Examining the Acoustic Feature Differences Between Voiced and Whispered Speech

Hannah Browning, Helen A. Boyd-Pratt, and Jeremy J. Donai
Department of Communication Sciences and Disorders, West Virginia University, Morgantown, WV 26506

Presentation Category: Health Sciences (Poster Presentation)

Student’s Major: Communication Sciences and Disorders

The purpose of this study is to examine acoustic feature differences between voiced and whispered speech. Whispered speech is an aspect of the speech signal that is not readily studied. A better understanding of the acoustic properties and differences of whispered speech compared to normal speech may allow for better automated speech and speaker recognition when the speaker is whispering. In this study, we collected and analyzed audio recordings of six female and six male subjects that spoke English. Each subject read five lists of ten sentences each in a normal speaking voice and again in a whispered speaking voice. Commonly used IEEE lists for the sentences of use in the study due to the consistency seen in other similar research studies. Spectral moments and spectral center of gravity measures were obtained. Preliminary results using stepwise discriminant function analyses showed the first spectral moment (i.e., mean) was somewhat useful for classifying mode of production (voiced vs whispered) for male (77%) and female (68%) signals. Additional features are currently being explored to improve classification accuracy.

Funding:

Program/mechanism supporting research/creative efforts: a West Virginia SURE program