The analysis, synthesis, manipulation, and feature extraction of the electromyogram of lip muscles during speech

Abstract

In the initial phase of a research program to study neural control of the speech musculature a need arose for a powerful and convenient methodology to analyze, synthesize, and manipulate electromyographic (EMG) signals. Various techniques were considered until it was determined that a signal processing algorithm originally developed for handling the acoustic speech signal was also applicable to electromyographic signals. This article will discuss the use of a certain class of wavefunctions and how they seem to be a particularly powerful analytical tool in processing electromyographic data. The electromyograms of selected motor units from three muscles of the lower lip, the Orbiculares Oris Inferior, the Depressor Labii Inferioris, and the Mentalis, were studied by the use of the Hanning Cosine Modulated (HCM) wavefunction, time domain model. Using digital processing techniques, an EMG signal is transformed by an analysis procedure into a set of time domain based HCM parameter lists. To verify the analysis completeness, the HCM parameter lists are used by a synthesis procedure to generate a synthetic form of the orginal EMG signal. A comparison shows an exact correspondence between the original and synthetic signals. A Feature Extraction procedure may then be used to extract information of interest from the HCM parameter lists. The HCM parameter lists may also be modified so that the EMG signal may be investigated through an Analysis through Synthesis approach. A summary of the HCM signal processing algorithm is presented. Then through the use of illustrative examples, it is shown that the HCM parameter lists provide a general purpose data base. This data base may then be used for simulation or Synthesis of the EMG signal, diagnostic studies, time and frequency domain analysis, and other important activities.