MODEL ESTIMATION SYSTEM, SOUND SOURCE SEPARATION SYSTEM, MODEL ESTIMATION METHOD, SOUND SOURCE SEPARATION METHOD, AND PROGRAM

摘要

PROBLEM TO BE SOLVED: To separate sound sources without causing a permutation problem and performing post-processing of permutation solution.;SOLUTION: A feature extraction unit 2 extracts a feature vector x_τω associated with a measurement signal vector y_τω including time frequency components of each of measurement signals outputted from M microphones respectively which have measured a mixed signal into which signals from K sound sources are mixed. A model estimation unit 3 adapts the feature vector x_τω to a predetermined probability model, and estimates a model parameter of the probability model using an evaluation function that gives a higher evaluation value according to a degree to which time-sequential data items of likelihoods of the respective sound sources are synchronized with frequency bins. The probability model is a mixed model expressed with a weighted sum of distributions of the feature vector x_τω concerning the sound sources. A mixed weight for the probability model is a weight that depends on a time frame τ but does not depend on an angular frequency ω. The model parameter of the probability model is a parameter for the mixed weight and the distributions of the feature vector x_τω concerning the sound sources.;COPYRIGHT: (C)2015,JPO&INPIT