Unspoken vowel recognition using facial electromyogram.

The paper aims to identify speech using the facial muscle activity without the audio signals. The paper presents an effective technique that measures the relative muscle activity of the articulatory muscles. Five English vowels were used as recognition variables. This paper reports using moving root mean square (RMS) of surface electromyogram (SEMG) of four facial muscles to segment the signal and identify the start and end of the utterance. The RMS of the signal between the start and end markers was integrated and normalised. This represented the relative muscle activity of the four muscles. These were classified using back propagation neural network to identify the speech. The technique was successfully used to classify 5 vowels into three classes and was not sensitive to the variation in speed and the style of speaking of the different subjects. The results also show that this technique was suitable for classifying the 5 vowels into 5 classes when trained for each of the subjects. It is suggested that such a technology may be used for the user to give simple unvoiced commands when trained for the specific user.