Combining temporal and cepstral features for the automatic perceptual categorization of disordered connected speech

摘要

The objective of the presentation is to report experiments involving the automatic classification of disordered connected speech into multiple (modal, moderately hoarse, severely hoarse) categories. Support vector machines, used for the classification, have been fed with temporal signal-to-dysperiodicity ratios, the first rahmonic amplitude as well as mel-frequency cepstral coefficients. The signal-to-dysperiodicity ratio complements the first rahmonic amplitude when categorizing voice samples according to the degree of hoarseness yielding 77% of correct classification.