Constant Q-value filter banks with spectral analysis using LMS algorithm

Abstract

In spectral analysis of temporally varying signals, constant Q-value filter banks using the short time spectral analysis method is known to be effective because the frame length can be changed freely with the method depending on the frequency. We have already reported an analysis with the LMS algorithm. But, the constant Q-value filter banks couldn't be constructed by this method because the parameters to control the stability and convergence factor of systems were scalar values. In order to realize the constant Q-value filter banks with this spectral analysis method, it is necessary to change the parameters of the stability and convergence factor from scalar to diagonal matrix. We first derive a LMS algorithm for this method. Then the characteristics of this spectral analysis method were found by extending the discussion to stability conditions of the method. Furthermore, the effectiveness of the proposed method for transfer performance was experimentally demonstrated by applying it to the analysis of human voices and acoustic waves generated by musical instruments.