Text-Independent Phone Segmentation Method Using Gaussian Function

In this paper, an effective method is proposed for the automatic phone segmentation of speech signal without using prior information about the transcript of utterance. The spectral change is used as the criterion for hypothesizing the phone boundary. Gaussian function can be used to measure the similarity of two vectors. Then a dissimilarity function is derived from the Gaussian function to measure the variation of speech spectra between mean feature vectors before and after the considered location. The peaks in the dissimilarity curve indicate locations of phone boundaries. Experiments on the TIMIT corpus show that the proposed method is more accurate than previous methods.