input_class.schema(by_alias=True).get(
+            "definitions", {}
+        )
+
+    def render_ui(self, streamlit_app_root) -> None:
+        if has_input_ui_renderer(self._input_class):
+            # The input model has a rendering function
+            # The rendering also returns the current state of input data
+            self._session_state.input_data = self._input_class.render_input_ui(  # type: ignore
+                st, self._session_state.input_data
+            )
+            return
+
+        # print(self._schema_properties)
+        for property_key in self._schema_properties.keys():
+            property = self._schema_properties[property_key]
+
+            if not property.get("title"):
+                # Set property key as fallback title
+                property["title"] = name_to_title(property_key)
+
+            try:
+                if "input_data" in self._session_state:
+                    self._store_value(
+                        property_key,
+                        self._render_property(streamlit_app_root, property_key, property),
+                    )
+            except Exception as e:
+                print("Exception!", e)
+                pass
+
+    def _get_default_streamlit_input_kwargs(self, key: str, property: Dict) -> Dict:
+        streamlit_kwargs = {
+            "label": property.get("title"),
+            "key": key,
+        }
+
+        if property.get("description"):
+            streamlit_kwargs["help"] = property.get("description")
+        return streamlit_kwargs
+
+    def _store_value(self, key: str, value: Any) -> None:
+        data_element = self._session_state.input_data
+        key_elements = key.split(".")
+        for i, key_element in enumerate(key_elements):
+            if i == len(key_elements) - 1:
+                # add value to this element
+                data_element[key_element] = value
+                return
+            if key_element not in data_element:
+                data_element[key_element] = {}
+            data_element = data_element[key_element]
+
+    def _get_value(self, key: str) -> Any:
+        data_element = self._session_state.input_data
+        key_elements = key.split(".")
+        for i, key_element in enumerate(key_elements):
+            if i == len(key_elements) - 1:
+                # add value to this element
+                if key_element not in data_element:
+                    return None
+                return data_element[key_element]
+            if key_element not in data_element:
+                data_element[key_element] = {}
+            data_element = data_element[key_element]
+        return None
+
+    def _render_single_datetime_input(
+        self, streamlit_app: st, key: str, property: Dict
+    ) -> Any:
+        streamlit_kwargs = self._get_default_streamlit_input_kwargs(key, property)
+
+        if property.get("format") == "time":
+            if property.get("default"):
+                try:
+                    streamlit_kwargs["value"] = datetime.time.fromisoformat(  # type: ignore
+                        property.get("default")
+                    )
+                except Exception:
+                    pass
+            return streamlit_app.time_input(**streamlit_kwargs)
+        elif property.get("format") == "date":
+            if property.get("default"):
+                try:
+                    streamlit_kwargs["value"] = datetime.date.fromisoformat(  # type: ignore
+                        property.get("default")
+                    )
+                except Exception:
+                    pass
+            return streamlit_app.date_input(**streamlit_kwargs)
+        elif property.get("format") == "date-time":
+            if property.get("default"):
+                try:
+                    streamlit_kwargs["value"] = datetime.datetime.fromisoformat(  # type: ignore
+                        property.get("default")
+                    )
+                except Exception:
+                    pass
+            with streamlit_app.container():
+                streamlit_app.subheader(streamlit_kwargs.get("label"))
+                if streamlit_kwargs.get("description"):
+                    streamlit_app.text(streamlit_kwargs.get("description"))
+                selected_date = None
+                selected_time = None
+                date_col, time_col = streamlit_app.columns(2)
+                with date_col:
+                    date_kwargs = {"label": "Date", "key": key + "-date-input"}
+                    if streamlit_kwargs.get("value"):
+                        try:
+                            date_kwargs["value"] = streamlit_kwargs.get(  # type: ignore
+                                "value"
+                            ).date()
+                        except Exception:
+                            pass
+                    selected_date = streamlit_app.date_input(**date_kwargs)
+
+                with time_col:
+                    time_kwargs = {"label": "Time", "key": key + "-time-input"}
+                    if streamlit_kwargs.get("value"):
+                        try:
+                            time_kwargs["value"] = streamlit_kwargs.get(  # type: ignore
+                                "value"
+                            ).time()
+                        except Exception:
+                            pass
+                    selected_time = streamlit_app.time_input(**time_kwargs)
+                return datetime.datetime.combine(selected_date, selected_time)
+        else:
+            streamlit_app.warning(
+                "Date format is not supported: " + str(property.get("format"))
+            )
+
+    def _render_single_file_input(
+        self, streamlit_app: st, key: str, property: Dict
+    ) -> Any:
+        streamlit_kwargs = self._get_default_streamlit_input_kwargs(key, property)
+        file_extension = None
+        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
+        )
+        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_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:
+        streamlit_kwargs = self._get_default_streamlit_input_kwargs(key, property)
+
+        if property.get("default"):
+            streamlit_kwargs["value"] = property.get("default")
+        elif property.get("example"):
+            # TODO: also use example for other property types
+            # Use example as value if it is provided
+            streamlit_kwargs["value"] = property.get("example")
+
+        if property.get("maxLength") is not None:
+            streamlit_kwargs["max_chars"] = property.get("maxLength")
+
+        if (
+            property.get("format")
+            or (
+                property.get("maxLength") is not None
+                and int(property.get("maxLength")) < 140  # type: ignore
+            )
+            or property.get("writeOnly")
+        ):
+            # If any format is set, use single text input
+            # If max chars is set to less than 140, use single text input
+            # If write only -> password field
+            if property.get("writeOnly"):
+                streamlit_kwargs["type"] = "password"
+            return streamlit_app.text_input(**streamlit_kwargs)
+        else:
+            # Otherwise use multiline text area
+            return streamlit_app.text_area(**streamlit_kwargs)
+
+    def _render_multi_enum_input(
+        self, streamlit_app: st, key: str, property: Dict
+    ) -> Any:
+        streamlit_kwargs = self._get_default_streamlit_input_kwargs(key, property)
+        reference_item = schema_utils.resolve_reference(
+            property["items"]["$ref"], self._schema_references
+        )
