Add Audio8-TTS-Preview-0.6b

[johappel]

What

I'd like to propose support for [Audio8-TTS-Preview-0.6b](https://huggingface.co/Audio8/Audio8-TTS-Preview-0.6b).

Audio8-TTS is a relatively small multilingual TTS model (~0.6B parameters) with support for multiple languages, including German, and zero-shot voice cloning from reference audio.

Would this model be a good fit for audio.cpp, and is anyone already working on a port?

Why it might fit audio.cpp

• Relatively small model (~0.6B), making it attractive for local inference.
• Multilingual TTS, including German.
• Zero-shot voice cloning from reference audio.
• Could provide another lightweight multilingual alternative to larger TTS/voice-cloning models already supported by audio.cpp.
• The model is available through Hugging Face and currently has a Transformers/PyTorch reference implementation that could serve as the parity reference.

Model

https://huggingface.co/Audio8/Audio8-TTS-Preview-0.6b

I haven't investigated the architecture deeply enough yet to estimate how much of it can reuse existing audio.cpp components, so this is primarily a model proposal rather than a port announcement.

If it looks interesting, I'd also be happy to help test the model, especially German output, voice cloning quality, and CUDA performance.

[0xShug0]

@johappel It's the same Dual AR architecture as Fish Audio. I tested this model when it was released, and I was a little disappointed with the quality compared with Fish Audio (just my personal opinion). But I completely missed the "small size" part. I will add it to todos.

[0xShug0]

BTW PRs are welcome if you’re interested in porting it as a community model!