VOICE ANALYZING METHOD, DEVICE THEREOF, VOICE ANALYSIS PROGRAM AND RECORDING MEDIUM THEREOF

摘要

PROBLEM TO BE SOLVED: To accurately calculate an articulation movement trajectory at voice speaking and spoken contents from a given voice. SOLUTION: A state sequence in which the appearance probability of an input voice becomes maximum is determined by using a probabilitical dynamic model representing the time series (articulation movement trajectory) of the positions of a vocal organ at voice speaking and a hidden Markov model composed of a voice generation model for determining the occurrence probability of a voice spectrum (voice parameter) to the position (articulation parameter) of the vocal organ, and an articulation parameter vector sequence wherein the occurrence probability of the input voice becomes maximum to the state sequence, that is, an articulation trajectory is created. The phoneme information of the model in each state of the determined state sequence is outputted to perform voice recognition. COPYRIGHT: (C)2003,JPO