Myo-To-Speech - Evolving Fuzzy-Neural Network Prediction of Speech Utterances from Myoelectric Signals

Abstract

Voice rehabilitation is needed after several diseases, when a subject’s vocal ability is compromised by surgical interference or removal of phonation organs (e.g. the larynx), by neural degeneration or by neurological injury to the motor component of the motor-speech system in the phonation area of the brain (e.g. dysarthria in Parkinson disease). A novel approach to voice rehabilitation consists of predicting the phonetic control sequence of the voice-production apparatus (larynx, tongue, etc.) by drawing inferences on the basis of myoelectric (EMG) signals captured by a set of contact electrodes, applied to the neck area of a subject with important phonatory alteration (e.g. laryngectomised) and intact neural control. The inference paradigm is based on an EFuNN (Evolving Fuzzy Neural Network) that has been trained to use the sampled EMG signal to predict the phoneme that corresponds to the motor control of the sublingual muscle movements monitored at phonation time. A phoneme-to-speech synthesizer generates audio output corresponding to the utterance the subject has tried to enunciate.