OpenVINO Biometric Extraction Pipelines (Face & Voice) (US-4.2) - 5 SP

[sainijit]

User story
As a machine learning engineer, I want to implement OpenVINO face processing and ECAPA-TDNN voice processing pipelines, so we can generate biometric embeddings.

Affected Components / Files

• identity-service/

Key Technical Tasks

• Ingest video/image frames and run OpenVINO models (face-detection-retail-0005, landmarks-regression-retail-0009, face-reidentification-retail-0095).
• Ingest audio PCM buffers and compute 192-d ECAPA-TDNN vectors.

Acceptance Criteria

• High-quality face vectors (256-d) and speaker vectors (192-d) are successfully computed within latency parameters (< 200 ms).

Estimate: 5 SP