Skip to content
DefendableDocs

Whisper Transcription

Installation

Section titled “Installation”
Terminal window
make install-whisper    # adds faster-whisper (~1.5GB model on first run)

Usage

Section titled “Usage”
Terminal window
streetchat transcribe --session latest

Designed default model: base.en · CPU mode · int8 compute. Tunable via .env.

Output

Section titled “Output”

Per-session transcript JSON with:

  • per-segment timestamps
  • speaker placeholder (diarization Phase 2)
  • per-segment confidence
  • language detection
  • duration · engine · model · generated_at

Paste fallback

Section titled “Paste fallback”

If faster-whisper is not installed (or transcription was done elsewhere) · the design falls back to ingest-transcript mode. The pipeline is intended to not require Whisper to function — paste always works.

🐝 Operator-grade · books and records · to the shed.