Skip to content

v2.3 Final Release
Compare
Choose a tag to compare
@Stardust-minus Stardust-minus released this 19 Dec 11:17
· 98 commits to master since this release
v2.3
76653b5
This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
GPG key ID: 4AEE18F83AFDEB23
Expired
Learn about vigilant mode.

本次更新内容如下：
1.完善了 webui_preprocess.py（By @OedoSoldier ）
2.修正了存在于 Duration Predictor Discriminator 的 bug。并添加了 LSTM。对 SDP 有较大提升。
3.支持 BF16 训练。
4.对英语的 G2P 进行修复。
5.添加基于 WavLM 的 Discriminator（来源于 StyleTTS2）
6.因情感控制效果不佳，去除了 CLAP。更换了更简单的 BERT 融合语义方式。（By @litagin02 ）

这可能是 Bert-VITS2 近期的 final release，接下来我们将会加入对 Fish Speech 的开发。

Bert-VITS2项目组
敬上

This might be our final release for the time being as we shift our focus toward our new project Fish Speech.

We continue to appreciate your constructive advice for further improvements. Please feel free to share your thoughts in pull requests.

Base model: https://huggingface.co/OedoSoldier/Bert-VITS2-2.3

Update: The configuration for the gin_channels parameter in the pre-trained 2.3 weights has been adjusted to 512. To align with this change, we've updated the corresponding entry in the config.json file.

Contributors

OedoSoldier and litagin02
Assets 3
Loading