A new pitch synchronous V/U/M/N/S classification algorithm

In this paper, we study a pitch-synchronous V/U/M/N/S classification method. The decision making process is viewed as a pattern recognition problem. Two aspects of the algorithm are considered: feature selection and classifier type. The feature selection procedure is studied for identifying a set of features to make V/U/M/N/S classification. The classifier used is a vector quantization (VQ). The method provides the information that is useful for line segmentation of the speech signal and also provides a classification rate as good as previous pitch-asynchronous algorithms. Five actual sentences spoken by the six speakers, three male and female, are tested with the proposed method. It shows the comparative accuracy for V/U/M/N/S decision with pitch asynchronous algorithm