whisper audio.mp3 --model base --initial_prompt 'This is a technical podcast about Linux'
Guide transcription with context
Seeds the decoder with a hint about the audio's style and vocabulary, which often fixes repeated hallucinations and improves spelling of domain terms. Whisper weighs the prompt's style more than its exact words. Useful for names, jargon, and accented speech that default transcription mangles.
Looking for more? Search all 7,657 commands — works offline, in English or Spanish, and fixes typos.