Sesap Audio Companion [Windows Proven]

| Aspect | Implementation | Metrics | |--------|----------------|---------| | | Fixed delay‑and‑sum (DS) with adaptive nulling | 15 dB SNR improvement vs. omnidirectional mic | | S2T Model | Conformer‑Tiny (12 M) + LM‑Lite (4 M) | WER = 7.2 % on in‑the‑wild dataset (SESAP‑Eval) | | Intent Classifier | MobileBERT‑Tiny (3 M) | Top‑1 accuracy = 94 % (10 intent categories) | | Voice Synthesis | Diffusion‑TTS (3 M) | MOS = 4.3/5 (naturalness) | | Latency | End‑to‑end (wake → response) | 120 ms average (99‑th percentile 210 ms) | | Power | Continuous listening + inference | 12 mW avg; 24‑h battery life (typical day) |

Stay up-to-date with the latest news, updates, and tips on how to get the most out of Sesap Audio Companion. Follow us on social media or subscribe to our blog for exclusive content, promotions, and more. sesap audio companion