Update README.md

Update README.md (#1011 )
Update README.md (#1009 )
2026-02-03 18:43:41 +08:00 · 2025-11-13 10:03:10 +08:00 · 2024-11-14 21:00:29 -08:00 · 2024-11-01 12:57:10 +08:00 · 2024-10-22 10:21:44 +08:00 · 2024-10-16 22:48:15 +08:00
7 changed files with 279 additions and 80 deletions
--- a/README-CN.md
+++ b/README-CN.md
@@ -29,6 +29,7 @@
 > 如果在用 pip 方式安装的时候出现 `ERROR: Could not find a version that satisfies the requirement torch==1.9.0+cu102 (from versions: 0.1.2, 0.1.2.post1, 0.1.2.post2)` 这个错误可能是 python 版本过低，3.9 可以安装成功
 * 安装 [ffmpeg](https://ffmpeg.org/download.html#get-packages)。
 * 运行`pip install -r requirements.txt` 来安装剩余的必要包。
+> 这里的环境建议使用 `Repo Tag 0.0.1` `Pytorch1.9.0 with Torchvision0.10.0 and cudatoolkit10.2` `requirements.txt` `webrtcvad-wheels` 因为 `requiremants.txt` 是在几个月前导出的，所以不适配新版本
 * 安装 webrtcvad `pip install webrtcvad-wheels`。

 或者
@@ -113,7 +114,7 @@
 > 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：
-`python synthesizer_train.py mandarin <datasets_root>/SV2TTS/synthesizer`
+`python ./control/cli/synthesizer_train.py mandarin <datasets_root>/SV2TTS/synthesizer`

 * 当您在训练文件夹 *synthesizer/saved_models/* 中看到注意线显示和损失满足您的需要时，请转到`启动程序`一步。

@@ -124,7 +125,7 @@
 | --- | ----------- | ----- | ----- |
 | 作者 | https://pan.baidu.com/s/1iONvRxmkI-t1nHqxKytY3g  [百度盘链接](https://pan.baidu.com/s/1iONvRxmkI-t1nHqxKytY3g) 4j5d |  | 75k steps 用3个开源数据集混合训练
 | 作者 | https://pan.baidu.com/s/1fMh9IlgKJlL2PIiRTYDUvw  [百度盘链接](https://pan.baidu.com/s/1fMh9IlgKJlL2PIiRTYDUvw) 提取码：om7f |  | 25k steps 用3个开源数据集混合训练, 切换到tag v0.0.1使用
-|@FawenYo | https://drive.google.com/file/d/1H-YGOUHpmqKxJ9FRc6vAjPuqQki24UbC/view?usp=sharing [百度盘链接](https://pan.baidu.com/s/1vSYXO4wsLyjnF3Unl-Xoxg) 提取码：1024  | [input](https://github.com/babysor/MockingBird/wiki/audio/self_test.mp3) [output](https://github.com/babysor/MockingBird/wiki/audio/export.wav) | 200k steps 台湾口音需切换到tag v0.0.1使用
+|@FawenYo | https://yisiou-my.sharepoint.com/:u:/g/personal/lawrence_cheng_fawenyo_onmicrosoft_com/EWFWDHzee-NNg9TWdKckCc4BC7bK2j9cCbOWn0-_tK0nOg?e=n0gGgC  | [input](https://github.com/babysor/MockingBird/wiki/audio/self_test.mp3) [output](https://github.com/babysor/MockingBird/wiki/audio/export.wav) | 200k steps 台湾口音需切换到tag v0.0.1使用
 |@miven| https://pan.baidu.com/s/1PI-hM3sn5wbeChRryX-RCQ 提取码：2021 | https://www.bilibili.com/video/BV1uh411B7AD/ | 150k steps 注意：根据[issue](https://github.com/babysor/MockingBird/issues/37)修复 并切换到tag v0.0.1使用

 #### 2.4训练声码器 (可选)
@@ -135,14 +136,14 @@


 * 训练wavernn声码器:
-`python vocoder_train.py <trainid> <datasets_root>`
+`python ./control/cli/vocoder_train.py <trainid> <datasets_root>`
 > `<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hifigan声码器:
-`python vocoder_train.py <trainid> <datasets_root> hifigan`
+`python ./control/cli/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`
+`python ./control/cli/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
@@ -173,14 +174,14 @@

