A novel approach to speech segmentation using the wavelet transform

This paper presents a novel approach to segmenting speech using the wavelet transform. The wavelet transform produces a flexible decomposition of signals in both time domain and frequency domains which is useful in segmenting parts of speech containing fast varying changes, like stops, as well as slow varying ones, like vowels. Results using six wavelet parameters are found to be comparable to those obtained using sixteen mel-scale spectrum coefficients