Request to add Vosk speech-to-text models

[Mokhyy]

Hi,

I wanted to request that, if possible, you add the Vosk speech-to-text models to your application. This engine supports many languages and is also very lightweight.

This model is used in the Sayboard speech-to-text keyboard and you can reuse its code:

https://github.com/ElishaAz/Sayboard

https://alphacephei.com/vosk/models

[BryceWG]

The model indeed appears to be very small, but from the provided error rate data, it doesn't seem very promising, so further evaluation will be needed later on.

[Mokhyy]

Thanks for the quick reply.
Please evaluate this Vosk model for me and let me know the results.

Honestly, among available STT models I only found Vosk that is both fast/lightweight and supports Persian well.

I just want you to add a Persian STT model to your app if possible. It doesn’t have to be Vosk specifically — any suitable Persian STT model is fine.

Thanks for the complete and helpful app you built.

[BryceWG]

I understand what you mean. The currently supported local models are indeed primarily trained for major languages. I will evaluate adding models that support more languages in the future, but such models often sacrifice some accuracy for mainstream languages

[BryceWG]

What do you think of omnilingual ASR?
https://github.com/facebookresearch/omnilingual-asr