Supertone

/

supertonic

Like 488

Follow

Supertone 629

Summary

Requesting native support and model training recipes for Bangla (bn / bn_BD / bn_IN).

Core Statistics

• Rank: 5th most spoken native language, 7th overall globally.
• Speakers: ~274M total speakers (234M+ native).
• Current Status: Minimal lightweight/on-device neural TTS coverage compared to European/East Asian languages.

Integration Requirements

• Grapheme-to-Phoneme (G2P) / eSpeak-ng phonemizer mapping for Bangla script.
• Dataset formatting (e.g., open-source datasets like MMS, Google Crowdsourced Bangla, or custom VITS datasets).
• Training config & ONNX export specs for on-device inference.

Technical Resources & Datasets Available

• Open-source Bangla speech datasets exist on Hugging Face (e.g., google/expressive-bangla-speech, facebook/mms-tts-ben).
• eSpeak-ng has base bn phoneme tables ready for pipeline integration.

Are the maintainers open to accepting community-trained weights, or what is the preferred workflow to add bn_BD into the voice repository?