Add IndicF5 Support for Indian Languages

[hariOneb]

Summary

Would love to see IndicF5 added as a TTS backend in Voicebox! It would bring support for 11 Indian languages including Tamil, Hindi, Bengali, Telugu, Malayalam, Kannada, Gujarati, Marathi, Punjabi, Odia, and Assamese — languages spoken by over 1.4 billion people that are currently not supported by any existing Voicebox backend.

What is IndicF5?

IndicF5 is a near-human quality, open-source Text-to-Speech model by AI4Bharat, trained on 1417 hours of high-quality Indian language speech data.

• 🔗 GitHub: https://github.com/AI4Bharat/IndicF5
• 🤗 HuggingFace: https://huggingface.co/ai4bharat/IndicF5
• 📄 License: Open Source

Why it fits Voicebox perfectly

• ✅ Pure Python library — integrates cleanly with the existing FastAPI backend
• ✅ Follows the same reference-audio voice cloning approach as existing backends
• ✅ Works with PyTorch + CUDA — no new dependencies outside what Voicebox already uses
• ✅ Installable with a single pip command
• ✅ Voicebox's modular backends/ architecture in v0.3.0 makes this straightforward to add

How it works

IndicF5 takes 3 inputs — exactly like other Voicebox backends:

1. Text to synthesize
2. A reference audio clip (for voice cloning)
3. Transcript of the reference audio

from transformers import AutoModel
model = AutoModel.from_pretrained("ai4bharat/IndicF5", trust_remote_code=True)
audio = model(text, ref_audio_path="sample.wav", ref_text="transcript")

Impact

There is currently no good local, open-source voice cloning tool for Indian languages. Adding IndicF5 would make Voicebox the go-to tool for Indian language TTS and would open the app to a massive new user base.

Happy to help!

I'm willing to contribute a PR if that helps move this forward. Would love to know if this is something you'd consider including!