Harmonics-to-noise ratio as an index of the degree of hoarseness

TLDR

Hoarseness is traditionally judged by how much noise replaces harmonic structure in a vowel spectrogram, but this visual assessment is subjective. This study aimed to create the harmonics‑to‑noise (H/N) ratio as an objective, quantitative measure of hoarseness. The H/N ratio was computed by averaging 50 pitch periods of a sustained /a/ vowel, taking the energy of the mean waveform as H and the mean energy of the period differences as N, and applying this to 42 normal and 41 hoarse recordings while experts rated spectrogram noise levels. The H/N ratio correlated highly with expert spectrogram ratings (rank correlation 0.849) and proved useful for quantitatively evaluating hoarseness treatment outcomes.

Abstract

Degree of hoarseness can be evaluated by judging the extent to which noise replaces the harmonic structure in the spectrogram of a sustained vowel. However, this visual method is subjective. The present study was undertaken to develop the harmonics-to-noise (H/N) ratio as an objective and quantitative evaluation of the degree of hoarseness. The computation is conceptually straightforward; 50 consecutive pitch periods of a sustained vowel /a/ are averaged; H is the energy of the averaged waveform, while N is the mean energy of the differences between the individual periods and the averaged waveform. Recordings of 42 normal voices and 41 samples with varying degrees of hoarseness were analyzed. Two experts rated the spectrogram of each voice sample, based on the amount of noise relative to that of the harmonic component. The results showed a highly significant agreement (the rank correlation coefficient = 0.849) between H/N calculations and the subjective evaluations of the spectrograms. The H/N ratio also proved useful in quantitatively assessing the results of treatment for hoarseness.