 * 下载aidatatang_200zh数据集并解压：确保您可以访问 *train* 文件夹中的所有音频文件（如.wav）
 * 进行音频和梅尔频谱图预处理：
-`python pre4ppg.py <datasets_root> -d {dataset} -n {number}`
+`python ./control/cli/pre4ppg.py <datasets_root> -d {dataset} -n {number}`
 可传入参数：
 * `-d {dataset}` 指定数据集，支持 aidatatang_200zh, 不传默认为aidatatang_200zh
 * `-n {number}` 指定并行数，CPU 11700k在8的情况下，需要运行12到18小时！待优化
 > 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：
-`python ppg2mel_train.py --config .\ppg2mel\saved_models\ppg2mel.yaml --oneshotvc `
+`python ./control/cli/ppg2mel_train.py --config .\ppg2mel\saved_models\ppg2mel.yaml --oneshotvc `
 * 如果想要继续上一次的训练，可以通过`--load .\ppg2mel\saved_models\<old_pt_file>` 参数指定一个预训练模型文件。

 #### 4.2 启动工具箱VC模式
--- a/README-LINUX-CN.md
+++ b/README-LINUX-CN.md
@@ -0,0 +1,223 @@
+## 实时语音克隆 - 中文/普通话
+![mockingbird](https://user-images.githubusercontent.com/12797292/131216767-6eb251d6-14fc-4951-8324-2722f0cd4c63.jpg)
+
+[![MIT License](https://img.shields.io/badge/license-MIT-blue.svg?style=flat)](http://choosealicense.com/licenses/mit/)
+
+### [English](README.md)  | 中文
+
+### [DEMO VIDEO](https://www.bilibili.com/video/BV17Q4y1B7mY/) | [Wiki教程](https://github.com/babysor/MockingBird/wiki/Quick-Start-(Newbie)) ｜ [训练教程](https://vaj2fgg8yn.feishu.cn/docs/doccn7kAbr3SJz0KM0SIDJ0Xnhd)
+
+## 特性
+🌍 **中文** 支持普通话并使用多种中文数据集进行测试：aidatatang_200zh, magicdata, aishell3, biaobei, MozillaCommonVoice, data_aishell 等
+
+🤩 **Easy & Awesome** 仅需下载或新训练合成器（synthesizer）就有良好效果，复用预训练的编码器/声码器，或实时的HiFi-GAN作为vocoder
+
+🌍 **Webserver Ready** 可伺服你的训练结果，供远程调用。
+
+🤩 **感谢各位小伙伴的支持，本项目将开启新一轮的更新**
+
+## 1.快速开始
+### 1.1 建议环境
+- Ubuntu 18.04 
+- Cuda 11.7 && CuDNN 8.5.0 
+- Python 3.8 或 3.9 
+- Pytorch 2.0.1 <post cuda-11.7>
+### 1.2 环境配置
+```shell
+# 下载前建议更换国内镜像源
+
+conda create -n sound python=3.9
+
+conda activate sound
+
+git clone https://github.com/babysor/MockingBird.git
+
+cd MockingBird
+
+pip install -r requirements.txt
+
+pip install webrtcvad-wheels
+
+pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu117
+```
+
+### 1.3 模型准备
+> 当实在没有设备或者不想慢慢调试，可以使用社区贡献的模型(欢迎持续分享):
+
+| 作者 | 下载链接 | 效果预览 | 信息 |
+| --- | ----------- | ----- | ----- |
+| 作者 | https://pan.baidu.com/s/1iONvRxmkI-t1nHqxKytY3g  [百度盘链接](https://pan.baidu.com/s/1iONvRxmkI-t1nHqxKytY3g) 4j5d |  | 75k steps 用3个开源数据集混合训练
+| 作者 | https://pan.baidu.com/s/1fMh9IlgKJlL2PIiRTYDUvw  [百度盘链接](https://pan.baidu.com/s/1fMh9IlgKJlL2PIiRTYDUvw) 提取码：om7f |  | 25k steps 用3个开源数据集混合训练, 切换到tag v0.0.1使用
