Measuring periodicity perturbations in pathological voice: general-purpose software vs. Custom-tailored methods

Abstract

The acoustic measurement of the severity of the symptoms present in pathological voice is an active research area, for being inexpensive and non invasive. Periodicity perturbations are among the most frequently used, requiring the previous extraction of the individual glottal pulse boundaries. In this paper we explore the performances of methods for detecting glottal pulse boundaries as implemented by freely available software (Praat, intended for phonetic studies) vs. a research-grade pulse cycle detector (reported as a superresolution method). We compare the sequences of pulse markers as obtained by two of Praat’s internal implementations and the super-resolution method against the hand-marked reference sequence in a dataset of pathological sustained vowels from a well-known database. A group of performance measures is extracted from this comparison, using a Dynamic-Time Warping alignment procedure. The measures obtained show the pros and cons of each alternative. Researchers and clinicians must be aware of the benefits of selecting either approach.