WhisperX
An open-source extension of Whisper adding accurate word-level timestamps and speaker diarisation.
WhisperX wraps Whisper to fix two of its practical gaps: it aligns the transcript to produce accurate word-level timestamps, and it adds speaker diarisation so you know who said what.
For anything beyond a plain transcript (captions, meeting notes, searchable audio), those two additions are usually what you actually need.
Where it fits
A fit when a transcript needs precise timing or speaker labels: captions, multi-speaker meeting notes, audio search.
Where it does not fit
More pipeline than needed when a plain, untimed transcript is all the task calls for.
Alternatives to WhisperX
The closest options on the map, and where each fits instead.
-
Whisper Speech-to-text model
The default when you need solid, multilingual transcription you can run yourself, with no audio leaving your environment.
-
faster-whisper Speech-to-text runtime
The default when you want Whisper-quality transcription self-hosted, but the reference implementation is too slow or memory-hungry.
Pairs with
Commonly used alongside it.