Fact · Speech recognition
Whisper's own timestamps are per utterance rather than per word, and can be inaccurate by several seconds.
Checked against its source on October 17, 2026
In the source's own words
“the corresponding timestamps are at the utterance-level, not per word, and can be inaccurate by several seconds”
WhisperX: README (GitHub, m-bain/whisperX), m-bain/whisperX project, GitHub (m-bain/whisperX), 2026, Introduction.
Open the original- What it measures
- WhisperX adds a wav2vec2 forced-alignment step to get word-level timestamps, and its README says so.
- What it does not show
- This is the project's description of Whisper's timestamps. It is a reason to use WhisperX for subtitles that must line up with words, not a general accuracy measure.
Source
Cited on
- How to Transcribe Audio to TextGuide
- How to Run Whisper Locally on Your Own ComputerGuide
- Captions and Subtitles: Formats, Rules and ToolsGuide
- Timestamps and Speaker Labels in TranscriptsGuide
- How do you make subtitles with Whisper?Question
- How do you transcribe audio with timestamps for free?Question
- How often should you add timestamps to a transcript?Question
- Timestamp (transcription)Glossary