Very low bit rate voice compression for mobile communications

Brooks, Fiona Clare Angharad (1998) Very low bit rate voice compression for mobile communications. University of Southampton, Doctoral Thesis.

Abstract

This thesis is concerned with very low bit rate voice compression for mobile communications, concentrating exclusively on speech encoders beneath 4kbps, namely Prototype Waveform Interpolation (PWI), Multiband Excitation (MBE) and Sinusoidal Transform Coding (STC).

Specifically, a 1.9kbps PWI speech encoder was developed, which employed Zinc Function Excitation (ZFE) to represent the voiced speech. Analysis-by-synthesis was adopted to select the best ZFE for each voiced segment. The error sensitivity of this speech coder was assessed together with harnessing several ZFEs for each excitation optimization sub-segment, in order to create a higher speech quality, higher bit rate speech encoder.

Furthermore, several MBE speech encoders were developed, where the first version adopted simple pulse excitation with five MBE bands to create a 2.3kbps speech coder. Another speech encoder incorporated adds three MBE in the previously developed PWI-ZFE speech coder, in order to encode speech at 2.35kbps. The performance of MBE at a higher bit rate, where more frequency bands have been added, is also discussed. It was found that the PWI-ZFE speech coder with three MBE bands was preferred by 64.10% of listeners.

The third technique, namely STC was developed into a STC-PWI speech coder operating at 2.4 and 3.8kpbs. Analysis-by-synthesis was harnessed to determine the best Fourier coefficients for each segment.

A number of wavelet techniques were also examined in this thesis, which were employed in the process of pitch detection and voiced-unvoiced determination. A pitch detector with an overall pitch estimation error rate of 3.9% was developed.

This record has no associated files available for download.

Contributors

Download statistics