281-line version with the person-identity DDL sketch, per-model license
audit table, VRAM/cost budget, GDPR Art. 9 analysis, and all primary-source
citations — replaces the 53-line executive summary.
Research: use PeerTube
native transcription (core since 6.2, whisper-ctranslate2, remote runners on
GPU nodes; large-v3-turbo MIT via model_path); pyannote community-1
diarization (CC-BY-4.0, clean); WhisperX fusion.
Face recognition license audit: NO accurate open FR model has clean weights
(InsightFace/SCRFD non-commercial, yolo-face GPL, AdaFace unstated) —
posture is detection+tracking+blur by default; identity matching opt-in,
human-review-gated, collective decision + DPIA required (Art. 9 biometric
data; automated FR over non-consenting third parties effectively unlawful).
DB DDL sketch: persons/person_face_embeddings/video_face_detections/
transcript_segments/video_speaker_person/identity_audit with review gating,
visibility tiers, erasure workflow.
