Purfview
Purfview
>Expected Behavior: >The transcription results from the asynchronous calls should be identical to the synchronous call's results. >I have tried the solution which you have mentioned. but still the same...
Read it again: >Whisper model is non-deterministic if temperature is not 0, try to set `temperature=0`, but results may degrade.
> yes @Purfview it is producing new results every time. It doesn't when you set temperature to 0.
> yes, @Purfview but the generated text is incorrect. Whisper doesn't guaranty a correct result, you can try a bigger model like `large-v2`. And: > ...but results may degrade.
BTW, your result looks like it contains hallucinations, try `hallucination_silence_threshold=2`
> can you please help me where can I use `hallucination_silence_threshold=2` in the given code? `model.transcribe('audio_path', language="en", hallucination_silence_threshold=1)` doesn't work? > I process this audio data by converting the bytes...
Check if VAD didn't cut off those missing segments.
I think you don't need long audio, it should crash with any audio where it hits fallback, try this with `tiny` model: [error_withtc_5s.zip](https://github.com/user-attachments/files/19929155/error_withtc_5s.zip) Better question, why it doesn't crash for...
>I am seeing the error in the Windows Event Log (not as a pop-up); so script crashes -> I am checking Event Viewer -> Filter Critical Then I guess that...
If unloading to cpu doesn't produce crash then maybe it can be used as workaround, unload to cpu then "del model". I can't test it as I don't use Windows+CUDA.