Acoustic Differentiation of Laryngeal, Esophageal, and Tracheoesophageal Speech

Abstract

This investigation was designed to determine if a multivariate acoustic classifier could effectively discriminate group membership for 15 tracheoesophageal, esophageal, and laryngeal speakers. Seven intensity, 10 frequency, and 13 duration measures were quantified from recorded voice samples. Using principal components analysis, a subset of the 13 least redundant acoustic and temporal measures was systematically selected from the 30 original measures and analyzed singly and jointly in terms of its ability to discriminate among the three speaker groups. Discriminant function analysis revealed perfect categorization of the 45 subjects, indicating that the three methods of speech production are acoustically and temporally distinct from one another. The relative importance of the selected variables which, in combination, significantly differentiated the three groups is discussed in relation to physiologic differences among groups and clinical application for postlaryngectomy vocal rehabilitation.