Parameter analysis for the generalized LZ compression of audio

Abstract

Summary form only given. We introduced (Kirovski and Landau (2004)) a memory-based model of the source signal, which explores multimedia repetitiveness to improve upon compression rates achieved by classic memoryless or simple prediction-based audio compression algorithms such as MP3. The representation error is masked using a psycho-acoustic filter. The goal of the masking function is to set the error such that reconstruction of audible samples is exact whereas the reconstruction of inaudible samples is such that the absolute magnitude of the error is minimized. We compute the entropy of the quantized pointers to all blocks, the quantized pointers to the applied transforms, the quantized scalars used to create the linear combination of transformed blocks, and the error vector returned.