Cochlea-scaled entropy predicts intelligibility of Mandarin Chinese sentences

Abstract

Cochlea-scaled entropy (CSE) is a measure of psychoacoustic and physiologic change in complex acoustic signals such as speech. Based upon information-theoretic considerations, this measure of signal unpredictability as potential information has been successful in predicting English sentence intelligibility. CSE is based on Euclidean distances between outputs from 33 relatively broad filters across spectral slices (16-ms) of speech (0-8 kHz), but it does not explicitly measure fundamental frequency (f0) or changes in f0. Due to this potential shortcoming, the present study addresses whether CSE effectively predicts speech intelligibility in a tone language for which changes in f0 are essential to phonetic and semantic distinctions. Twenty-five native-Mandarin listeners transcribed Mandarin sentences in which consonant-length (80-ms) and vowel-length (112-ms) segments with either high or low CSE were replaced with speech-shaped noise. Results demonstrate that CSE is a significant predictor of speech intelligibility for Mandarin sentences, even though CSE does not explicitly provide information about f0. It is possible that some f0 changes were reflected in changes in filter outputs, and listeners may have used f0 in maintained signal intervals to interpolate contours in replaced segments. © 2013 Acoustical Society of America.