vits-av_ru-mms-graphemes
Avar text-to-speech in ONNX format
1 min readupdated
A text-to-speech (TTS) model for the Avar language (locale code av-RU). It is based on Meta's original VITS-based facebook/mms-tts-ava model, converted to ONNX and packaged by Phoonnx to run with the phoonnx library. Unlike the original PyTorch version, this build is ready for the ONNX Runtime and uses the built-in scriptconv frontend for text normalization. It is a solid fit for developers building voice assistants and local services with Avar language support.
Key features
- VITS architecture adapted from
facebook/mms-tts-ava(originally created by Meta). - Ships in ONNX format for inference via the
phoonnxPython library. - Takes raw text characters (
graphemes) as input, converting them withGraphemePhonemizerfrom thescriptconvlibrary. - Integrates directly into the OVOS voice assistant through the
ovos-tts-plugin-phoonnxplugin.
What you can do
- Add Avar speech synthesis to a Python script or local service and save the output as WAV audio.
- Configure the
facebook/mms-tts-ava-Avarvoice in the OVOS voice assistant settings (mycroft.conf).
Example
import wave
from phoonnx.model_manager import TTSModelManager
manager = TTSModelManager()
manager.load()
manager.merge_default_voices()
voice = manager.voices["facebook/mms-tts-ava-Avar"].load()
with wave.open("output.wav", "wb") as wav_file:
voice.synthesize_wav("...", wav_file)Read more
DetailsLanguage: Avar (av-RU).
Voice ID in phoonnx: facebook/mms-tts-ava-Avar.
License: cc-by-nc-4.0 (inherited from facebook/mms-tts-ava, non-commercial use only).
Installation: pip install --pre phoonnx.










Comments