enhancement: external stt
Stargate256 opened this issue · comments
Is your feature request related to a problem? Please describe.
Would it be possible to allow running whisper on another server?
Describe the solution you'd like
Adding the ability to connect to whisper via some kind of API.
Additional context
I am running open-webui on a server without GPUs, and LLMs, TTS on another server.
I've recently developed faster-whisper-server which provides an OpenAI compatible transcription/translation (and some extras as well) which could be used as a transcription backend if OpenWebUI adds support for having an external STT
STT & TTS settings have been refactored on dev and you can now configure external OpenAI STT from Admin Settings!
I tried this and the model name for STT is not correctly used.
ValueError: Invalid model size 'whisper-1', expected one of: tiny.en, tiny, base.en, base, small.en, small, medium.en, medium, large-v1, large-v2, large-v3, large, distil-large-v2, distil-medium.en, distil-small.en, distil-large-v3
STT & TTS settings have been refactored on dev and you can now configure external OpenAI STT from Admin Settings!
Amazing.. well done!!
I tried this and the model name for STT is not correctly used.
ValueError: Invalid model size 'whisper-1', expected one of: tiny.en, tiny, base.en, base, small.en, small, medium.en, medium, large-v1, large-v2, large-v3, large, distil-large-v2, distil-medium.en, distil-small.en, distil-large-v3
@tjbck The model name is being ignored. As a result we can't use @fedirz's faster-whisper-server because it doesn't know what "whisper-1" is.
I modified faster-whisper-server to treat "whisper-1" as the default model and it worked great! @fedirz, maybe this is a good feature you could add? @tjbck, could you make the fix so we can choose the models?
Thank you both for your work!
@JohnSmithToYou yeah, I'll add special handling for whisper-1