Relevance of time-frequency features for phonetic and speaker-channel classification

The mutual information concept is used to study the distribution of speech information in frequency and in time. The main focus is on the information that is relevant for phonetic classification. A large database of hand-labe led fluent speech is used to (a) compute the mutual information (MI) betwee n a phonetic classification variable and one spectral feature variable in t he time-frequency plane, and (b) compute the joint mutual information (JMI) between the phonetic classification variable and two feature variables in the time-frequency plane. The MI and the JMI of the feature variables are u sed as relevance measures to select inputs for phonetic classifiers. Multi- layer perceptron (MLP) classifiers with one or two inputs are trained to re cognize phonemes to examine the effectiveness of the input selection method based on the MI and the JMI, To analyze the non-linguistic sources of vari ability, we use speaker-channel labels to represent different speakers and different telephone channels and estimate the MI between the speaker-channe l variable and one or two feature variables. (C) 2000 Elsevier Science B.V. All rights reserved.