Privacy controls implemented

This document summarizes implemented privacy-preserving transforms and storage/retention guidance.

Implemented transforms - Audio anonymization: anonymize_audio_add_noise(wave, snr_db) — adds Gaussian noise to reduce identifiability - Audio prefix masking: mask_audio_zero_first_seconds(wave, sample_rate, seconds) — zeroes first N seconds - Face blur: blur_frames(frames) — coarse per-frame Gaussian blur (uses OpenCV if available)

API usage - /process accepts flags: anonymize_audio, mask_audio_prefix_s, blur_faces - /submit accepts same flags; they are persisted with the job and applied by background workers

Data retention & storage - Default job store is JSON-file fallback; use secure Redis or DB with TTL for production - Implement automatic deletion of stored media and results after retention period

Legal / compliance note - These transforms do not guarantee legal anonymization (e.g., under GDPR). Consult privacy/legal experts for compliance-critical deployments.