On determining EMG onset time.

Each of two methods was used to determine the onset time of EMG activity associated with speech gestures in two data sets. Each data set contained between 10 and 20 tokens (repetitions) of several utterances. Onset times were specified in relation to a reference point in the speech signal. In one method, measurements were made from the digitized waveforms for the individual tokens. Average onset time for each utterance type was calculated by averaging measurements from its tokens. In the other method, the ensemble average of the EMG signals from tokens of each utterance type was calculated after carefully aligning them at the reference point. Onset time was then measured for each utterance type from the ensemble-average waveform. Onset-time measurements from ensemble averages were consistently earlier than corresponding averages of individual-token measurements. For some utterances, the ensemble-average onset time was earlier than any individual-token measurement. These results confirm the value of ensemble averaging for reducing the inherent noise in EMG signals associated with speech gestures. [Work supported by NINCDS and BRSG.] [ABSTRACT FROM AUTHOR]