+        # TODO: how to select defaults
+        return streamlit_app.multiselect(
+            **streamlit_kwargs, options=reference_item["enum"]
+        )
+
+    def _render_single_enum_input(
+        self, streamlit_app: st, key: str, property: Dict
+    ) -> Any:
+
+        streamlit_kwargs = self._get_default_streamlit_input_kwargs(key, property)
+        reference_item = schema_utils.get_single_reference_item(
+            property, self._schema_references
+        )
+
+        if property.get("default") is not None:
+            try:
+                streamlit_kwargs["index"] = reference_item["enum"].index(
+                    property.get("default")
+                )
+            except Exception:
+                # Use default selection
+                pass
+
+        return streamlit_app.selectbox(
+            **streamlit_kwargs, options=reference_item["enum"]
+        )
+
+    def _render_single_dict_input(
+        self, streamlit_app: st, key: str, property: Dict
+    ) -> Any:
+
+        # Add title and subheader
+        streamlit_app.subheader(property.get("title"))
+        if property.get("description"):
+            streamlit_app.markdown(property.get("description"))
+
+        streamlit_app.markdown("---")
+
+        current_dict = self._get_value(key)
+        if not current_dict:
+            current_dict = {}
+
+        key_col, value_col = streamlit_app.columns(2)
+
+        with key_col:
+            updated_key = streamlit_app.text_input(
+                "Key", value="", key=key + "-new-key"
+            )
+
+        with value_col:
+            # TODO: also add boolean?
+            value_kwargs = {"label": "Value", "key": key + "-new-value"}
+            if property["additionalProperties"].get("type") == "integer":
+                value_kwargs["value"] = 0  # type: ignore
+                updated_value = streamlit_app.number_input(**value_kwargs)
+            elif property["additionalProperties"].get("type") == "number":
+                value_kwargs["value"] = 0.0  # type: ignore
+                value_kwargs["format"] = "%f"
+                updated_value = streamlit_app.number_input(**value_kwargs)
+            else:
+                value_kwargs["value"] = ""
+                updated_value = streamlit_app.text_input(**value_kwargs)
+
+        streamlit_app.markdown("---")
+
+        with streamlit_app.container():
+            clear_col, add_col = streamlit_app.columns([1, 2])
+
+            with clear_col:
+                if streamlit_app.button("Clear Items", key=key + "-clear-items"):
+                    current_dict = {}
+
+            with add_col:
+                if (
+                    streamlit_app.button("Add Item", key=key + "-add-item")
+                    and updated_key
+                ):
+                    current_dict[updated_key] = updated_value
+
+        streamlit_app.write(current_dict)
+
+        return current_dict
+
+    def _render_single_reference(
+        self, streamlit_app: st, key: str, property: Dict
+    ) -> Any:
+        reference_item = schema_utils.get_single_reference_item(
+            property, self._schema_references
+        )
+        return self._render_property(streamlit_app, key, reference_item)
+
+    def _render_multi_file_input(
+        self, streamlit_app: st, key: str, property: Dict
+    ) -> Any:
+        streamlit_kwargs = self._get_default_streamlit_input_kwargs(key, property)
+
+        file_extension = None
+        if "mime_type" in property:
+            file_extension = mimetypes.guess_extension(property["mime_type"])
+
+        uploaded_files = streamlit_app.file_uploader(
+            **streamlit_kwargs, accept_multiple_files=True, type=file_extension
+        )
+        uploaded_files_bytes = []
+        if uploaded_files:
+            for uploaded_file in uploaded_files:
+                uploaded_files_bytes.append(uploaded_file.read())
+        return uploaded_files_bytes
+
+    def _render_single_boolean_input(
+        self, streamlit_app: st, key: str, property: Dict
+    ) -> Any:
+        streamlit_kwargs = self._get_default_streamlit_input_kwargs(key, property)
+
+        if property.get("default"):
+            streamlit_kwargs["value"] = property.get("default")
+        return streamlit_app.checkbox(**streamlit_kwargs)
+
+    def _render_single_number_input(
+        self, streamlit_app: st, key: str, property: Dict
+    ) -> Any:
+        streamlit_kwargs = self._get_default_streamlit_input_kwargs(key, property)
+
+        number_transform = int
+        if property.get("type") == "number":
+            number_transform = float  # type: ignore
+            streamlit_kwargs["format"] = "%f"
+
+        if "multipleOf" in property:
+            # Set stepcount based on multiple of parameter
+            streamlit_kwargs["step"] = number_transform(property["multipleOf"])
+        elif number_transform == int:
+            # Set step size to 1 as default
+            streamlit_kwargs["step"] = 1
+        elif number_transform == float:
+            # Set step size to 0.01 as default
+            # TODO: adapt to default value
+            streamlit_kwargs["step"] = 0.01
+
+        if "minimum" in property:
+            streamlit_kwargs["min_value"] = number_transform(property["minimum"])
+        if "exclusiveMinimum" in property:
+            streamlit_kwargs["min_value"] = number_transform(
+                property["exclusiveMinimum"] + streamlit_kwargs["step"]
+            )
+        if "maximum" in property:
+            streamlit_kwargs["max_value"] = number_transform(property["maximum"])
+
+        if "exclusiveMaximum" in property:
+            streamlit_kwargs["max_value"] = number_transform(
+                property["exclusiveMaximum"] - streamlit_kwargs["step"]
+            )
+
+        if property.get("default") is not None:
+            streamlit_kwargs["value"] = number_transform(property.get("default"))  # type: ignore
+        else:
+            if "min_value" in streamlit_kwargs:
+                streamlit_kwargs["value"] = streamlit_kwargs["min_value"]
+            elif number_transform == int:
+                streamlit_kwargs["value"] = 0
+            else:
+                # Set default value to step
+                streamlit_kwargs["value"] = number_transform(streamlit_kwargs["step"])
+
+        if "min_value" in streamlit_kwargs and "max_value" in streamlit_kwargs:
+            # TODO: Only if less than X steps
+            return streamlit_app.slider(**streamlit_kwargs)
+        else:
+            return streamlit_app.number_input(**streamlit_kwargs)
+
+    def _render_object_input(self, streamlit_app: st, key: str, property: Dict) -> Any:
+        properties = property["properties"]
+        object_inputs = {}
+        for property_key in properties:
+            property = properties[property_key]
+            if not property.get("title"):
+                # Set property key as fallback title
+                property["title"] = name_to_title(property_key)
+            # construct full key based on key parts -> required later to get the value
+            full_key = key + "." + property_key
+            object_inputs[property_key] = self._render_property(
+                streamlit_app, full_key, property
+            )
+        return object_inputs
+
+    def _render_single_object_input(
+        self, streamlit_app: st, key: str, property: Dict
+    ) -> Any:
+        # Add title and subheader
+        title = property.get("title")
+        streamlit_app.subheader(title)
+        if property.get("description"):
+            streamlit_app.markdown(property.get("description"))
+
+        object_reference = schema_utils.get_single_reference_item(
