Automatic determination of pauses in speech for classification of stuttering disorder

Abstract

An algorithm to automatically identify segments of silence or speech is presented. The algorithm was developed to measure the silence periods in spontaneous and read speech. These silence periods are one of the parameters used to know the degree of severity of stuttered speech. For this purpose the three longer disfluent events (pauses or other disfluent events) and also the percentage of silence are useful. The algorithm is based on the evaluation of the energy and the zero crossing rate of the signal compared to the threshold values previously determined in silence. One experiment with eight subjects is described using the Stuttering Severity Instrument for Children and Adults – SSI and the percentage of silence in speech. It was concluded that the percentage of silence is good enough to separate stuttered from the normal speech but alone is not capable of measuring the degree of severity of the stuttered speech.