+|@FawenYo | https://drive.google.com/file/d/1H-YGOUHpmqKxJ9FRc6vAjPuqQki24UbC/view?usp=sharing [百度盘链接](https://pan.baidu.com/s/1vSYXO4wsLyjnF3Unl-Xoxg) 提取码：1024  | [input](https://github.com/babysor/MockingBird/wiki/audio/self_test.mp3) [output](https://github.com/babysor/MockingBird/wiki/audio/export.wav) | 200k steps 台湾口音需切换到tag v0.0.1使用
+|@miven| https://pan.baidu.com/s/1PI-hM3sn5wbeChRryX-RCQ 提取码：2021 | https://www.bilibili.com/video/BV1uh411B7AD/ | 150k steps 注意：根据[issue](https://github.com/babysor/MockingBird/issues/37)修复 并切换到tag v0.0.1使用
+
+### 1.4 文件结构准备
+文件结构准备如下所示，算法将自动遍历synthesizer下的.pt模型文件。
+``` 
+#  以第一个 pretrained-11-7-21_75k.pt 为例
+
+└── data
+      └── ckpt 
+            └── synthesizer 
+                     └── pretrained-11-7-21_75k.pt
+```
+### 1.5 运行
+```
+python web.py 
+```
+
+## 2.模型训练
+### 2.1 数据准备
+#### 2.1.1 数据下载
+``` shell
+# aidatatang_200zh 
+ 
+wget https://openslr.elda.org/resources/62/aidatatang_200zh.tgz
+```
+``` shell
+# MAGICDATA  
+
+wget https://openslr.magicdatatech.com/resources/68/train_set.tar.gz
+
+wget https://openslr.magicdatatech.com/resources/68/dev_set.tar.gz
+
+wget https://openslr.magicdatatech.com/resources/68/test_set.tar.gz
+```
+``` shell
+# AISHELL-3 
+
+wget https://openslr.elda.org/resources/93/data_aishell3.tgz
+```
+```shell
+# Aishell  
+
+wget https://openslr.elda.org/resources/33/data_aishell.tgz
+```
+#### 2.1.2 数据批量解压
+```shell
+# 该指令为解压当前目录下的所有压缩文件 
+
+for gz in *.gz; do tar -zxvf $gz; done
+```
+### 2.2 encoder模型训练
+#### 2.2.1 数据预处理：
+需要先在`pre.py `头部加入：
+```python
+import torch
+torch.multiprocessing.set_start_method('spawn', force=True)
+```
+使用以下指令对数据预处理：
+```shell
+python pre.py <datasets_root> \
+           -d <datasets_name> 
+```
+其中`<datasets_root>`为原数据集路径，`<datasets_name>` 为数据集名称。
+
+支持 `librispeech_other`，`voxceleb1`，`aidatatang_200zh`，使用逗号分割处理多数据集。
+
+### 2.2.2 encoder模型训练：
+超参数文件路径：`models/encoder/hparams.py`
+```shell
+python encoder_train.py <name> \
+                        <datasets_root>/SV2TTS/encoder
+```
+其中 `<name>` 是训练产生文件的名称，可自行修改。
+
+其中 `<datasets_root>` 是经过 `Step 2.1.1` 处理过后的数据集路径。
+#### 2.2.3 开启encoder模型训练数据可视化（可选）
+```shell
+visdom
+```
+
+### 2.3 synthesizer模型训练
+#### 2.3.1 数据预处理：
+```shell
+python pre.py    <datasets_root> \
+              -d <datasets_name> \
+              -o <datasets_path> \
+              -n <number>
+```
+`<datasets_root>` 为原数据集路径，当你的`aidatatang_200zh`路径为`/data/aidatatang_200zh/corpus/train`时，`<datasets_root>` 为 `/data/`。
+ 
+`<datasets_name>` 为数据集名称。
+ 
+`<datasets_path>` 为数据集处理后的保存路径。
+
+`<number>` 为数据集处理时进程数，根据CPU情况调整大小。
+
+#### 2.3.2 新增数据预处理：
+```shell
