cjm-capability-pyannote 0.0.3

A pyannote.audio speaker-diarization capability for the cjm-substrate runtime — source audio in, anonymous time-ranged speaker turns out (community-1 pipeline by default; HF-gated). Identity stays downstream human judgment: the correction TUI's assignment lane.

cjm-capability-pyannote

Speaker-diarization capability backed by pyannote.audio (community-1) — full-source decoded-PCM audio → time-ranged anonymous speaker turns (SPEAKER_00-style cluster labels, never identities), implementing the speaker-diarization adapter interface; pure compute — caching lives in the generic adapter. Default-on rung in the transcription workflow; the correction TUI's assign lane consumes the turns.

Modules

• cjm_capability_pyannote.__init__
• cjm_capability_pyannote.capability — Speaker-diarization tool capability using pyannote.audio (session-D DECs 18d7de80 + d6df3a8e): source audio in, ANONYMOUS time-ranged speaker turns out — the machine half of speaker assignment; identity stays the correction TUI's human lane (DEC 44afb2df).

API

cjm_capability_pyannote.capability

• PyannoteConfig class — Configuration for the pyannote diarization pipeline.
• PyannoteDiarizationCapability class — Speaker-diarization tool capability wrapping a pyannote.audio Pipeline.

Dependencies

Depends on: cjm-capability-primitives, cjm-substrate, cjm-substrate-torch-utils, pyannote.audio