Is it possible to use more CPU cores for Whisper?

[peter1rhodes]

I found I got terrible results when using any of the models that would fit on my RTX3070, so I am using CPU models instead with Whisper large v3, and getting good results (though it takes a couple of hours to process and diarise an hour of audio).

I've noticed that whisperx is only using four cores. I have 16 available. Is it possible to tell Scriberr to use more cores for transcription?

[dukei]

try to set one of the following env variables:

OMP_NUM_THREADS=16
MKL_NUM_THREADS=16

However, it affects diarization only. If you want to increase number of threads for WisperX, you need to pass --threads parameter to it. Unfortunately Scribber currently does not pass this param. I managed to hardcode --threads 24 into Scriberr/internal/transcription/adapters/whisperx_adapter.go at line 507:

	args = append(args, "--threads", "24")

and then recompile with docker-compose.build.yml

[dukei]

I created pull request #407 to support WHX_NUM_THREADS environment variable for setting number of threads for WhisperX

[rishikanthc]

@dukei is right. I need to modify my implementation to support threads.. will look into the PR

[rishikanthc]

Hi all ! I know it's been a while since this issue was opened and folks might be wondering why there hasn't been any updates.
I just wanted to let you know that development is temporarily paused but not abandoned. I recently got laid off and am urgently trying to find my next job. See README for details, and thanks for your patience! Things should resume soon once I get back up on my feet.