Preliminary Test of a Wireless Magnetic Tongue Tracking System for Silent Speech Interface

Abstract

A silent speech interface (SSI) converts non-audio information such as articulatory movement to text that can then drive a text-to-speech synthesizer. SSI has potential to assist the oral communication for individuals with impaired voice. In this paper, we present a prototype, wireless, permanent magnet articulography (PMA)-based tongue motion tracking system for SSI purpose. An array of magnetometers, embedded in this device, measures the changes in magnetic field induced by the movements of a magnetic tracer attached near the tongue tip. The magnetic field recordings are directly mapped to text by a deep neural network-based silent speech recognition algorithm. The recognition accuracy of target phonemes was evaluated in a speaker-specific task using a data set collected from six American English speakers. The average phoneme error rate was 37.3% and the word error rate was 32.1%. These encouraging results demonstrated the potential of our wireless device for silent speech interfaces.