VITS2 Backbone with bert
基于b站up主团子是咸鱼的视频教程制作【Bert-Vits2 手把手本地部署录屏教程【已适配V1.1】】 https://www.bilibili.com/video/BV18N4y1Q7JK/?share_source=copy_web&vd_source=0850903c2c3f5ed7b220c9bda4e285f6
【免费显卡?Bert-VITS2 colab脚本在线训练 教你免费在线训练属于自己的虚拟音声-哔哩哔哩】 https://b23.tv/gJWyXEs
1.若干时长的高质量语音文本
2.音频切割工具slicer-gui
3.数据处理和标注工具,来自b站up主领航员未鸟【Bert-VITS2/VITS】更快更准确!自动标注一键包来啦】 https://www.bilibili.com/video/BV1dr4y1X7RL/?share_source=copy_web&vd_source=0850903c2c3f5ed7b220c9bda4e285f6
!git clone https://github.com/MengXing15646/Bert-VITS2
1.原始数据音频必须是.wav格式,使用slicer-gui将文件切割为短于20s的音频
2.按照【Bert-VITS2/VITS】更快更准确!自动标注一键包来啦】 https://www.bilibili.com/video/BV1dr4y1X7RL/?share_source=copy_web&vd_source=0850903c2c3f5ed7b220c9bda4e285f6 教程处理原始数据集
3.将处理后的clean_barbara.list文件上传到filelists文件夹,并改名成【自己需要的名字】
4.将preprocess_text文件中的名字改为【自己需要的名字】
5.创建raw文件夹和它的子文件夹【自己需要的名字】,将处理后的音频上传到【自己需要的名字】文件夹
%cd /content/Bert-VITS2/
!pip install -r requirements.txt
hugging face不提供直链下载,使用二次分发下载模型
!wget -P bert/chinese-roberta-wwm-ext-large/ https://huggingface.co/hfl/chinese-roberta-wwm-ext-large/resolve/main/flax_model.msgpack
!wget -P bert/chinese-roberta-wwm-ext-large/ https://huggingface.co/hfl/chinese-roberta-wwm-ext-large/resolve/main/pytorch_model.bin
!wget -P bert/chinese-roberta-wwm-ext-large/ https://huggingface.co/hfl/chinese-roberta-wwm-ext-large/resolve/main/tf_model.h5
!wget -P logs/{datasetname}/ https://huggingface.co/Erythrocyte/bert-vits2_base_model/resolve/main/DUR_0.pth
!wget -P logs/{datasetname}/ https://huggingface.co/Erythrocyte/bert-vits2_base_model/resolve/main/D_0.pth
!wget -P logs/{datasetname}/ https://huggingface.co/Erythrocyte/bert-vits2_base_model/resolve/main/G_0.pth
!python resample.py
!python preprocess_text.py
!python bert_gen.py
!python train_ms.py -m {datasetname} -c configs/config.json
!python webui.py