Automatic Detection of Pharyngeal Fricatives in Cleft Palate Speech

摘要

Pharyngeal fricatives are produced as substitutes of normal fricatives by speakers with cleft palate. This chapter introduces a method of extracting acoustic features of pharyngeal fricatives by means of one-third octave analysis. The results show that the pharyngeal fricatives have higher amplitude below 4,000 Hz and lower amplitude above 4,000 Hz than the normal fricatives. Amplitudes of eight one-third octave bands are extracted to represent the acoustic features of pharyngeal fricatives. A pattern classifier based on logistic regression is also proposed here and obtains acceptable classification accuracy (90 %) when these acoustic features are used as the input factors. Detection of the pharyngeal fricatives is also automated in combination of such pattern classifier and an automatic selection of consonant component.