パルス計数法による筋電図からの神経指令の抽出

抄録

Merits of a pulse counting method for extraction of the motor command, or the control signal of muscle contraction, from electromyogram (hereafter EMG) recorded with a pair of hooked-wire electrodes are discussed. First, based on the consideration of the mechanisms of the transmission of the motor command and the observation of EMG, it is discussed that the motor command of speech gesture can be extracted more accurately by the pulse counting method, that is, by detecting as, many neural impulses as possible, and extracting the analog time pattern of change in the pulse density, rather than by the rectifying-smoothing method which has been used in experimental phonetics. Then the optimum condition of the pulse counting method is examined and it is compared quantitatively with the rectifying-smoothing method by computer simulation. The material used here is the EMG of the thyrohyoid muscle recorded during the utterances of a nonsense mora sequence with four accentuation types. The accuracy of extraction of the motor command of accentuation type for each utterance is evaluated by an average correlation ratio which indicates both degree of similarity in an accentuation type and degree of dissimilarity among accentuation types. As the results, it is shown, in the pulse counting method with some modification for reducing the miscounting, that the optimum threshold for the detection of pulse train lies at 1/40 of the maximum pulse amplitude, and the optimum width of time window for extraction of analog time pattern of change in the pulse density is about 170 ms which is. somewhat shorter than the duration of one mora. Then it is also shown that the modified pulse counting method with the optimum threshold and width of time window gives better results compared with the rectifying-smoothing method. Finally, it is demonstrated that the modified pulse counting method can be carried out in real time by an electronic circuit.