An adaptive algorithm for the automatic segmentation of continuous stuttered speech

This paper presents an adaptive algorithm for the automatic segmentation of continuous speech into voiced, unvoiced, and silence regions (V/U/S). The procedure presented is based on the use of a digital filter bank with control parameters established using formant frequencies of voiced segments of speech. This procedure makes it possible to separate continuous speech into syllable like structures. The use of these speech segments makes possible the determination of stuttered speech. Special care is given to the handling of prolongations and repetitions (audible and silent) to insure they are properly handled by the segmenting algorithm. Segmentation experiments have been conducted on male and female subjects to test the reliability of the system at segmenting fluent and stuttered continuous speech.