+python pre.py    <datasets_root> \
+              -d <datasets_name> \
+              -o <datasets_path> \
+              -n <number> \
+              -s
+```
+当新增数据集时，应加 `-s` 选择数据拼接，不加则为覆盖。
+#### 2.3.3 synthesizer模型训练：
+超参数文件路径：`models/synthesizer/hparams.py`，需将`MockingBird/control/cli/synthesizer_train.py`移成`MockingBird/synthesizer_train.py`结构。
+```shell
+python synthesizer_train.py <name> <datasets_path> \
+                                -m <out_dir>
+```
+其中 `<name>` 是训练产生文件的名称，可自行修改。
+
+其中 `<datasets_path>` 是经过 `Step 2.2.1` 处理过后的数据集路径。
+
+其中 `<out_dir> `为训练时所有数据的保存路径。
+
+### 2.4 vocoder模型训练
+vocoder模型对生成效果影响不大，已预置3款。
+#### 2.4.1 数据预处理
+```shell
+python vocoder_preprocess.py <datasets_root> \
+                          -m <synthesizer_model_path>
+```
+
+其中`<datasets_root>`为你数据集路径。
+
+其中 `<synthesizer_model_path>`为synthesizer模型地址。
+
+#### 2.4.2 wavernn声码器训练:
+```
+python vocoder_train.py <name> <datasets_root>
+```
+#### 2.4.3 hifigan声码器训练:
+```
+python vocoder_train.py <name> <datasets_root> hifigan
+```
+#### 2.4.4 fregan声码器训练:
+```
+python vocoder_train.py <name> <datasets_root> \
+                        --config config.json fregan
+```
+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.致谢
+### 3.1 项目致谢
+该库一开始从仅支持英语的[Real-Time-Voice-Cloning](https://github.com/CorentinJ/Real-Time-Voice-Cloning) 分叉出来的，鸣谢作者。
+### 3.2 论文致谢
+| URL | Designation | 标题 | 实现源码 |
+| --- | ----------- | ----- | --------------------- |
+| [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 |
+
+### 3.3 开发者致谢
+
+作为AI领域的从业者，我们不仅乐于开发一些具有里程碑意义的算法项目，同时也乐于分享项目以及开发过程中收获的喜悦。
+
+因此，你们的使用是对我们项目的最大认可。同时当你们在项目使用中遇到一些问题时，欢迎你们随时在issue上留言。你们的指正这对于项目的后续优化具有十分重大的的意义。
+
+为了表示感谢，我们将在本项目中留下各位开发者信息以及相对应的贡献。
+
+- ------------------------------------------------  开 发 者 贡 献 内 容  ---------------------------------------------------------------------------------
+
--- a/README.md
+++ b/README.md
@@ -1,9 +1,11 @@
+> 🚧 While I no longer actively update this repo, you can find me continuously pushing this tech forward to good side and open-source. I'm also building an optimized and cloud hosted version: https://noiz.ai/ and it's free but not ready for commersial use now.
+>
 ![mockingbird](https://user-images.githubusercontent.com/12797292/131216767-6eb251d6-14fc-4951-8324-2722f0cd4c63.jpg)
-
+<a href="https://trendshift.io/repositories/3869" target="_blank"><img src="https://trendshift.io/api/badge/repositories/3869" alt="babysor%2FMockingBird | Trendshift" style="width: 250px; height: 55px;" width="250" height="55"/></a>

 [![MIT License](https://img.shields.io/badge/license-MIT-blue.svg?style=flat)](http://choosealicense.com/licenses/mit/)

-> English | [中文](README-CN.md)
+> English | [中文](README-CN.md)| [中文Linux](README-LINUX-CN.md)

 ## Features
 🌍 **Chinese** supported mandarin and tested with multiple datasets: aidatatang_200zh, magicdata, aishell3, data_aishell, and etc.
@@ -29,6 +31,7 @@
 > If you get an `ERROR: Could not find a version that satisfies the requirement torch==1.9.0+cu102 (from versions: 0.1.2, 0.1.2.post1, 0.1.2.post2 )` This error is probably due to a low version of python, try using 3.9 and it will install successfully
 * Install [ffmpeg](https://ffmpeg.org/download.html#get-packages).
 * Run `pip install -r requirements.txt` to install the remaining necessary packages.
+> The recommended environment here is `Repo Tag 0.0.1` `Pytorch1.9.0 with Torchvision0.10.0 and cudatoolkit10.2` `requirements.txt` `webrtcvad-wheels` because `requirements. txt` was exported a few months ago, so it doesn't work with newer versions
 * Install webrtcvad `pip install webrtcvad-wheels`(If you need)

 or
@@ -126,7 +129,7 @@ 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://yisiou-my.sharepoint.com/:u:/g/personal/lawrence_cheng_yisiou_onmicrosoft_com/EWFWDHzee-NNg9TWdKckCc4BC7bK2j9cCbOWn0-_tK0nOg?e=Cc4EFA 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
+|@FawenYo | https://yisiou-my.sharepoint.com/:u:/g/personal/lawrence_cheng_fawenyo_onmicrosoft_com/EWFWDHzee-NNg9TWdKckCc4BC7bK2j9cCbOWn0-_tK0nOg?e=n0gGgC  | [input](https://github.com/babysor/MockingBird/wiki/audio/self_test.mp3) [output](https://github.com/babysor/MockingBird/wiki/audio/export.wav) | 200k steps with local accent of Taiwan, only works under version 0.0.1
 |@miven| https://pan.baidu.com/s/1PI-hM3sn5wbeChRryX-RCQ code: 2021 https://www.aliyundrive.com/s/AwPsbo8mcSP code: z2m0 | https://www.bilibili.com/video/BV1uh411B7AD/ | only works under version 0.0.1

 #### 2.4 Train vocoder (Optional)
--- a/models/synthesizer/preprocess.py
+++ b/models/synthesizer/preprocess.py
@@ -39,6 +39,9 @@ data_info = {
    }
 }

+def should_skip(fpath: Path, skip_existing: bool) -> bool:
+    return skip_existing and fpath.exists()
+
 def preprocess_dataset(datasets_root: Path, out_dir: Path, n_processes: int,
                           skip_existing: bool, hparams, no_alignments: bool, 
                           dataset: str, emotion_extract = False, encoder_model_fpath=None):
@@ -99,7 +102,7 @@ def preprocess_dataset(datasets_root: Path, out_dir: Path, n_processes: int,
    print("Max mel frames length: %d" % max(int(m[4]) for m in metadata))
    print("Max audio timesteps length: %d" % max(int(m[3]) for m in metadata))

-def embed_utterance(fpaths, encoder_model_fpath):
+def _embed_utterance(fpaths: str, encoder_model_fpath: str):
    if not encoder.is_loaded():
        encoder.load_model(encoder_model_fpath)

@@ -110,15 +113,13 @@ def embed_utterance(fpaths, encoder_model_fpath):
    embed = encoder.embed_utterance(wav)
    np.save(embed_fpath, embed, allow_pickle=False)
    
-def _emo_extract_from_utterance(fpaths, hparams, skip_existing=False):
-    if skip_existing and fpaths.exists():
-        return
+def _emo_extract_from_utterance(fpaths, hparams):
    wav_fpath, emo_fpath = fpaths
    wav = np.load(wav_fpath)
    emo = extract_emo(np.expand_dims(wav, 0), hparams.sample_rate, True)
    np.save(emo_fpath, emo.squeeze(0), allow_pickle=False)
 
-def create_embeddings(synthesizer_root: Path, encoder_model_fpath: Path, n_processes: int):
+def create_embeddings(synthesizer_root: Path, encoder_model_fpath: Path, n_processes: int, skip_existing: bool):
    wav_dir = synthesizer_root.joinpath("audio")
    metadata_fpath = synthesizer_root.joinpath("train.txt")
    assert wav_dir.exists() and metadata_fpath.exists()
@@ -128,11 +129,11 @@ def create_embeddings(synthesizer_root: Path, encoder_model_fpath: Path, n_proce
    # Gather the input wave filepath and the target output embed filepath
    with metadata_fpath.open("r", encoding="utf-8") as metadata_file:
        metadata = [line.split("|") for line in metadata_file]
-        fpaths = [(wav_dir.joinpath(m[0]), embed_dir.joinpath(m[2])) for m in metadata]
-        
+        fpaths = [(wav_dir.joinpath(m[0]), embed_dir.joinpath(m[2])) for m in metadata if not should_skip(embed_dir.joinpath(m[2]), skip_existing)]
+
    # TODO: improve on the multiprocessing, it's terrible. Disk I/O is the bottleneck here.
    # Embed the utterances in separate threads
-    func = partial(embed_utterance, encoder_model_fpath=encoder_model_fpath)
+    func = partial(_embed_utterance, encoder_model_fpath=encoder_model_fpath)
    job = Pool(n_processes).imap(func, fpaths)
    tuple(tqdm(job, "Embedding", len(fpaths), unit="utterances"))

