---
license: mit
language:
- ru
library_name: phoonnx
pipeline_tag: text-to-speech
tags:
- tts
- vits2
- onnx
- phoonnx
- offline
---

# phoonnx-vits2

Mirror of VITS2 text-to-speech checkpoints exported to ONNX for
[phoonnx](https://github.com/TigreGotico/phoonnx), OpenVoiceOS's offline TTS
library. VITS2 is a single-stage TTS architecture, an evolution of VITS that
improves naturalness and inference speed through adversarial duration
modeling and a monotonic alignment search transformer.

This repository holds the raw ONNX weights and `config.json` files consumed
by `phoonnx`'s `voice_index`; it is not meant to be browsed for voices
directly — use `phoonnx`'s voice manager, which resolves voice IDs to the
files here.

## Contents

| Directory | Voice ID | Language | Source |
|---|---|---|---|
| `frappuccino-ru-natasha/` | `frappuccino/vits2-ru-natasha` | Russian (`ru`) | [`frappuccino/vits2_ru_natasha`](https://huggingface.co/frappuccino/vits2_ru_natasha), trained on the Natasha dataset, shared by LangSwap.app |

## Usage

```python
import wave
from phoonnx.model_manager import TTSModelManager

manager = TTSModelManager()
manager.load()
manager.merge_default_voices()

voice = manager.voices["frappuccino/vits2-ru-natasha"].load()
with wave.open("out.wav", "wb") as wav_file:
    voice.synthesize_wav("Привет, как дела?", wav_file)
```

## License

The `frappuccino-ru-natasha` checkpoint is released by its author under the
MIT license. The VITS2 architecture and training code (`p0p4k/vits2_pytorch`)
are also MIT licensed. `phoonnx` itself is Apache-2.0.