Assessing a Set of Glottal Features from Vocal Fold Biomechanics

This paper summarizes a statistical study of a set of glottal features with the ultimate aim of measuring their capacity to discriminate and detect vocal pathology. The study is concentrated in the analysis of relevance of a set of features obtained from the analysis of phonated speech, specifically an open vowel as /ah/. The speech signal was inversely filtered to obtain the glottal source, which on its turn was used to generate a set of 72 features, describing its biometrical and biomechanical properties, among others. The study of relevance is based on factorial analysis, parametrical and non-parametrical hypothesis tests and effect size analysis, with the aim of assessing the pathologic/normophonic condition of the speaker. The validation of the results is based on discriminant analysis. The conclusions allow establishing the most relevant features and feature families for pathological voice detection. High classification rates are obtained in many cases.