Interpretation: PSNR is far below conventional thresholds (≥ 30 dB). Nevertheless, the I‑frame carries a crude “key image” that the brain can extrapolate across subsequent frames. | Measure | Mean Score (1‑5) | |---------|------------------| | Overall intelligibility | 2.8 | | Acceptable for “audio‑only” context (visual as supplementary) | 3.4 | | Noticeable artifacts (blocking, smearing) | 4.7 |
Participants reported that (e.g., a static face) remained recognizable, while high‑motion sequences became indistinguishable. 4.3. LVCI The composite LVCI for the full 4‑minute clip was 0.31 , placing the stream in the “barely usable” category (thresholds: > 0.6 = good, 0.4‑0.6 = moderate, < 0.4 = poor). 4.4. Comparative Bandwidth Efficiency | Encoder (target) | Achieved PSNR (dB) | Bitrate (bps) | |------------------|--------------------|--------------| | H.263 (Katlynshine) | 9.3 | 720 | | H.264 (baseline) at 720 bps | 7.5 | 720 | | AV1 (experimental) at 720 bps | 10.2 | 720 | | VVC (theoretical) at 720 bps | 11.5 | 720 | Stickam Katlynshine 720bps Avi
OpenAI Research Team (ChatGPT)