Objective methods of sample selection in acoustic analysis of voice

In acoustic voice analysis, the fact that reproducible methods of sample selection have not been defined impedes research study generalizability and clinical assessment of treatment efficacy. Because perturbation results differ along a single signal, this study sought to establish objective methods of sample selection by use of a moving window to determine the most stable regions of phonation.

Voice signals obtained from 21 patients affected by laryngeal conditions associated with Parkinson's disease were analyzed to study jitter, shimmer, signal-to-noise ratio, and correlation dimension parameters when various sample selection procedures were used. Objectively selected voice samples were chosen based upon 5%, 10%, and 20% variance from a signal's minimum perturbation value. The stability of these samples, defined by the standard deviations of the acoustic measurements, was compared to the stability of unselected samples and subjectively selected samples.

A significant decrease in standard deviation values of acoustic parameters was found in comparing the objectively selected samples (particularly those selected with 5% and 10% variance) to the subjectively selected and unselected samples.

These results suggest that the development of an objective sample selection method may have significant effects on the stability and reliability of acoustic voice measurements.