The temporal sampling problem in electromyographic studies of speech musculature

Electromyographic signals are often averaged on the assumption that they represent a sum of a stable biological signal plus random noise. A muscle may be said to do the same thing on two occasions if the activity of all motor units involved is the same on both occasions. Since the activity of all units in a muscle cannot be monitored simultaneously [Cooper and Folkins, J. Acoust. Soc. Am. Suppl. I 70, S77 (1981)], we rely on the interference pattern to provide a signal representative of the activity of many units. The simultaneous monitoring of activity at multiple electrode sites allows the observation of a larger proportion of the occurring bioelectric activity than does the customary use of one derivation per muscle. Signals from three bipolar fine wire derivations high, medium, and low in the same palatoglossus muscle of speakers of American English were amplified, rectified, and integrated. The degree of similarity between pairs of simultaneous signals varied significantly across tokens of the same word, although it did not vary as greatly across tokens as it did across words. Consequently the pattern of muscle activity is not always constant across tokens of the same word, an effect which would be overlooked by averaging of tokens. [Work supported in part by NIDR.]