DEV.co
AI Frameworks · fishaudio

Bert-VITS2

Bert-VITS2 is a Python-based text-to-speech (TTS) synthesis system combining VITS2 architecture with multilingual BERT embeddings for voice generation. The project is under AGPL-3.0 license and receives occasional updates, though the maintainers recommend their newer Fish-Speech project as a replacement.

Source: GitHub — github.com/fishaudio/Bert-VITS2
8.8k
GitHub stars
1.3k
Forks
Python
Primary language
AGPL-3.0
License (OSI-approved)

Key facts

Objective fields from the source. Values we can't verify are shown as “Unknown” rather than guessed.

FieldValue
Repositoryfishaudio/Bert-VITS2
Ownerfishaudio
Primary languagePython
LicenseAGPL-3.0 — OSI-approved
Stars8.8k
Forks1.3k
Open issues1
Latest releaseJP-Exta (2024-02-01)
Last updated2026-06-29
Sourcehttps://github.com/fishaudio/Bert-VITS2

What Bert-VITS2 is

Bert-VITS2 integrates BERT for linguistic features with VITS2 neural vocoding backbone to enable multilingual TTS. It references MassTTS and original VITS implementations, providing training pipelines for custom voice models. The codebase is Python-based with recent commits as of June 2026.

Quickstart

Get the Bert-VITS2 source

Clone the repository and explore it locally.

terminalbash
git clone https://github.com/fishaudio/Bert-VITS2.gitcd Bert-VITS2# follow the project's README for install & configuration

Need it deployed, integrated, or customized instead? DEV.co ships production installs.

Best use cases

Multilingual voice synthesis research

Suitable for academic or research settings exploring TTS with BERT-based linguistic conditioning across multiple languages. Reference implementation for VITS2-based architectures.

Custom voice model training

Enables training on proprietary or niche language datasets where pre-trained models are unavailable. Supports fine-tuning for specific speakers or acoustic characteristics.

Voice agent development (non-production)

Can prototype conversational agents with neural TTS in development/demo environments where inference speed and production SLAs are not critical constraints.

Implementation considerations

  • Training pipeline complexity: requires audio preprocessing, BERT tokenization, and multi-stage optimization. Refer to webui_preprocess.py for quickstart.
  • GPU compute requirements unknown; typical VITS2 training demands NVIDIA GPUs with 24GB+ VRAM.
  • Multilingual support depends on underlying BERT model availability and language-specific phonetic/linguistic preprocessing.
  • No built-in inference serving framework provided; custom deployment scaffolding required.
  • Dependencies on third-party VITS2 and BERT implementations; versions and compatibility matrices not clearly documented.

When to avoid it — and what to weigh

  • Production SLA requirements — Project is stated to be under reduced maintenance. No guarantees on inference latency, uptime, or support. Fish-Speech recommended as maintained alternative.
  • Commercial SaaS or proprietary deployment — AGPL-3.0 requires source code disclosure to end users. Incompatible with closed-source commercial products unless entire stack is open-sourced.
  • Regulatory or compliance-sensitive voice applications — No explicit security audit, privacy controls, or compliance documentation provided. Unknown handling of training data or inference logging.
  • Immediate need for actively maintained codebase — Maintainers explicitly state the project is not actively developed. Dependency updates, security patches, and bug fixes not guaranteed.

License & commercial use

AGPL-3.0 (GNU Affero General Public License v3.0). This is a copyleft license requiring that any software using or modifying Bert-VITS2 must also be released under AGPL-3.0 and source code must be made available to end users if distributed as a service.

Commercial use is legally possible under AGPL-3.0, but with a critical condition: if the software is offered as a network service (SaaS, hosted API, etc.), you must offer users the right to download and run your complete source code, including all modifications and integrations. Closed-source commercial products cannot incorporate Bert-VITS2. Requires legal review before commercial deployment.

DEV.co evaluation signals

Editorial assessment — not user reviews. Directional, with an explicit confidence level.

SignalAssessment
MaintenanceModerate
DocumentationLimited
License clarityClear
Deployment complexityHigh
DEV.co fitPossible
Assessment confidenceHigh
Security considerations

No security audit, privacy policy, or data handling documentation provided. Training data lineage unknown. No stated controls for inference logging or voice model theft. AGPL source code availability requirement may expose implementation details. Recommend threat modeling before handling sensitive speaker data or proprietary datasets.

Alternatives to consider

Fish-Speech (fishaudio/fish-speech)

Actively maintained by same organization. Autoregressive TTS, stated as open-source SOTA with ongoing updates. Recommended replacement in Bert-VITS2 README.

Glow-TTS or FastSpeech2 (opensource)

Non-autoregressive baselines with better inference speed and clearer training/deployment docs. Trade-off: potentially lower naturalness than Bert-VITS2.

Coqui TTS (TTS library, permissive license)

Pre-trained models, multiple architectures, MIT/Apache license (permissive), better production-ready tooling. No multilingual BERT integration.

Software development agency

Build on Bert-VITS2 with DEV.co software developers

Bert-VITS2 is ideal for research, prototyping, and voice model training in open-source environments. For production deployments or commercial SaaS, consult our engineers on licensing constraints and explore maintained alternatives like Fish-Speech. Contact us to assess fit and licensing implications.

Talk to DEV.co

Related open-source tools

Surfaced by semantic similarity across the DEV.co open-source index.

Related on DEV.co

Explore the category and the services that help you build with it.

Bert-VITS2 FAQ

Can I use Bert-VITS2 in a commercial product?
AGPL-3.0 permits commercial use, but requires you to open-source your entire product (including modifications and integrations) if distributed as a service. Closed-source or SaaS deployments are not compatible. Requires legal review.
Is this project actively maintained?
No. The README states the project is no longer actively developed and recommends Fish-Speech as a maintained alternative. Recent commits exist, but no active bug-fix or feature roadmap is published.
What are the hardware requirements?
Unknown from provided data. Typical VITS2 training requires NVIDIA GPU (24GB+ VRAM); inference requirements depend on batch size and language complexity. Requires benchmarking on target hardware.
Does it support voice cloning or speaker adaptation?
Not clearly documented in provided README. Project supports training on custom audio datasets and references speaker-specific fine-tuning, but no pre-built zero-shot cloning mechanism is stated.

Software development & web development with DEV.co

From first prototype to production, DEV.co delivers software development services around tools like Bert-VITS2. Our software development agency staffs experienced software developers and web developers for custom software development, web development, integrations, and ongoing support across ai frameworks and beyond.

Evaluate Bert-VITS2 for Your TTS Architecture

Bert-VITS2 is ideal for research, prototyping, and voice model training in open-source environments. For production deployments or commercial SaaS, consult our engineers on licensing constraints and explore maintained alternatives like Fish-Speech. Contact us to assess fit and licensing implications.