mirror of
https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI.git
synced 2025-01-01 12:35:04 +08:00
一个基于VITS的简单易用的变声框架
audio-analysischangeconversational-aiconversionconverterretrieval-modelretrieve-datarvcso-vits-svcsovitsvcvitsvoicevoice-conversionvoice-convertervoiceconversion
ef738704d0
* Step 2a fix by quoting config.python_cmd (#1030) - Ensure `config.python_cmd` is properly quoted when constructing command strings. - Address issue where paths with spaces (e.g., "C:\Program Files\...") caused command execution failure on Windows during training. - This is a fix for step 2a of training. * Step 2b rmvpe_gpu fix by quoting config.python_cmd (#1030) - Ensure `config.python_cmd` is properly quoted when constructing command strings. - Address issue where paths with spaces (e.g., "C:\Program Files\...") caused command execution failure on Windows during training. - This is a fix for step 2b of training (rmvpe_gpu) mode specifically. * Step 2b pm fix by quoting config.python_cmd (#1030) - Quoted `config.python_cmd` to ensure proper command string construction. - Resolved an issue where paths with spaces, like "C:\Program Files\...", caused execution failures on Windows during training. - While this fix targets Step 2b of training (pm mode), it might also address issues in other modes (harvest, dio, rmvpe). I've only verified the fix post-change for these modes. * Step 3 Train model fix by quoting config.python_cmd (#1030) - Ensure `config.python_cmd` is properly quoted when constructing command strings. - Address issue where paths with spaces (e.g., "C:\Program Files\...") caused command execution failure on Windows during training. - This is a fix for step 3 of training: (Train model) button specifically. Train feature index seems to be working correctly, I have only verified this post-fix. I've not yet tested the "One-click training" button. * Quote config.python_cmd for Step 3 One-click training (#1030) - Addressed an issue where paths with spaces (e.g., "C:\Program Files\...") caused command execution failures on Windows during training. - Specifically, ensured `config.python_cmd` is properly quoted when constructing command strings for step 3 of training (One-click training button). Notes: - All remaining instances of unquoted `config.python_cmd` in `infer-web.py` seem to be piped into a command shell. These might exhibit similar issues when executed. - It might be simpler to replace `config.python_cmd` itself with its quoted version, unless there are instances where it's used differently. - This likely resolves the issue for the `Train` tab, but further review is advised. |
||
---|---|---|
.github | ||
configs | ||
docs | ||
lib | ||
logs/mute | ||
pretrained | ||
pretrained_v2 | ||
tools | ||
uvr5_weights | ||
weights | ||
.gitignore | ||
app.py | ||
config.py | ||
docker-compose.yml | ||
Dockerfile | ||
environment_dml.yaml | ||
extract_f0_print.py | ||
extract_f0_rmvpe_dml.py | ||
extract_f0_rmvpe.py | ||
extract_feature_print.py | ||
extract_locale.py | ||
go-realtime-gui.bat | ||
go-web.bat | ||
gui_v1.py | ||
i18n.py | ||
infer_batch_rvc.py | ||
infer_cli.py | ||
infer_uvr5.py | ||
infer-web.py | ||
LICENSE | ||
MDXNet.py | ||
MIT协议暨相关引用库协议 | ||
poetry.lock | ||
pyproject.toml | ||
README.md | ||
requirements-dml.txt | ||
requirements-win-for-realtime_vc_gui-dml.txt | ||
requirements-win-for-realtime_vc_gui.txt | ||
requirements.txt | ||
Retrieval_based_Voice_Conversion_WebUI_v2.ipynb | ||
Retrieval_based_Voice_Conversion_WebUI.ipynb | ||
run.sh | ||
rvc_for_realtime.py | ||
scan_i18n.py | ||
train_nsf_sim_cache_sid_load_pretrain.py | ||
trainset_preprocess_pipeline_print.py | ||
values1.json | ||
vc_infer_pipeline.py | ||
venv.sh |
Retrieval-based-Voice-Conversion-WebUI
一个基于VITS的简单易用的语音转换(变声器)框架更新日志 | 常见问题解答 | AutoDL·5毛钱训练AI歌手 | 对照实验记录 | 在线演示
English | 中文简体 | 日本語 | 한국어 (韓國語) | Türkçe
点此查看我们的演示视频 !
