Nyra Labs · Speech-to-Text

CrisperWhisper 2.0

Whisper-based verbatim speech-to-text with two modes: a literal transcript of every filler, repetition and stutter, or a cleaned-up "intended" read. The 2.0 Pro tier scores 93.5 disfluency F1 across 10 languages, times words within 30ms on read speech, and hits 96.1 percent rare-word recall. Open weights on Hugging Face.