+            property, self._schema_references
+        )
+        return self._render_object_input(streamlit_app, key, object_reference)
+
+    def _render_property_list_input(
+        self, streamlit_app: st, key: str, property: Dict
+    ) -> Any:
+
+        # Add title and subheader
+        streamlit_app.subheader(property.get("title"))
+        if property.get("description"):
+            streamlit_app.markdown(property.get("description"))
+
+        streamlit_app.markdown("---")
+
+        current_list = self._get_value(key)
+        if not current_list:
+            current_list = []
+
+        value_kwargs = {"label": "Value", "key": key + "-new-value"}
+        if property["items"]["type"] == "integer":
+            value_kwargs["value"] = 0  # type: ignore
+            new_value = streamlit_app.number_input(**value_kwargs)
+        elif property["items"]["type"] == "number":
+            value_kwargs["value"] = 0.0  # type: ignore
+            value_kwargs["format"] = "%f"
+            new_value = streamlit_app.number_input(**value_kwargs)
+        else:
+            value_kwargs["value"] = ""
+            new_value = streamlit_app.text_input(**value_kwargs)
+
+        streamlit_app.markdown("---")
+
+        with streamlit_app.container():
+            clear_col, add_col = streamlit_app.columns([1, 2])
+
+            with clear_col:
+                if streamlit_app.button("Clear Items", key=key + "-clear-items"):
+                    current_list = []
+
+            with add_col:
+                if (
+                    streamlit_app.button("Add Item", key=key + "-add-item")
+                    and new_value is not None
+                ):
+                    current_list.append(new_value)
+
+        streamlit_app.write(current_list)
+
+        return current_list
+
+    def _render_object_list_input(
+        self, streamlit_app: st, key: str, property: Dict
+    ) -> Any:
+
+        # TODO: support max_items, and min_items properties
+
+        # Add title and subheader
+        streamlit_app.subheader(property.get("title"))
+        if property.get("description"):
+            streamlit_app.markdown(property.get("description"))
+
+        streamlit_app.markdown("---")
+
+        current_list = self._get_value(key)
+        if not current_list:
+            current_list = []
+
+        object_reference = schema_utils.resolve_reference(
+            property["items"]["$ref"], self._schema_references
+        )
+        input_data = self._render_object_input(streamlit_app, key, object_reference)
+
+        streamlit_app.markdown("---")
+
+        with streamlit_app.container():
+            clear_col, add_col = streamlit_app.columns([1, 2])
+
+            with clear_col:
+                if streamlit_app.button("Clear Items", key=key + "-clear-items"):
+                    current_list = []
+
+            with add_col:
+                if (
+                    streamlit_app.button("Add Item", key=key + "-add-item")
+                    and input_data
+                ):
+                    current_list.append(input_data)
+
+        streamlit_app.write(current_list)
+        return current_list
+
+    def _render_property(self, streamlit_app: st, key: str, property: Dict) -> Any:
+        if schema_utils.is_single_enum_property(property, self._schema_references):
+            return self._render_single_enum_input(streamlit_app, key, property)
+
+        if schema_utils.is_multi_enum_property(property, self._schema_references):
+            return self._render_multi_enum_input(streamlit_app, key, property)
+
+        if schema_utils.is_single_file_property(property):
+            return self._render_single_file_input(streamlit_app, key, property)
+
+        if schema_utils.is_multi_file_property(property):
+            return self._render_multi_file_input(streamlit_app, key, property)
+
+        if schema_utils.is_single_datetime_property(property):
+            return self._render_single_datetime_input(streamlit_app, key, property)
+
+        if schema_utils.is_single_boolean_property(property):
+            return self._render_single_boolean_input(streamlit_app, key, property)
+
+        if schema_utils.is_single_dict_property(property):
+            return self._render_single_dict_input(streamlit_app, key, property)
+
+        if schema_utils.is_single_number_property(property):
+            return self._render_single_number_input(streamlit_app, key, property)
+
+        if schema_utils.is_single_string_property(property):
+            return self._render_single_string_input(streamlit_app, key, property)
+
+        if schema_utils.is_single_object(property, self._schema_references):
+            return self._render_single_object_input(streamlit_app, key, property)
+
+        if schema_utils.is_object_list_property(property, self._schema_references):
+            return self._render_object_list_input(streamlit_app, key, property)
+
+        if schema_utils.is_property_list(property):
+            return self._render_property_list_input(streamlit_app, key, property)
+
+        if schema_utils.is_single_reference(property):
+            return self._render_single_reference(streamlit_app, key, property)
+
+        streamlit_app.warning(
+            "The type of the following property is currently not supported: "
+            + str(property.get("title"))
+        )
+        raise Exception("Unsupported property")
+
+
+class OutputUI:
+    def __init__(self, output_data: Any, input_data: Any):
+        self._output_data = output_data
+        self._input_data = input_data
+
+    def render_ui(self, streamlit_app) -> None:
+        try:
+            if isinstance(self._output_data, BaseModel):
+                self._render_single_output(streamlit_app, self._output_data)
+                return
+            if type(self._output_data) == list:
+                self._render_list_output(streamlit_app, self._output_data)
+                return
+        except Exception as ex:
+            streamlit_app.exception(ex)
+            # Fallback to
+            streamlit_app.json(jsonable_encoder(self._output_data))
+
+    def _render_single_text_property(
+        self, streamlit: st, property_schema: Dict, value: Any
+    ) -> None:
+        # Add title and subheader
+        streamlit.subheader(property_schema.get("title"))
+        if property_schema.get("description"):
+            streamlit.markdown(property_schema.get("description"))
+        if value is None or value == "":
+            streamlit.info("No value returned!")
+        else:
+            streamlit.code(str(value), language="plain")
+
+    def _render_single_file_property(
+        self, streamlit: st, property_schema: Dict, value: Any
+    ) -> None:
+        # Add title and subheader
+        streamlit.subheader(property_schema.get("title"))
+        if property_schema.get("description"):
+            streamlit.markdown(property_schema.get("description"))
+        if value is None or value == "":
+            streamlit.info("No value returned!")
+        else:
+            # TODO: Detect if it is a FileContent instance
+            # TODO: detect if it is base64
+            file_extension = ""
+            if "mime_type" in property_schema:
+                mime_type = property_schema["mime_type"]
+                file_extension = mimetypes.guess_extension(mime_type) or ""
+
+                if is_compatible_audio(mime_type):
+                    streamlit.audio(value.as_bytes(), format=mime_type)
+                    return
+
+                if is_compatible_image(mime_type):
+                    streamlit.image(value.as_bytes())
+                    return
