vits-new-mms-graphemes
Text-to-speech model for Newar
1 min readupdated
An adaptation of Meta's VITS architecture based on the upstream facebook/mms-tts-new release, packaged for the phoonnx ONNX library. It handles Text-to-Speech synthesis for the Newar language, converting text graphemes into quality WAV audio. Built on Meta's proven MMS (Massively Multilingual Speech) framework, the model is optimized to run via phoonnx and integrate with the OVOS voice assistant. Ideal for developers building local voice interfaces and researchers working with low-resource languages.
Key features
- VITS architecture for quality generative speech synthesis
- Based on Meta's facebook/mms-tts-new project
- Fully packaged for the phoonnx ONNX library
- Supports integration into the OVOS voice assistant via the ovos-tts-plugin-phoonnx plugin
What you can do
- Synthesize Newar speech from text strings using a Python script
- Hook the model up to the OVOS ecosystem through the mycroft.conf configuration file
- Use the built-in scriptconv text frontend to normalize characters and graphemes
Example
Generating Newar speech with phoonnx in Python:
import wave
from phoonnx.model_manager import TTSModelManager
manager = TTSModelManager()
manager.load()
manager.merge_default_voices()
voice = manager.voices["facebook/mms-tts-new-Newar"].load()
with wave.open("output.wav", "wb") as wav_file:
text = "..." # a sentence in new
voice.synthesize_wav(text, wav_file)
Read more
DetailsArchitecture: VITS
Upstream: facebook/mms-tts-new
Language: Newar (`new`)
License: cc-by-nc-4.0










Comments