A waveform interpolation speech coder its enhancement and optimization

Abstract

One of the most promising methods for the low-bit-rate speech coding is waveform interpolation (WI).This research aims at developing a speech coder based on the waveform interpolation (WI) scheme, with an attempt to deliver near-toll-quality (wireline-quality) speech at rates around 4 kbps. A waveform interpolation speech coder has been proposed, implemented and evaluated in floating-point arithmetics using Matlab and C programming language environment. Most modules in this simulation platform are implemented in C language to speed up its execution time.In this thesis, the general structure of a conventional WI coder is described and analyzed to understand how different properties of WI coding parameters are reflected to the speech quality. Then a unique frame structure of a 10-ms processing frame with a 25-ms algorithmic lookahead is proposed. A coding scheme depending on input speech characteristics is suggested for an efficient quantization of coding parameters. A novel idea is proposed that makes use of variable extraction-rate of a characteristic waveform (CW).