+
+                if is_compatible_video(mime_type):
+                    streamlit.video(value.as_bytes(), format=mime_type)
+                    return
+
+            filename = (
+                (property_schema["title"] + file_extension)
+                .lower()
+                .strip()
+                .replace(" ", "-")
+            )
+            streamlit.markdown(
+                f'<a href="data:application/octet-stream;base64,{value}" download="{filename}"><input type="button" value="Download File"></a>',
+                unsafe_allow_html=True,
+            )
+
+    def _render_single_complex_property(
+        self, streamlit: st, property_schema: Dict, value: Any
+    ) -> None:
+        # Add title and subheader
+        streamlit.subheader(property_schema.get("title"))
+        if property_schema.get("description"):
+            streamlit.markdown(property_schema.get("description"))
+
+        streamlit.json(jsonable_encoder(value))
+
+    def _render_single_output(self, streamlit: st, output_data: BaseModel) -> None:
+        try:
+            if has_output_ui_renderer(output_data):
+                if function_has_named_arg(output_data.render_output_ui, "input"):  # type: ignore
+                    # render method also requests the input data
+                    output_data.render_output_ui(streamlit, input=self._input_data)  # type: ignore
+                else:
+                    output_data.render_output_ui(streamlit)  # type: ignore
+                return
+        except Exception:
+            # Use default auto-generation methods if the custom rendering throws an exception
+            logger.exception(
+                "Failed to execute custom render_output_ui function. Using auto-generation instead"
+            )
+
+        model_schema = output_data.schema(by_alias=False)
+        model_properties = model_schema.get("properties")
+        definitions = model_schema.get("definitions")
+
+        if model_properties:
+            for property_key in output_data.__dict__:
+                property_schema = model_properties.get(property_key)
+                if not property_schema.get("title"):
+                    # Set property key as fallback title
+                    property_schema["title"] = property_key
+
+                output_property_value = output_data.__dict__[property_key]
+
+                if has_output_ui_renderer(output_property_value):
+                    output_property_value.render_output_ui(streamlit)  # type: ignore
+                    continue
+
+                if isinstance(output_property_value, BaseModel):
+                    # Render output recursivly
+                    streamlit.subheader(property_schema.get("title"))
+                    if property_schema.get("description"):
+                        streamlit.markdown(property_schema.get("description"))
+                    self._render_single_output(streamlit, output_property_value)
+                    continue
+
+                if property_schema:
+                    if schema_utils.is_single_file_property(property_schema):
+                        self._render_single_file_property(
+                            streamlit, property_schema, output_property_value
+                        )
+                        continue
+
+                    if (
+                        schema_utils.is_single_string_property(property_schema)
+                        or schema_utils.is_single_number_property(property_schema)
+                        or schema_utils.is_single_datetime_property(property_schema)
+                        or schema_utils.is_single_boolean_property(property_schema)
+                    ):
+                        self._render_single_text_property(
+                            streamlit, property_schema, output_property_value
+                        )
+                        continue
+                    if definitions and schema_utils.is_single_enum_property(
+                        property_schema, definitions
+                    ):
+                        self._render_single_text_property(
+                            streamlit, property_schema, output_property_value.value
+                        )
+                        continue
+
+                    # TODO: render dict as table
+
+                    self._render_single_complex_property(
+                        streamlit, property_schema, output_property_value
+                    )
+            return
+
+    def _render_list_output(self, streamlit: st, output_data: List) -> None:
+        try:
+            data_items: List = []
+            for data_item in output_data:
+                if has_output_ui_renderer(data_item):
+                    # Render using the render function
+                    data_item.render_output_ui(streamlit)  # type: ignore
+                    continue
+                data_items.append(data_item.dict())
+            # Try to show as dataframe
+            streamlit.table(pd.DataFrame(data_items))
+        except Exception:
+            # Fallback to
+            streamlit.json(jsonable_encoder(output_data))
+
+
+def getOpyrator(mode: str) -> Opyrator:
+    if mode == None or mode.startswith('VC'):
+        from control.mkgui.app_vc import convert
+        return  Opyrator(convert)
+    if mode == None or mode.startswith('预处理'):
+        from control.mkgui.preprocess import preprocess
+        return  Opyrator(preprocess)
+    if mode == None or mode.startswith('模型训练'):
+        from control.mkgui.train import train
+        return  Opyrator(train)
+    if mode == None or mode.startswith('模型训练(VC)'):
+        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
+    session_state.input_data = {}
+    # Add custom css settings
+    st.markdown(f"<style>{CUSTOM_STREAMLIT_CSS}</style>", unsafe_allow_html=True)
+
+    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)")
+        )
+        if "mode" in session_state:
+            mode = session_state.mode
+        else:
+            mode = ""
+        opyrator = getOpyrator(mode)
+    title = opyrator.name + mode
+
+    col1, col2, _ = st.columns(3)
+    col2.title(title)
+    col2.markdown("欢迎使用MockingBird Web 2")
+
+    image = Image.open(path.join('control','mkgui', 'static', 'mb.png'))
+    col1.image(image)
+
+    st.markdown("---")
+    left, right = st.columns([0.4, 0.6])
+
+    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:
+            with st.spinner("Executing operation. Please wait..."):
+                try:
+                    input_data_obj = parse_obj_as(
+                        opyrator.input_type, session_state.input_data
+                    )
+                    session_state.output_data = opyrator(input=input_data_obj)
+                    session_state.latest_operation_input = input_data_obj  # should this really be saved as additional session object?
+                except ValidationError as ex:
+                    st.error(ex)
+                else:
+                    # st.success("Operation executed successfully.")
+                    pass
+
+    with right:
+        st.header("Result 结果")
+        if 'output_data' in session_state:
+            OutputUI(
+                session_state.output_data, session_state.latest_operation_input
+            ).render_ui(st)
+            if st.button("Clear"):
+            # Clear all state
+                for key in st.session_state.keys():
+                    del st.session_state[key]   
+                session_state.input_data = {}
+                st.experimental_rerun()
+        else:
+            # placeholder
+            st.caption("请使用左侧控制板进行输入并运行获得结果")
+        
+   

