Laryngeal‐Disorder Detection through Speech Analysis

This talk reports on a study into the feasibility of using automatic voice analysis for the detection of laryngeal disorder. Over 200 normal and pathological subjects were examined and had their voices recorded at the Massachusetts Eye and Ear Infirmary. The first stage of the automatic analysis was a highly accurate peak detection and pitch tracking on a throat microphone signal [CRYSTAL, T. H., and C. L. JACKSON (1970), J. Acoust. Soc. Amer. 48, 118(A)]. Lieberman's pitch perturbation technique [P. LIEBERMAN (1963). J. Acoust. Soc. Amer. 35, 334–353] was extended to amplitudes. Our analysis was made on “jitter” defined as the difference between smoothed data and actual data. The normalized mean‐square jitter was then used as an index of vocal quality. A set of such indices, obtained by using different smoothing algorithms, was used in a multivariate discriminant analysis of the data. Linear discrimination on the training data allowed the separation of 83% of the male pathologicals when the decision thre...