Acoustical analysis for voice disorders: An attempt towards clinical applications

Glottal sound waveforms were derived by an inverse filtering method from sustained vowels uttered by 217 patients suffering from 15 kinds of laryngeal diseases and by 20 normal speakers. Various parameters which indicate individual characteristics were extracted from autocorrelation function, correlograms of time series of fundamental period and maximum amplitude in each fundamental period, and spectral envelope of the glottal sound waveform. Then, dependency on each other among acoustical parameters was examined by principal component analysis. Meanwhile, medical parameters utilized in diagnosis were evaluated quantitatively, and the relation between those and the acoustical parameters was examined also by principal‐component analysis. Combining those results, more than 10 acoustical parameters which seemed to be useful for diagnosis and to be rather independent of each other were selected. Using canonical analysis the axes which discriminate the kind of disease best were obtained based on those acoustical parameters of the patients of several diseases and normal speakers, and the degree of discrimination and the contribution of each acoustical parameters to each axis were examined.