control/mkgui/base/ui/streamlit_utils.py

@@ -0,0 +1,13 @@
+CUSTOM_STREAMLIT_CSS = """
+div[data-testid="stBlock"] button {
+  width: 100% !important;
+  margin-bottom: 20px !important;
+  border-color: #bfbfbf !important;
+}
+section[data-testid="stSidebar"] div {
+  max-width: 10rem;
+}
+pre code {
+    white-space: pre-wrap;
+}
+"""

control/mkgui/preprocess.py

@@ -0,0 +1,96 @@
+from pydantic import BaseModel, Field
+import os
+from pathlib import Path
+from enum import Enum
+from typing import Any, Tuple
+
+
+# Constants
+EXT_MODELS_DIRT = f"data{os.sep}ckpt{os.sep}ppg_extractor"
+ENC_MODELS_DIRT = f"data{os.sep}ckpt{os.sep}encoder"
+
+
+if os.path.isdir(EXT_MODELS_DIRT):    
+    extractors =  Enum('extractors', list((file.name, file) for file in Path(EXT_MODELS_DIRT).glob("**/*.pt")))
+    print("Loaded extractor models: " + str(len(extractors)))
+else:
+    raise Exception(f"Model folder {EXT_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):
+    VC_PPG2MEL = "ppg2mel"
+
+class Dataset(str, Enum):
+    AIDATATANG_200ZH = "aidatatang_200zh"
+    AIDATATANG_200ZH_S = "aidatatang_200zh_s"
+
+class Input(BaseModel):
+    # def render_input_ui(st, input) -> Dict: 
+    #     input["selected_dataset"] = st.selectbox(
+    #         '选择数据集', 
+    #         ("aidatatang_200zh", "aidatatang_200zh_s")
+    #     )
+    # return input
+    model: Model = Field(
+        Model.VC_PPG2MEL, title="目标模型",
+    )
+    dataset: Dataset = Field(
+        Dataset.AIDATATANG_200ZH, title="数据集选择",
+    )
+    datasets_root: str = Field(
+        ..., alias="数据集根目录", description="输入数据集根目录（相对/绝对）",
+        format=True,
+        example="..\\trainning_data\\"
+    )
+    output_root: str = Field(
+        ..., alias="输出根目录", description="输出结果根目录（相对/绝对）",
+        format=True,
+        example="..\\trainning_data\\"
+    )
+    n_processes: int = Field(   
+        2, alias="处理线程数", description="根据CPU线程数来设置",
+        le=32, ge=1
+    )
+    extractor: extractors = Field(
+        ..., alias="特征提取模型", 
+        description="选择PPG特征提取模型文件."
+    )
+    encoder: encoders = Field(
+        ..., alias="语音编码模型", 
+        description="选择语音编码模型文件."
+    )
+
+class AudioEntity(BaseModel):
+    content: bytes
+    mel: Any
+
+class Output(BaseModel):
+    __root__: Tuple[str, int]
+
+    def render_output_ui(self, streamlit_app, input) -> None:  # type: ignore
+        """Custom output UI.
+        If this method is implmeneted, it will be used instead of the default Output UI renderer.
+        """
+        sr, count = self.__root__
+        streamlit_app.subheader(f"Dataset {sr} done processed total of {count}")
+
+def preprocess(input: Input) -> Output:
+    """Preprocess(预处理)"""
+    finished = 0
+    if input.model == Model.VC_PPG2MEL:
+        from models.ppg2mel.preprocess import preprocess_dataset
+        finished = preprocess_dataset(
+            datasets_root=Path(input.datasets_root),
+            dataset=input.dataset,
+            out_dir=Path(input.output_root),
+            n_processes=input.n_processes,
+            ppg_encoder_model_fpath=Path(input.extractor.value),
+            speaker_encoder_model=Path(input.encoder.value)
+        )
+    # TODO: pass useful return code
+    return Output(__root__=(input.dataset, finished))

