Design of Low-Complex Linear-Phase Non-uniform Filter Bank to Realize Wavelet Approximation of Bark Frequency Partitioning for Real-Time Applications

Abstract

This paper proposes the realization of perceptual wavelet approximation of Bark scale using a linear-phase integer decimated non-uniform filter bank (IDNUFB), designed by merging the channels of a partially cosine-modulated uniform filter bank. In effect, the channels with different sampling factors of the IDNUFB can be derived from the same prototype filter using the proposed method of design. Also, the proposed IDNUFB is employed for the realization of Bark frequency partitioning. The proposed method of IDNUFB design using PCM and merging is found to have linear-phase property and reduced hardware complexity, when compared to the existing filter bank-based methods for Bark frequency partitioning.