whisper audio.mp3 --model base --beam_size 5

Category: AI/ML Tooling

whisper audio.mp3 --model base --beam_size 5

Use beam search decoding

Runs beam search with a width of 5 instead of the greedy default, exploring several candidate transcripts and keeping the most probable one. Higher beam sizes improve accuracy on hard audio at a real speed cost. Values of 5 are a common sweet spot; the default is 1 (greedy).
Looking for more? Search all 7,657 commands — works offline, in English or Spanish, and fixes typos.