What is best quality settings (speed not important) #541
-
|
Hi! I’m using faster-whisper-xxl.exe and I want maximum transcription/subtitle quality (I don’t care about speed). This is my current command: $env:CUDA_VISIBLE_DEVICES="0" $purfExe="faster-whisper-xxl.exe" & $purfExe For best accuracy, should I switch to -m large-v3 (if supported) and/or --compute_type float32? What are the recommended values for --beam_size, --best_of, --temperature, and --temperature_increment_on_fallback when optimizing for quality? Are there any other flags (e.g., VAD options, punctuation/diarization options, etc.) that improve subtitle quality in this build? Does --batched affect accuracy, or only speed/VRAM? Environment (if useful): OS: Windows Device: CUDA GPU (can provide GPU model/VRAM if needed) Output: .srt Thanks! |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 5 replies
-
|
The best is whatever works best for you. IMO, these are generally the best: Current public version: Current Pro version: |
Beta Was this translation helpful? Give feedback.
The best is whatever works best for you. IMO, these are generally the best:
Current public version:
...exe "my.mp4" -l en -m large-v2 -ct float16 --vad_method pyannote_v3 --standardCurrent Pro version:
...exe "my.mp4" -l en -m large-v2 -ct float16 --vad_method pyannote_v3 --standard --ff_vocal_extract mb-roformer --realign