Speech and wideband audio compression using filter banks and wavelets
Abstract
The rapidly expanding volume of multimedia data transfers over the Internet and current progress in wireless communications have led to burgeoning research activity in voice and wideband audio compression for novel applications. This thesis presents audio and speech compression methodologies under the framework of sampling the time-scale plane. We outline the different schemes of time-scale tiling, their corresponding implementations, and the compression scenarios that they are most suitable for. We present a theoretical framework using sampling to describe the traditional decimated wavelet transform, wavelet packets, the extrema representation, spline spaces and adaptive sampling. A technique for speech compression using the decimated wavelet transform is presented with a demonstration. The extrema representation is presented with an example showing its performance in the presence of road noise as in a moving car. We show that the extrema representation of spline wavelet decompositions yields a non-iterative reconstruction scheme. It is also amenable to a least squares formulation for reconstruction under noisy conditions. A new scheme for compressing CD quality music is presented. This method computes a signal-dependant filter bank structure based on a cost function. It has an adaptive filter bank structure which changes according to computational power available at the decoder, to facilitate real time decoding. It also has the multi-resolution property, such that successive bit streams yield improved quality of music. It incorporates a psychoacoustic model into its framework to provide transparent compression. Finally, it can be used as a variable-rate scheme, with a bit budget dictating the compression efficiency. Perceptual experiments at about 50 KBits/sec for monophonic CD quality music samples indicate that the reconstruction from compressed signals are virtually indistinguishable from the original samples.
Degree
Ph.D.
Advisors
Jamieson, Purdue University.
Subject Area
Electrical engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.