ASR Transcribe to Text

Transcribe audio/video to speaker-labeled text. Default pipeline (decoupled, WhisperX-style): Qwen3-ASR transcribes the full audio with context intact, mlx-whisper supplies a word-level timing lattice, pyannote supplies speaker segments, and an aligner merges the three — the audio is never cut before ASR, so transcription quality stays at full-audio fidelity.

Configuration persists in ${CLAUDE_PLUGIN_DATA}/config.json.