@@ -142,14 +143,14 @@ def create_emo(synthesizer_root: Path, n_processes: int, skip_existing: bool, hp
    assert wav_dir.exists() and metadata_fpath.exists()
    emo_dir = synthesizer_root.joinpath("emo")
    emo_dir.mkdir(exist_ok=True)
-    
+
    # Gather the input wave filepath and the target output embed filepath
    with metadata_fpath.open("r", encoding="utf-8") as metadata_file:
        metadata = [line.split("|") for line in metadata_file]
-        fpaths = [(wav_dir.joinpath(m[0]), emo_dir.joinpath(m[0].replace("audio-", "emo-"))) for m in metadata]
+        fpaths = [(wav_dir.joinpath(m[0]), emo_dir.joinpath(m[0].replace("audio-", "emo-"))) for m in metadata if not should_skip(emo_dir.joinpath(m[0].replace("audio-", "emo-")), skip_existing)]
        
    # TODO: improve on the multiprocessing, it's terrible. Disk I/O is the bottleneck here.
    # Embed the utterances in separate threads
-    func = partial(_emo_extract_from_utterance, hparams=hparams, skip_existing=skip_existing)
+    func = partial(_emo_extract_from_utterance, hparams=hparams)
    job = Pool(n_processes).imap(func, fpaths)
    tuple(tqdm(job, "Emo", len(fpaths), unit="utterances"))
--- a/models/synthesizer/preprocess_audio.py
+++ b/models/synthesizer/preprocess_audio.py
@@ -45,7 +45,7 @@ def extract_emo(
    return y

 def _process_utterance(wav: np.ndarray, text: str, out_dir: Path, basename: str, 
-                      skip_existing: bool, hparams, encoder_model_fpath):
+                      mel_fpath: str, wav_fpath: str, hparams, encoder_model_fpath):
    ## FOR REFERENCE:
    # For you not to lose your head if you ever wish to change things here or implement your own
    # synthesizer.
@@ -58,13 +58,6 @@ def _process_utterance(wav: np.ndarray, text: str, out_dir: Path, basename: str,
    #   without extra padding. This means that you won't have an exact relation between the length
    #   of the wav and of the mel spectrogram. See the vocoder data loader.
        
