GitHub - byzp/svc: SoftVC VITS Singing Voice Conversion

此分支的readme仅记录一些琐碎的改动, 使用方法与原项目一致

重写webui, 仅保留推理功能

训练中途ctrl+c后保存当前步数的模型, 两次ctrl+c立即终止

移除了whisper-large-v3 encoder以外的层, 默认推理精度改为int4, vits的默认推理精度改为fp16, 推理仅需3GB VRAM

添加训练参数grad_accumulate, 当batch_size受显存容量限制并且梯度震荡严重时可以尝试增大此参数, 它与batch_size的乘积近似于实际的batch_size

Name		Name	Last commit message	Last commit date
Latest commit History 873 Commits
.github		.github
cluster		cluster
configs		configs
configs_template		configs_template
dataset_raw		dataset_raw
diffusion		diffusion
edgetts		edgetts
filelists		filelists
inference		inference
logs/44k		logs/44k
modules		modules
onnxexport		onnxexport
pretrain		pretrain
raw		raw
trained		trained
vdecoder		vdecoder
vencoder		vencoder
.gitattributes		.gitattributes
.gitignore		.gitignore
.ruff.toml		.ruff.toml
LICENSE		LICENSE
README.md		README.md
compress_model.py		compress_model.py
data_utils.py		data_utils.py
export_index_for_onnx.py		export_index_for_onnx.py
flask_api.py		flask_api.py
flask_api_full_song.py		flask_api_full_song.py
gui.py		gui.py
gui_i18.py		gui_i18.py
inference_main.py		inference_main.py
models.py		models.py
onnx_export.py		onnx_export.py
onnx_export_old.py		onnx_export_old.py
preprocess_flist_config.py		preprocess_flist_config.py
preprocess_hubert_f0.py		preprocess_hubert_f0.py
requirements.txt		requirements.txt
requirements_onnx_encoder.txt		requirements_onnx_encoder.txt
requirements_win.txt		requirements_win.txt
resample.py		resample.py
sovits4_for_colab.ipynb		sovits4_for_colab.ipynb
spkmix.py		spkmix.py
train.py		train.py
train_broken.py		train_broken.py
train_diff.py		train_diff.py
train_index.py		train_index.py
utils.py		utils.py
wav_upload.py		wav_upload.py
webUI.py		webUI.py

Provide feedback