Skip to content

v1.48.0

Choose a tag to compare

@sblattj sblattj released this 21 Nov 01:26
· 13 commits to main since this release

Opt-in NVIDIA Parakeet speech-to-text for whosaid live. Whisper stays the default for live and batch.

Added

  • #235: whosaid live --asr parakeet (or WHOSAID_LIVE_ASR=parakeet) transcribes live chunks with parakeet-mlx (mlx-community/parakeet-tdt-0.6b-v2, or WHOSAID_LIVE_PARAKEET_MODEL). Faster per chunk than turbo, and word times always on, so a speaker change inside a line splits at a word. The workspace's [transcribe] vocabulary (--ws, else $WHOSAID_WORKSPACE) plus --vocabulary TERM is boosted on every chunk (beam 5, weight 1.0; on by default when there are terms, --no-boost / WHOSAID_LIVE_BOOST=0 for greedy). The launcher adds parakeet-mlx==0.7.0 to the live uv env only when Parakeet is picked; pip installs use the whosaid[parakeet] extra, and a missing package exits 2 naming it. --model is the chosen backend's model; WHOSAID_LIVE_MODEL stays Whisper's.
  • test/asr_bench/bench_asr.py: Whisper vs Parakeet on your own recording (whosaid-shaped transcript, term spelling, live chunk timing).

Fixes #235