-    # Skip existing utterances if needed
-    mel_fpath = out_dir.joinpath("mels", "mel-%s.npy" % basename)
-    wav_fpath = out_dir.joinpath("audio", "audio-%s.npy" % basename)
-    
-    if skip_existing and mel_fpath.exists() and wav_fpath.exists():
-        return None
-
    # Trim silence
    if hparams.trim_silence:
        if not encoder.is_loaded():
@@ -112,50 +105,27 @@ def _split_on_silences(wav_fpath, words, hparams):
 def preprocess_general(speaker_dir, out_dir: Path, skip_existing: bool, hparams, dict_info, no_alignments: bool, encoder_model_fpath: Path):
    metadata = []
    extensions = ("*.wav", "*.flac", "*.mp3")
-    if skip_existing:
-        for extension in extensions:
-            wav_fpath_list = speaker_dir.glob(extension)
-            # Iterate over each wav
-            for wav_fpath in wav_fpath_list:
-                words = dict_info.get(wav_fpath.name.split(".")[0])
+    for extension in extensions:
+        wav_fpath_list = speaker_dir.glob(extension)
+        # Iterate over each wav
+        for wav_fpath in wav_fpath_list:
+            words = dict_info.get(wav_fpath.name.split(".")[0])
+            if not words:
+                words = dict_info.get(wav_fpath.name) # try with extension 
                if not words:
-                    words = dict_info.get(wav_fpath.name) # try with extension 
-                    if not words:
-                        print("no wordS")
-                        continue
-                sub_basename = "%s_%02d" % (wav_fpath.name, 0)
-                
-                mel_fpath = out_dir.joinpath("mels", f"mel-{sub_basename}.npy")
-                wav_fpath_ = out_dir.joinpath("audio", f"audio-{sub_basename}.npy")
-                
-                if mel_fpath.exists() and wav_fpath_.exists():
+                    print(f"No word found in dict_info for {wav_fpath.name}, skip it")
                    continue
+            sub_basename = "%s_%02d" % (wav_fpath.name, 0)
+            mel_fpath_out = out_dir.joinpath("mels", f"mel-{sub_basename}.npy")
+            wav_fpath_out = out_dir.joinpath("audio", f"audio-{sub_basename}.npy")
+            
+            if skip_existing and mel_fpath_out.exists() and wav_fpath_out.exists():
+                continue
+            wav, text = _split_on_silences(wav_fpath, words, hparams)
+            result = _process_utterance(wav, text, out_dir, sub_basename, mel_fpath_out, wav_fpath_out, hparams, encoder_model_fpath)
+            if result is None:
+                continue
+            wav_fpath_name, mel_fpath_name, embed_fpath_name, wav, mel_frames, text = result
+            metadata.append ((wav_fpath_name, mel_fpath_name, embed_fpath_name, len(wav), mel_frames, text))

-                wav, text = _split_on_silences(wav_fpath, words, hparams)
-                result = _process_utterance(wav, text, out_dir, sub_basename, 
-                                                    False, hparams, encoder_model_fpath) # accelarate
-                if result is None:
-                    continue
-                wav_fpath_name, mel_fpath_name, embed_fpath_name, wav, mel_frames, text = result
-                metadata.append ((wav_fpath_name, mel_fpath_name, embed_fpath_name, len(wav), mel_frames, text))
-    else:
-        for extension in extensions:
-            wav_fpath_list = speaker_dir.glob(extension)
-            # Iterate over each wav
-            for wav_fpath in wav_fpath_list:
-                words = dict_info.get(wav_fpath.name.split(".")[0])
-                if not words:
-                    words = dict_info.get(wav_fpath.name) # try with extension 
-                    if not words:
-                        print("no wordS")
-                        continue
-                sub_basename = "%s_%02d" % (wav_fpath.name, 0)
-
-                wav, text = _split_on_silences(wav_fpath, words, hparams)
-                result = _process_utterance(wav, text, out_dir, sub_basename, 
-                                                    False, hparams, encoder_model_fpath)
-                if result is None:
-                    continue
-                wav_fpath_name, mel_fpath_name, embed_fpath_name, wav, mel_frames, text = result
-                metadata.append ((wav_fpath_name, mel_fpath_name, embed_fpath_name, len(wav), mel_frames, text))
    return metadata
--- a/pre.py
+++ b/pre.py
@@ -71,7 +71,7 @@ if __name__ == "__main__":
    del args.n_processes_embed
    preprocess_dataset(**vars(args))
    
-    create_embeddings(synthesizer_root=args.out_dir, n_processes=n_processes_embed, encoder_model_fpath=encoder_model_fpath)
+    create_embeddings(synthesizer_root=args.out_dir, n_processes=n_processes_embed, encoder_model_fpath=encoder_model_fpath, skip_existing=args.skip_existing)
    
