Insanely Fast Whisper — быстрая транскрипция аудио

★ 13K

Insanely Fast Whisper is a tool for very fast audio transcription with Whisper models: it optimizes the run to turn speech into text several times faster than usual. Reach for it when you need to quickly transcribe a lot of, or long, audio on your GPU: convert recordings, podcasts, and interviews into text in minimal time, from the command line or a script. It uses accelerations (batching, flash-attention, and more) on top of Whisper and runs locally. Its focus is the fastest possible speech-to-text on a GPU via CLI, not transcription with precise timestamps and speaker diarization (whisperX for that) or a desktop app with a UI (buzz): the value is throughput on large volumes of audio. It suits developers and anyone who needs to run many recordings through Whisper quickly without waiting.