The USC Andrew and Erna Viterbi School of Engineering USC Signal and Image Processing Institute USC Ming Hsieh Department of Electrical and Computer Engineering University of Southern California

Technical Report USC-SIPI-196

“Cumulant-Based Adaptive Analysis of Speech Signals”

by Mithat C. Dogan and Jerry M. Mendel

January 1992

This report describes a speech processing method consisting of an adaptive predictor, a voicing decision (V/UV), and a pitch period estimator. The focus of this report is on robust detection of speech state and estimation of pitch period. This is accomplished by observing the behavior of an adaptive predictor which processes the speech signal. Higher-order-statistical analysis is proposed for discrimination of speech states. Comparing the energy of the original speech signal with that of the prediction-error residual yields the decision method. Both covariance and cumulant-based prediction methods are investigated and the latter is shown to be a more robust way of making (V/UV) decision. Pitch estimation is accomplished by using correlation-based approaches that operate on the energy estimate of the cumulant-based prediction residual rather than the original speech signal. Pitch estimation by our method yields better performance than currently existing batch procedures.

To download the report in PDF format click here: USC-SIPI-196.pdf (1.1Mb)