nyrahealth/CrisperWhisper
A Whisper-based automatic speech recognition model that provides verbatim transcription with accurate word-level timestamps and filler detection.
Not currently ranked — collecting fresh signals.
star history
CrisperWhisper extends OpenAI’s Whisper to produce exact transcriptions of spoken audio, including disfluencies, fillers like um and uh, pauses, and false starts. It achieves improved word-level timestamp accuracy through an adjusted tokenizer and custom attention loss during training. The model was trained to minimize hallucinations and achieved 1st place on the OpenASR Leaderboard.
Frequently asked
- What is nyrahealth/CrisperWhisper?
- A Whisper-based automatic speech recognition model that provides verbatim transcription with accurate word-level timestamps and filler detection.
- Is CrisperWhisper open source?
- Yes — nyrahealth/CrisperWhisper is an open-source project tracked on heatdrop.
- What language is CrisperWhisper written in?
- nyrahealth/CrisperWhisper is primarily written in Python.
- How popular is CrisperWhisper?
- nyrahealth/CrisperWhisper has 975 stars on GitHub.
- Where can I find CrisperWhisper?
- nyrahealth/CrisperWhisper is on GitHub at https://github.com/nyrahealth/CrisperWhisper.