m-bain / whisperX

WhisperX: Automatic Speech Recognition with Word-level Timestamps (& Diarization)
BSD 2-Clause "Simplified" License
11.23k stars 1.18k forks source link

force CPU only #748

Open Abudinka opened 5 months ago

Abudinka commented 5 months ago

Hi,

i fooled around with whisperx for a project and tried it out on my computer. relly like the speed of it compared to other solutions, but i want to test performance and resource consumption without gpu. unfortunaltey --compute_type int8 doesnt do the trick with my machine (5800X3D, RTX2080) processing time of (a short audio file) is still within 5% and i cant see any difference in cpu gpu load trought execution (running on win11)

the command i executed (tried different switch regarding literally anthing than compute_type) is: whisperx zweisprecher.mp3 --model large-v2 --diarize --hf_token mytoken --language de --min_speakers 2 --compute_type int8

proycon commented 5 months ago

I also ran into this issue when trying to measure CPU/GPU difference. Try passing --device cpu.