control/mkgui/train.py

@@ -0,0 +1,106 @@
+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)

control/mkgui/train_vc.py

@@ -0,0 +1,155 @@
+from pydantic import BaseModel, Field
+import os
+from pathlib import Path
+from enum import Enum
+from typing import Any, Tuple
+import numpy as np
+from utils.hparams import HpsYaml
+from utils.util import AttrDict
+import torch
+
+# Constants
+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"
+
+
+if os.path.isdir(EXT_MODELS_DIRT):    
+    extractors =  Enum('extractors', list((file.name, file) for file in Path(EXT_MODELS_DIRT).glob("**/*.pt")))
+    print("Loaded extractor models: " + str(len(extractors)))
+else:
+    raise Exception(f"Model folder {EXT_MODELS_DIRT} doesn't exist.")
+
+if os.path.isdir(CONV_MODELS_DIRT):    
+    convertors =  Enum('convertors', list((file.name, file) for file in Path(CONV_MODELS_DIRT).glob("**/*.pth")))
+    print("Loaded convertor models: " + str(len(convertors)))
+else:
+    raise Exception(f"Model folder {CONV_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):
+    VC_PPG2MEL = "ppg2mel"
+
+class Dataset(str, Enum):
+    AIDATATANG_200ZH = "aidatatang_200zh"
+    AIDATATANG_200ZH_S = "aidatatang_200zh_s"
+
+class Input(BaseModel):
+    # def render_input_ui(st, input) -> Dict: 
+    #     input["selected_dataset"] = st.selectbox(
+    #         '选择数据集', 
+    #         ("aidatatang_200zh", "aidatatang_200zh_s")
+    #     )
+    # return input
+    model: Model = Field(
+        Model.VC_PPG2MEL, 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\\"
+    # )
+    output_root: str = Field(
+        ..., alias="输出目录(可选)", description="建议不填，保持默认",
+        format=True,
+        example=""
+    )
+    continue_mode: bool = Field(
+        True, alias="继续训练模式", description="选择“是”，则从下面选择的模型中继续训练",
+    )
+    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",
+    )
+    # TODO: Move to hiden fields by default
+    convertor: convertors = Field(
+        ..., alias="转换模型", 
+        description="选择语音转换模型文件."
+    )
+    extractor: extractors = Field(
+        ..., alias="特征提取模型", 
+        description="选择PPG特征提取模型文件."
+    )
+    encoder: encoders = Field(
+        ..., alias="语音编码模型", 
+        description="选择语音编码模型文件."
+    )
+    njobs: int = Field(
+        8, alias="进程数", description="适用于ppg2mel",
+    )
+    seed: int = Field(
+        default=0, alias="初始随机数", description="适用于ppg2mel",
+    )
+    model_name: str = Field(
+        ..., alias="新模型名", description="仅在重新训练时生效,选中继续训练时无效",
+        example="test"
+    )
+    model_config: str = Field(
+        ..., alias="新模型配置", description="仅在重新训练时生效,选中继续训练时无效",
+        example=".\\ppg2mel\\saved_models\\seq2seq_mol_ppg2mel_vctk_libri_oneshotvc_r4_normMel_v2"
+    )
+
+class AudioEntity(BaseModel):
+    content: bytes
+    mel: Any
+
+class Output(BaseModel):
+    __root__: Tuple[str, int]
+
+    def render_output_ui(self, streamlit_app, input) -> None:  # type: ignore
+        """Custom output UI.
+        If this method is implmeneted, it will be used instead of the default Output UI renderer.
+        """
+        sr, count = self.__root__
+        streamlit_app.subheader(f"Dataset {sr} done processed total of {count}")
+
+def train_vc(input: Input) -> Output:
+    """Train VC(训练 VC)"""
+
+    print(">>> OneShot VC training ...")
+    params = AttrDict()
+    params.update({
+        "gpu": input.gpu,
+        "cpu": not input.gpu,
+        "njobs": input.njobs,
+        "seed": input.seed,
+        "verbose": input.verbose,
+        "load": input.convertor.value,
+        "warm_start": False,
+    })
+    if input.continue_mode: 
+        # trace old model and config
+        p = Path(input.convertor.value)
+        params.name = p.parent.name
+        # search a config file
+        model_config_fpaths = list(p.parent.rglob("*.yaml"))
+        if len(model_config_fpaths) == 0:
+            raise "No model yaml config found for convertor"
+        config = HpsYaml(model_config_fpaths[0])
+        params.ckpdir = p.parent.parent
+        params.config = model_config_fpaths[0]
+        params.logdir = os.path.join(p.parent, "log")
+    else:
+        # Make the config dict dot visitable
+        config = HpsYaml(input.config)    
+    np.random.seed(input.seed)
+    torch.manual_seed(input.seed)
+    if torch.cuda.is_available():
+        torch.cuda.manual_seed_all(input.seed)
+    mode = "train"
+    from models.ppg2mel.train.train_linglf02mel_seq2seq_oneshotvc import Solver
+    solver = Solver(config, params, mode)
+    solver.load_data()
+    solver.set_model()
+    solver.exec()
+    print(">>> Oneshot VC train finished!")
+
+    # TODO: pass useful return code
+    return Output(__root__=(input.dataset, 0))

