A Method of Evaluating Characteristic Parameters of Unvoiced/Voiced Segment Based on Granular Computing

It is a difficult question How to evaluate a parameter´s significance of segmenting speech signal to unvoiced or voiced. An application of granular computing with this topic is presented in this paper. By analyzing eight parameters of speech signal, a decision table has been constructed depending on the granulation theory, which provides a method to find of the significance of parameters judging unvoiced/voiced decision. an then finally carries out the optimum scheme of choosing parameters set MFCC, wavelet cross-correlative, wavelet coefficient and short-time self-correlative to process speech segment and decision rules. It has the excellent performance of unvoiced/voiced segment.