Audio Coding with Signal Adaptive Block based Filter Bank Switching

Abstract

An improved efficiency perceptual audio codec is presented which analyzes each block of input signal and selects a suitable time/frequency mapping transform for it. The selection is based on statistics of the input signal vis-a-vis energy compaction and resolution power properties of the transforms employed which include the DFT (uniform subbands), DFT (critical subbands), DCT and CELP( for speech only blocks). The performance of the codec is compared with the widely famous MPEG-1 Layer-III algorithm. Efficiency enhancement is indicated by improved grades of subjective listening test for the proposed codec compared to those for MPEG-1 Layer-III at similar bit rates.. The paper concludes with a discussion of future research implications of the work.