Recognition of Single Japanese Sounds using sEMG Signals and Membership Functions

Abstract

We report a method for recognizing single Japanese sounds including a consonant using surface electromyography (sEMG) signals and membership functions. We focused on six features of the waveforms of four muscles (a total of 24 indexes) and examined the consistency between the sEMG signals and the distribution of each index regarded as membership functions. We also introduced π-type membership functions and confirmed an improvement in the recognition rate (average recognition rate 62.3%, recognition of a series of Japanese syllables 99.1%) by varying the parameter values. These methods allow greater recognition of single Japanese sounds by users.