SINUSOIDAL WAVE CONVOLUTION MODEL PARAMETER ESTIMATING METHOD AND THE SOUND SOURCE ISOLATING METHOD USING THE SAME

摘要

PROBLEM TO BE SOLVED: To derive a new estimation algorithm for calculating a most likelihood parameter of a sinusoidal wave convolution model. SOLUTION: A parameter is estimated when an input signal is expressed with the sinusoidal wave convolution model. At first, integral of an square error of the input signal and the sinusoidal wave convolution model is set as an objective functionΦ(Θ). Then, an auxiliary functionΦ+(Θ, m) in which the objective functionΦ(Θ)≤the auxiliary functionΦ+(Θ, m) for the objective functionΦ(Θ), and its auxiliary variable m are calculated. A step for updating the auxiliary variable m and a step for updating the parameterΘare executed by turns. This is repeated until an update amount or the parameterΘbecomes lower than a prescribed value. The parameterΘof the sinusoidal wave convolution model is calculated by sequentially updating the parameter by using the auxiliary function. COPYRIGHT: (C)2009,JPO&INPIT