Psychoacoustically guided midfrequency band-limiting improves the diagnostic utility of classical acoustic measures in dysphonia

Abstract

Abstract Human hearing is most sensitive around 3 kHz, where small changes in harmonic structure and noise are particularly noticeable. However, traditional or standard acoustic measures of dysphonia—jitter, shimmer, harmonics-to-noise ratio, and cepstral peak prominence smoothed—are usually calculated from full-band signals and do not account for this psychoacoustic weighting. Consequently, their diagnostic accuracy is often reduced in connected speech and cases of mild dysphonia, which are common clinical presentations. We tested a simple, low-cost preprocessing strategy that limited analysis to the 2–4 kHz frequency band before computing these measures. Using sustained vowels and connected speech from Japanese speakers across a wide range of dysphonia severities, we compared band-limited and full-band results against the auditory-perceptual judgments of breathiness, roughness, and the grade of hoarseness. Across utterance and perceptual dimensions, band-limiting consistently improved diagnostic performance, with particularly notable gains in mild dysphonia. These findings suggest that restricting analysis to the high-sensitivity auditory range reveals microperturbations that are masked by dominant low-frequency energy in full-band signals. This psychoacoustically guided approach enhances the clinical usefulness of traditional acoustic measures without requiring complex modeling. It is device-independent and suitable for telemedicine and real-world voice-assessment workflows.