Physical AI Reads Spoken Digits Directly from Throat Vibrations—96.8% Recognition without Preprocessing
ICYS Research Fellow Dr. NISHIOKA
【Key Points of This Research】
This study demonstrates a speech-recognition system that seamlessly integrates sensing and AI processing. Throat-surface vibrations during speech are detected by a throat-mounted π-gel-electret mechanoelectric generator (MEG) sensor and directly fed into an artificial intelligence device: an ion-gel/graphene-based ion-gating reservoir (IGR).
Unlike conventional microphone-based speech recognition, the system does not require frequency-domain analysis, handcrafted feature extraction, or other digital preprocessing. Instead, the nonlinear ion–electron dynamics inside the material device physically transform raw biomechanical vibration signals into features suitable for classification.
The system achieved 96.8% accuracy in spoken-digit recognition, highlighting its potential for compact and low-power edge-AI applications, including noise-resilient wearable speech interfaces, assistive communication technologies, and spoof-resistant biometric authentication based on throat-vibration signatures.
【Title of the Paper】
Ion-Gating Reservoir Computing for Preprocessing-Free Speech Recognition from Throat Vibrationsr【Publication Details】
This research was presented by Dr. Daiki NISHIOKA, ICYS Research Fellow.
Published in ADVANCED ELECTRONIC MATERIALS on 26 April 2026.
DOI: 10.1002/aelm.202600006