Phoneme recognition using wavelet based features

Abstract

This paper proposes the use of the discrete wavelet transform (DWT) for the extraction of features from phonemes. Instead of using the short time Fourier transform for feature extraction a new set of features is obtained from the DWT. The new set of features overcomes the previously reported problem of shift variance in DWT based features. Training and test samples of the phonemes were obtained from the TIMIT database. To account for the fast changes in the phonemes, the features were calculated for different phoneme durations and the performance was compared. For the classification of the phonemes, two different classifiers were used, based on linear discriminant analysis and multi-layer perceptron.