    if args.emotion_extract:
        create_emo(synthesizer_root=args.out_dir, n_processes=n_processes_embed, skip_existing=args.skip_existing, hparams=args.hparams)
--- a/requirements.txt
+++ b/requirements.txt
@@ -2,7 +2,8 @@ umap-learn
 visdom
 librosa
 matplotlib>=3.3.0
-numpy
+numpy==1.19.3; platform_system == "Windows"
+numpy==1.20.3; platform_system != "Windows"
 scipy>=1.0.0
 tqdm
 sounddevice
@@ -12,8 +13,8 @@ inflect
 PyQt5
 multiprocess
 numba
-webrtcvad
-pypinyin
+webrtcvad; platform_system != "Windows"
+pypinyin==0.44.0
 flask
 flask_wtf
 flask_cors
@@ -25,9 +26,9 @@ PyYAML
 torch_complex
 espnet
 PyWavelets
-monotonic-align==0.0.3
-transformers
 fastapi
 loguru
-typer[all]
-click
+click==8.0.4
+typer
+monotonic-align==1.0.0
+transformers
Author	SHA1	Message	Date
Vega	918933ea86	Update README.md	2025-11-13 10:03:10 +08:00
Vega	1cde29d5f3	Update README.md (#1011 )	2024-11-14 21:00:29 -08:00
Vega	4b8fa992b7	Update README.md (#1009 )	2024-11-01 12:57:10 +08:00
Bob Conan	42789babd8	Update README.md, fix a typo (#1007 )	2024-10-22 10:21:44 +08:00
Vega	2354bb42d1	Update README.md (#1005 )	2024-10-16 22:48:15 +08:00
Vega	4358f6f353	Update README.md	2024-08-29 17:52:56 +08:00
xxxxx	5971555319	Update requirements.txt (#747 ) Ubuntu 20.04.1 CUDA 11.3 缺少依赖，还有依赖冲突 Co-authored-by: Vega <babysor00@gmail.com>	2024-08-22 15:06:40 +08:00
Emma Thompson	6f84026c51	Env update 添加环境需求注释 (#660 ) * Update Readme Doc 添加环境需求注释 * Update Readme Doc Add environmental requirement notes --------- Co-authored-by: Limingrui0 <65227354+Limingrui0@users.noreply.github.com>	2024-07-06 10:13:09 +08:00
Terminal	a30657ecf5	fix:preprocess_audio.py--The .npy file failed to save (#988 )	2024-07-06 10:12:36 +08:00
Terminal	cc250af1f6	fix requirements monotonic-align error (#989 )	2024-07-06 10:12:06 +08:00
Vega	156723e37c	Skip embedding (#950 ) * Skip embedding * Skip earlier * Remove unused paramater * Pass param	2023-09-05 23:15:04 +08:00
Vega	1862d2145b	Merge pull request #953 from babysor/babysor-patch-3 Update README.md	2023-08-31 11:42:15 +08:00
Vega	72a22d448b	Update README.md	2023-08-31 11:42:05 +08:00
Vega	98d38d84c3	Merge pull request #952 from SeaTidesPro/main add readme-linux-zh	2023-08-31 11:41:10 +08:00
Tide	7ab86c6f4c	Update README-LINUX-CN.md	2023-08-30 14:41:45 +08:00
Tide	ab79881480	Update README-LINUX-CN.md	2023-08-30 14:40:30 +08:00
Tide	fd93b40398	Update README-LINUX-CN.md	2023-08-30 14:35:34 +08:00
Tide	dbf01347fc	Update README-LINUX-CN.md	2023-08-30 14:35:12 +08:00
Tide	28f9173dfa	Update README-LINUX-CN.md	2023-08-30 14:34:20 +08:00
Tide	d073e1f349	Update README-LINUX-CN.md	2023-08-30 14:24:05 +08:00
Tide	baa8b5005d	Update README-LINUX-CN.md	2023-08-30 14:05:40 +08:00
Tide	d54f4fb631	Update README-LINUX-CN.md	2023-08-30 13:18:33 +08:00
Tide	7353888d35	Create README-LINUX-CN.md	2023-08-30 12:20:29 +08:00
Vega	e9ce943f6c	Merge pull request #947 from FawenYo/doc/update_link 📝 Update model download link	2023-08-11 22:02:41 +08:00
FawenYo	77c145328c	📝 Update model download link	2023-08-11 14:31:39 +08:00
Vega	3bce6bbbe7	Merge pull request #945 from babysor/babysor-patch-1 Update README.md	2023-08-10 15:54:23 +08:00
Vega	9dd8ea11e5	Merge pull request #944 from babysor/babysor-patch-2 Update README-CN.md	2023-08-10 15:53:52 +08:00
Vega	5a0d77e699	Update README-CN.md	2023-08-10 15:53:42 +08:00