使用了RVC的实时语音转换: w-okada/voice-changer
底模使用接近50小时的开源高质量VCTK训练集训练,无版权方面的顾虑,请大家放心使用
请期待RVCv3的底模,参数更大,数据更大,效果更好,基本持平的推理速度,需要训练数据量更少。
简介
本仓库具有以下特点
- 使用top1检索替换输入源特征为训练集特征来杜绝音色泄漏
- 即便在相对较差的显卡上也能快速训练
- 使用少量数据进行训练也能得到较好结果(推荐至少收集10分钟低底噪语音数据)
- 可以通过模型融合来改变音色(借助ckpt处理选项卡中的ckpt-merge)
- 简单易用的网页界面
- 可调用UVR5模型来快速分离人声和伴奏
- 使用最先进的人声音高提取算法InterSpeech2023-RMVPE根绝哑音问题。效果最好(显著地)但比crepe_full更快、资源占用更小
- A卡I卡加速支持
环境配置
以下指令需在 Python 版本大于3.8的环境中执行。
(Windows/Linux)
首先通过 pip 安装主要依赖:
# 安装Pytorch及其核心依赖,若已安装则跳过
# 参考自: https://pytorch.org/get-started/locally/
pip install torch torchvision torchaudio
#如果是win系统+Nvidia Ampere架构(RTX30xx),根据 #21 的经验,需要指定pytorch对应的cuda版本
#pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu117
可以使用 poetry 来安装依赖:
# 安装 Poetry 依赖管理工具, 若已安装则跳过
# 参考自: https://python-poetry.org/docs/#installation
curl -sSL https://install.python-poetry.org | python3 -
# 通过poetry安装依赖
poetry install
你也可以通过 pip 来安装依赖:
N卡:
pip install -r requirements.txt
A卡/I卡:
pip install -r requirements-dml.txt
Mac 用户可以通过 run.sh
来安装依赖:
sh ./run.sh
其他预模型准备
RVC需要其他一些预模型来推理和训练。
你可以从我们的Hugging Face space下载到这些模型。
以下是一份清单,包括了所有RVC所需的预模型和其他文件的名称:
hubert_base.pt
./pretrained
./uvr5_weights
想测试v2版本模型的话,需要额外下载
./pretrained_v2
如果你正在使用Windows,则你可能需要这个文件,若ffmpeg和ffprobe已安装则跳过; ubuntu/debian 用户可以通过apt install ffmpeg来安装这2个库, Mac 用户则可以通过brew install ffmpeg来安装 (需要预先安装brew)
./ffmpeg
https://huggingface.co/lj1995/VoiceConversionWebUI/blob/main/ffmpeg.exe
./ffprobe
https://huggingface.co/lj1995/VoiceConversionWebUI/blob/main/ffprobe.exe
如果你想使用最新的RMVPE人声音高提取算法,则你需要下载音高提取模型参数并放置于RVC根目录
https://huggingface.co/lj1995/VoiceConversionWebUI/blob/main/rmvpe.pt
A卡I卡用户需要的dml环境要请下载
https://huggingface.co/lj1995/VoiceConversionWebUI/blob/main/rmvpe.onnx
之后使用以下指令来启动WebUI:
python infer-web.py
如果你正在使用Windows 或 macOS,你可以直接下载并解压RVC-beta.7z
,前者可以运行go-web.bat
以启动WebUI,后者则运行命令sh ./run.sh
以启动WebUI。
仓库内还有一份小白简易教程.doc
以供参考。
参考项目
- ContentVec
- VITS
- HIFIGAN
- Gradio
- FFmpeg
- Ultimate Vocal Remover
- audio-slicer
- Vocal pitch extraction:RMVPE