Physics / mechanism
Hearables combine in-ear or on-ear transducers with signal processing to capture biosignals — EMG from facial/jaw muscles, EEG from temporal cortex, ultrasonic Doppler of articulatory movement, or bone-conducted vibration — and decode intended speech without audible vocalisation. Silent-speech interfaces (SSIs) typically chain a sensor array, analog front-end ASIC, edge ML inference (transformer or RNN), and BLE/UWB radio. Key metrics: word-error rate (<10% is competitive; best lab demos hit ~5% on constrained vocabularies), inference latency (<200 ms end-to-end for conversational use), and power budget (<5 mW continuous for all-day wear). Current SoA: Meta/CTRL-Labs wrist EMG, Whisper Chip integration in TWS, Halo Neuroscience ear EEG. Commercialisation bottleneck is electrode-skin contact variability and per-user calibration overhead.
Competitive landscape
Companies using
Connected ideas
Sources
Frontier (open questions)
- To be added.