CrisperWhisper 2

by Nyra Labs · Model · Voice & audio

Open transcription family with verbatim and intended modes plus word-level timestamps; free Hugging Face Space

Evidence

  1. Aug 2
    On its self-made benchmark it beats ElevenLabs and other leading transcription tools, with the fewest word-timestamp errors
  2. Aug 2
    Four sizes: smallest 0.2B parameters (under 500 MB, runs without a GPU), largest 2B parameters (3 GB)

Cost to run