control/toolbox/__init__.py

@@ -1,11 +1,12 @@
-from toolbox.ui import UI
-from encoder import inference as encoder
-from synthesizer.inference import Synthesizer
-from vocoder.wavernn import inference as rnn_vocoder
-from vocoder.hifigan import inference as gan_vocoder
+from control.toolbox.ui import UI
+from models.encoder import inference as encoder
+from models.synthesizer.inference import Synthesizer
+from models.vocoder.wavernn import inference as rnn_vocoder
+from models.vocoder.hifigan import inference as gan_vocoder
+from models.vocoder.fregan import inference as fgan_vocoder
 from pathlib import Path
 from time import perf_counter as timer
-from toolbox.utterance import Utterance
+from control.toolbox.utterance import Utterance
 import numpy as np
 import traceback
 import sys
@@ -37,8 +38,8 @@ recognized_datasets = [
     "VoxCeleb2/dev/aac",
     "VoxCeleb2/test/aac",
     "VCTK-Corpus/wav48",
-    "aidatatang_200zh/corpus/dev",
     "aidatatang_200zh/corpus/test",
+    "aidatatang_200zh/corpus/train",
     "aishell3/test/wav",
     "magicdata/train",
 ]
@@ -396,7 +397,7 @@ class Toolbox:
         self.ui.log("Loading the extractor %s... " % model_fpath)
         self.ui.set_loading(1)
         start = timer()
-        import ppg_extractor as extractor
+        import models.ppg_extractor as extractor
         self.extractor = extractor.load_model(model_fpath)
         self.ui.log("Done (%dms)." % int(1000 * (timer() - start)), "append")
         self.ui.set_loading(0)
@@ -405,16 +406,11 @@ class Toolbox:
         if self.ui.current_convertor_fpath is None:
             return
         model_fpath = self.ui.current_convertor_fpath
-        # search a config file
-        model_config_fpaths = list(model_fpath.parent.rglob("*.yaml"))
-        if self.ui.current_convertor_fpath is None:
-            return
-        model_config_fpath = model_config_fpaths[0]
         self.ui.log("Loading the convertor %s... " % model_fpath)
         self.ui.set_loading(1)
         start = timer()
-        import ppg2mel as convertor
-        self.convertor = convertor.load_model(model_config_fpath, model_fpath)
+        import models.ppg2mel as convertor
+        self.convertor = convertor.load_model( model_fpath)
         self.ui.log("Done (%dms)." % int(1000 * (timer() - start)), "append")
         self.ui.set_loading(0)
         
@@ -447,7 +443,7 @@ class Toolbox:
             return 
         # Sekect vocoder based on model name
         model_config_fpath = None
-        if model_fpath.name[0] == "g":
+        if model_fpath.name is not None and model_fpath.name.find("hifigan") > -1:
             vocoder = gan_vocoder
             self.ui.log("set hifigan as vocoder")
             # search a config file
@@ -456,6 +452,15 @@ class Toolbox:
                 return
             if len(model_config_fpaths) > 0:
                 model_config_fpath = model_config_fpaths[0]
+        elif model_fpath.name is not None and model_fpath.name.find("fregan") > -1:
+            vocoder = fgan_vocoder
+            self.ui.log("set fregan as vocoder")
+            # search a config file
+            model_config_fpaths = list(model_fpath.parent.rglob("*.json"))
+            if self.vc_mode and self.ui.current_extractor_fpath is None:
+                return
+            if len(model_config_fpaths) > 0:
+                model_config_fpath = model_config_fpaths[0]
         else:
             vocoder = rnn_vocoder
             self.ui.log("set wavernn as vocoder")

control/toolbox/ui.py

@@ -3,9 +3,8 @@ from PyQt5 import QtGui
 from PyQt5.QtWidgets import *
 import matplotlib.pyplot as plt
 from matplotlib.backends.backend_qt5agg import FigureCanvasQTAgg as FigureCanvas
-from matplotlib.figure import Figure
-from encoder.inference import plot_embedding_as_heatmap
-from toolbox.utterance import Utterance
+from models.encoder.inference import plot_embedding_as_heatmap
+from control.toolbox.utterance import Utterance
 from pathlib import Path
 from typing import List, Set
 import sounddevice as sd
@@ -274,7 +273,9 @@ class UI(QDialog):
             if datasets_root is None or len(datasets) == 0:
                 msg = "Warning: you d" + ("id not pass a root directory for datasets as argument" \
                     if datasets_root is None else "o not have any of the recognized datasets" \
-                                                  " in %s" % datasets_root) 
+                                                  " in %s \n" \
+                                                  "Please note use 'E:\datasets' as root path " \
+                                                  "instead of 'E:\datasets\aidatatang_200zh\corpus\test' as an example "   % datasets_root) 
                 self.log(msg)
                 msg += ".\nThe recognized datasets are:\n\t%s\nFeel free to add your own. You " \
                        "can still use the toolbox by recording samples yourself." % \

