Yingyong Qi
Dept. of Speech and Hearing Sciences, Univ. of Arizona, Tucson, AZ 85721
Current methods of computing noise present in human voices depend upon being able to determine fundamental periods. In this paper, a method of estimating noise present in human voices that does not depend on the determination of the boundaries between fundamental periods is described. In this method, noise was computed as the uncorrelated component in the voice signal. The magnitude of the uncorrelated component was obtained by computing the average, maximum cross correlation between many pairs of short frames of voice signals. The method is evaluated using synthetic and natural voice signals. Results indicate that it is an effective tool for estimating the noise component in voice signals. [Work supported by NIH.]