Parameter analysis for GLZ audio compression

The generalized Lempel-Ziv (GLZ) paradigm for lossy compression for audio relies upon the fact that music, in particular electronically generated sound, has a substantial level of repetitiveness within a single clip. Thus, GLZ compresses each of the overlapped and transformed windows of the audio using a linear combination of filtered past windows. Following the introduction of the basic GLZ algorithm, in this paper, we empirically analyze several key algorithm components. We analyze the design of simple band-pass filters used during the similarity search, we investigate the distributions of weights used to create the linear combinations, and finally, we explore how beat detection can be used to significantly speed up the similarity search process. We present preliminary experimental results on a benchmark of electronically generated musical pieces.