Sounds pretty cool 
today i will test the whisper model on our nvidia t4 card with this locally hosted api version.
I have to figure it out how can do the timing for precise results, but somehow i can do it i think
So, i have one number, but going forward for testing the whole thing. I have played with faster-whisper and i think its looks good for first tries.
I have an audio sample in our language (Hungarian, which is 8 seconds long) I put the audio to ASR and the outcome is generated 1.9 seconds.This is 1:4 performance on a local nVidia T4 card which is not the worst option regarding the budget (like A10 or A100 cards)
I will do some more testing and also i will try your Voicebot in our test environment with local whisper, but i need some time for that
1 Like
@kissze I merged the PR, if you have any problem, please open an issue in the project and we continue from there