Can you please add MetaVoice 1B

[priya-dwivedi]

Awesome library. Can you integrate Metavoice 1B
https://github.com/metavoiceio/metavoice-src

[KoljaB]

Hi Priyanka,

Thank you for your interest in the library, and I must say, your Github profile is impressive!

Regarding MetaVoice, I've done some research. While its voice quality is decent (not at par with XTTS though) it currently falls short for real-time applications. Here's why:

MetaVoice lacks support for streaming (as noted in this issue), which means we're limited to processing only complete sentences. It operates at a realtime factor of approximately 0.66 on an RTX 4090. This translates to a latency of around 6.6 seconds for a 10-second sentence, which is quite substantial for real-time applications. Additionally, the limitation to English is a significant drawback and also another notable concern is its high VRAM requirement of 12GB, which could be prohibitive for many users and systems.

In summary, given these constraints, I'd hesitate to recommend MetaVoice for real-time usage at this stage.

[priya-dwivedi]

I appreciate your detailed answer on above. Thanks a lot!

[mercuryyy]

@KoljaB There is this git for metavoice live - https://github.com/ferret99gt/GradioMVL

claiming 80ms to first audio with streaming.

Worth taking a look

[KoljaB]

Can't test, 8 GB VRAM GPU poor here (won't upgrade until RTX 5xxx). Edit: Metavoice needs >= 12GB VRAM...