Feature evaluation for recognition of unspoken small vocabulary using facial EMG

Abstract

This paper presents a novel technique for feature evaluation of some common vocabulary terms employed during the process of speech without producing audio signals using facial EMG. This paper aims to present an effective mechanism to measure the relative activity of the facial muscles related to speech production for person’s suffering from speech-related disabilities. The features were extracted from raw EMG signals of four important facial muscles (zygomaticus major, orbicularis oris inferior, levator labii superioris, anterior belly of diagastric) using root mean square (RMS) values for both genders. These values presented the relative activity of the four chosen muscles showing significant changes only in orbicularis oris inferior (OOI) muscle both in males and females during the process of speech production. Further, the fast Fourier transform (FFT) was used to validate this hypothesis and verify the results obtained. From these results, it may concluded that this muscle is the most active and useful in the speech production. This evaluation will be very helpful in designing speech prosthesis for the people with speech disabilities in near future.