[Task] speaker-diarization model support

Overview

Speaker diarization models answer the question "who spoke when?" by segmenting and clustering an audio recording by speaker identity. The pyannote suite includes end-to-end diarization pipelines (3.x generation) as well as individual segmentation and overlapped-speech detection components that underpin them.

Agent Scenarios

• Meeting notes agent: produce transcripts annotated with speaker turns ("Alice: ...", "Bob: ...") for post-meeting summaries and action item extraction
• Call center analytics agent: separate agent voice from customer voice to compute per-speaker metrics (talk ratio, interruption rate, sentiment)
• Podcast / media production agent: auto-generate speaker-labeled chapters or subtitles for multi-speaker recordings
• Legal / compliance agent: create speaker-attributed transcripts of depositions or earnings calls for searchable archiving

ModelKit Integration

Models must pass the full wmk pipeline on all EPs:

wmk config → wmk build (ONNX export) → wmk perf → wmk eval

Acceptance Criteria

• pyannote/speaker-diarization-3.1
• pyannote/speaker-diarization-3.0
• pyannote/speaker-diarization-community-1
• pyannote/speaker-diarization
• pyannote/segmentation-3.0
• pyannote/segmentation
• pyannote/overlapped-speech-detection

Start by locating the existing ModelKit model integrations and the entry points for the wmk pipeline. Run wmk config, wmk build, wmk perf, and wmk eval for each listed pyannote model; done means all seven models pass the full pipeline.