A bounded-Q fast filter bank for audio signal analysis

Abstract

This paper describes a new spectral analysis tool, the bounded-Q fast filter bank (BQFFB), which consists of a high selectivity filter bank with a piecewise linear spacing between its channels. This approximately geometric representation yields relatively low computational cost and a suitable frequency resolution, thus making the BQFFB an efficient tool for musical analysis applications. A description of the BQFFB algorithm is presented along with a complete evaluation of its corresponding computational complexity. Computer simulations are then included to illustrate the BQFFB performance against other methods from the literature, such as the fast Fourier transform, the standard fast filter bank, and the constant-Q transform.