Repolode
Submit
fishaudio

fishaudio/Bert-VITS2

A VITS2-based text-to-speech project that adds multilingual BERT to improve prosody and pronunciation in synthesized speech.

View on GitHub
Stars
8,801
Forks
1,307
Open issues
2
Last commit
5 days ago

What is it?

Bert-VITS2 is a text-to-speech project built on the VITS2 architecture. It feeds text features from a multilingual BERT model into the VITS2 backbone, aiming to make synthesized speech more natural in prosody and pronunciation.

The repository provides preprocessing and training workflows, with a web UI entry point (webui_preprocess.py) that serves as the quick guide for getting started. The project's core approach draws on the MassTTS project, and it includes references to related work such as vits, vits2_pytorch and so-vits-svc.

The maintainers note that this project is no longer actively maintained and recommend Fish-Speech as the successor for users who want a current autoregressive TTS option.

Key features

  • 01Combines VITS2 with multilingual BERT text features for speech synthesis.
  • 02Provides a WebUI-based preprocessing and training workflow.
  • 03Built on the VITS2 backbone with a vocoder component.
  • 04Source is licensed under AGPL-3.0.

Use cases

  • Training a custom voice model from your own recorded speech data.
  • Studying how BERT text embeddings can be integrated into a VITS-style TTS model.
  • Experimenting with Chinese and multilingual speech synthesis pipelines.

Things to know

  • The project is licensed under AGPL-3.0, a strong copyleft license.
  • The README says the project will not be maintained in the short term and recommends Fish-Speech instead.
  • The README bans use for political purposes and for activities that violate Chinese law.

In collections

AI video & audio tools

Topics

#tts#text-to-speech#vits#bert#speech-synthesis#voice-cloning#deep-learning#agent#bert-vits#bert-vits2

FAQ

What is Bert-VITS2?
A VITS2-based text-to-speech project that adds multilingual BERT to improve prosody and pronunciation in synthesized speech. Bert-VITS2 is a text-to-speech project built on the VITS2 architecture. It feeds text features from a multilingual BERT model into the VITS2 backbone, aiming to make synthesized speech more natural in prosody and pronunciation.
Is Bert-VITS2 free to use? What license does it have?
Bert-VITS2 is open source under the AGPL-3.0 license. Check the license terms before using it in commercial products.
Is Bert-VITS2 still maintained?
Actively maintained. The last commit to Bert-VITS2 was on Nov 2, 2026.
What can I use Bert-VITS2 for?
Training a custom voice model from your own recorded speech data. Studying how BERT text embeddings can be integrated into a VITS-style TTS model. Experimenting with Chinese and multilingual speech synthesis pipelines.
Where can I get Bert-VITS2?
The source code and installation instructions are on GitHub: https://github.com/fishaudio/Bert-VITS2

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects