EchoWhisper: Exploring an Acoustic-based Silent Speech Interface for Smartphone Users

on YouTube ↗06m

" summary

Smartphone emits a 20 kHz ultrasound and uses the dual microphones to record micro-Doppler reflections from mouth and tongue motions, then applies noise cancellation, segmentation, spectrogram features and a custom Echo-Net model. System recognizes 45 daily words with 91% accuracy (reported 8.33% WER) from five users on three phones, works best at 3–7 cm, and shows robustness to orientation, ambient noise, body motion, and nearby 20 kHz interference.

AI-written summary · about

" chapters (9)

1. 0:00Introduction
2. 1:26Interesting Findings
3. 2:11Proposed EchoWhisper
4. 3:08Experiments
5. 3:29Recognition Performance
6. 4:04Usability Analysis
7. 4:26Robotness Analysis
8. 4:51Distance Analysis
9. 5:34Conclusion