A source localizer using a steering vector estimator based on an online complex Gaussian mixture model using recursive least squares

摘要

The present invention relates to a method for estimating a sound source direction. The sound source direction estimation method relates to a method of estimating a steering vector corresponding to the direction of a sound source by using input signals input from a plurality of microphones, wherein (a) a signal mixed with voice and noise is a CGMM Assuming that the model is followed, obtaining a log likelihood function expressed as a cost function; (b) obtaining an equation for estimating parameter values of the log likelihood function in a direction recursively maximizing the value of the log likelihood function every time frame; (c) recursively estimating a covariance matrix every time frame; (d) recursively estimating a steering vector using the estimated covariance matrix; To estimate the direction vector of the sound source from the input signals.