Intelligibility in dysphonic speech: Landmark-based measures

Abstract

Speakers with voice disorders frequently report reduced intelligibility in ordinary communication situations. This effect is typically attributed to reduced loudness; however, other source/vocal tract interactions may be at work. The acoustic landmark theory of speech perception postulates that specific acoustic events, called “landmarks,” contain particularly salient information about acoustic cues used by listeners. The current study examined acoustic profiles of dysphonic speech with the publically available landmark-based automatic speech analysis software, SpeechMark™. In this study, we focused on burst landmarks, which aim to identify onsets and offsets of affricate/stop bursts. The study tested two hypotheses: (1) normal and dysphonic speech samples differ in the number of burst landmarks because laryngeal pathology affects the consistency of airflow control, and (2) the number of burst landmarks will correlate with cepstral peak prominence values, which have been shown to correlate well with perceptual judgment of dysphonia severity. Speech samples of 36 normal and 33 dysphonic speakers from KAY Elemetrics database of Disordered Voice were subjected to the analysis. Results will be discussed in the context of clinical assessment of intelligibility for dysphonic voices.