A new method of speech model based on periodic expanded

Abstract

A novel method is proposed for speech model, which is used for speech transformation and speech recognition. This method divides segmentations of a speech into an adaptive segment, in which its speech wave is integrated and has approximate periods. A speech is divided into voiced speech and unvoiced speech by voice activity detection, which is widely used in speech dividing. This speech model mainly deals with sonant, which is divided into approximate periodical waves in our currently work. Segment's parameter contains primary periods' parameter, secondary periods' parameter and the length of the segment. Primary periodic signal's expandability is to form speech segmentation. The paper also gives some results of experiments to show the effectively of the model. The accuracy and few coefficients of model are due to using Morlet wavelet to extract primary periods.