datasets_download/CN.txt

@@ -0,0 +1,8 @@
+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

@@ -0,0 +1,8 @@
+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

@@ -0,0 +1,8 @@
+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

@@ -0,0 +1,4 @@
+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

@@ -0,0 +1,8 @@
+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

@@ -0,0 +1,8 @@
+#!/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

@@ -0,0 +1,29 @@
+#!/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 toolbox import Toolbox
+from control.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="encoder/saved_models", 
+    parser.add_argument("-e", "--enc_models_dir", type=Path, default=f"data{os.sep}ckpt{os.sep}encoder", 
                         help="Directory containing saved encoder models")
-    parser.add_argument("-s", "--syn_models_dir", type=Path, default="synthesizer/saved_models", 
+    parser.add_argument("-s", "--syn_models_dir", type=Path, default=f"data{os.sep}ckpt{os.sep}synthesizer", 
                         help="Directory containing saved synthesizer models")
-    parser.add_argument("-v", "--voc_models_dir", type=Path, default="vocoder/saved_models", 
+    parser.add_argument("-v", "--voc_models_dir", type=Path, default=f"data{os.sep}ckpt{os.sep}vocoder", 
                         help="Directory containing saved vocoder models")
-    parser.add_argument("-ex", "--extractor_models_dir", type=Path, default="ppg_extractor/saved_models", 
+    parser.add_argument("-ex", "--extractor_models_dir", type=Path, default=f"data{os.sep}ckpt{os.sep}ppg_extractor", 
                         help="Directory containing saved extrator models")
-    parser.add_argument("-cv", "--convertor_models_dir", type=Path, default="ppg2mel/saved_models", 
+    parser.add_argument("-cv", "--convertor_models_dir", type=Path, default=f"data{os.sep}ckpt{os.sep}ppg2mel", 
                         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

@@ -0,0 +1,23 @@
+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

@@ -0,0 +1,17 @@
+#!/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/data_objects/__init__.py

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

gen_voice.py

@@ -0,0 +1,120 @@
+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)
+
+

models/encoder/audio.py

@@ -1,5 +1,5 @@
 from scipy.ndimage.morphology import binary_dilation
-from encoder.params_data import *
+from models.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(
-        wav,
-        sampling_rate,
+        y=wav,
+        sr=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

models/encoder/data_objects/__init__.py

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

models/encoder/data_objects/speaker.py

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

models/encoder/data_objects/speaker_batch.py

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

models/encoder/data_objects/speaker_verification_dataset.py

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

models/encoder/inference.py

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

models/encoder/model.py

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

models/encoder/preprocess.py

@@ -1,8 +1,8 @@
 from multiprocess.pool import ThreadPool
-from encoder.params_data import *
-from encoder.config import librispeech_datasets, anglophone_nationalites
+from models.encoder.params_data import *
+from models.encoder.config import librispeech_datasets, anglophone_nationalites
 from datetime import datetime
-from encoder import audio
+from models.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 encoder import params_data
+        from models.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)

models/encoder/train.py

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

models/encoder/visualizations.py

@@ -1,4 +1,4 @@
-from encoder.data_objects.speaker_verification_dataset import SpeakerVerificationDataset
+from models.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 encoder import params_data
-        from encoder import params_model
+        from models.encoder import params_data
+        from models.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)

models/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.load_yaml import HpsYaml
+from utils.hparams import HpsYaml
 
 class MelDecoderMOLv2(AbsMelDecoder):
     """Use an encoder to preprocess ppg."""
@@ -191,12 +191,15 @@ class MelDecoderMOLv2(AbsMelDecoder):
         
         return mel_outputs[0], mel_outputs_postnet[0], alignments[0]
 
-def load_model(train_config, model_file, device=None):
-    
+def load_model(model_file, device=None):
+    # search a config file
+    model_config_fpaths = list(model_file.parent.rglob("*.yaml"))
+    if len(model_config_fpaths) == 0:
+        raise "No model yaml config found for convertor"
     if device is None:
         device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 
-    model_config = HpsYaml(train_config)
+    model_config = HpsYaml(model_config_fpaths[0])
     ppg2mel_model = MelDecoderMOLv2(
         **model_config["model"]
     ).to(device)

models/ppg2mel/preprocess.py

@@ -7,10 +7,10 @@ from pathlib import Path
 import soundfile
 import resampy
 
-from ppg_extractor import load_model
+from models.ppg_extractor import load_model
 import encoder.inference as Encoder
-from encoder.audio import preprocess_wav
-from encoder import audio
+from models.encoder.audio import preprocess_wav
+from models.encoder import audio
 from utils.f0_utils import compute_f0
 
 from torch.multiprocessing import Pool, cpu_count
@@ -110,3 +110,4 @@ def preprocess_dataset(datasets_root, dataset, out_dir, n_processes, ppg_encoder
     t_fid_file.close()
     d_fid_file.close()
     e_fid_file.close()
+    return len(wav_file_list)

models/ppg2mel/train.py

@@ -0,0 +1,62 @@
+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='ckpt/', 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('--no-msg', action='store_true', help='Hide all messages.')
+    
+    ###
+
+    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()   

models/ppg2mel/train/solver.py

@@ -8,7 +8,6 @@ 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():
@@ -93,6 +92,7 @@ class BaseSolver():
 
     def load_ckpt(self):
         ''' Load ckpt if --load option is specified '''
+        print(self.paras)
         if self.paras.load is not None:
             if self.paras.warm_start:
                 self.verbose(f"Warm starting model from checkpoint {self.paras.load}.")
@@ -100,7 +100,7 @@ class BaseSolver():
                     self.paras.load, map_location=self.device if self.mode == 'train'
                                                         else 'cpu')
                 model_dict = ckpt['model']
-                if len(self.config.model.ignore_layers) > 0:
+                if "ignore_layers" in self.config.model and len(self.config.model.ignore_layers) > 0:
                     model_dict = {k:v for k, v in model_dict.items()
                                   if k not in self.config.model.ignore_layers}
                     dummy_dict = self.model.state_dict()

models/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 ppg2mel import MelDecoderMOLv2
+from models.ppg2mel import MelDecoderMOLv2
 
 
 class Solver(BaseSolver):