VOICE PROCESSOR AND KARAOKE DEVICE

摘要

PROBLEM TO BE SOLVED: To perform a voice process for making an object voice to be aligned correspond to an input voice in real time with a small storage capacity. SOLUTION: Data obtained by analyzing an object voice in frame unit sectioned by specific time unit are stored and an input voice is also analyzed in frame unit sectioned by similar time unit. A hidden Markov model is generated according to a phoneme dictionary and the frame of an object person corresponding to the time of a frame of the input voice is specified in phoneme unit by using one-pass viterbi algorithm to match time relation. Data corresponding to the frame wherein the phonemes of the input voice and object voice are made to match with each other are composed.