SPEECH TRANSFORM CODING USING RANKED VECTOR QUANTIZATION
Ying Zhang · Summit (Simon Fraser University) · 1996
As the demand for mobile communications continues to grow, speech codec designers are faced with the challenge of providing high quality speech coding systems at low bit rate.New efficient speech coding algorithms are required to reduce the bit rates and obtain high quality reproduced speech signal.Transform coding is a frequency-domain coding technique which has been studied extensively and used widely in low bit rate speech coding systems.The Vector Transform Quantization (VTQ) system is an example of transform coding, where a set of vector quantizers are used to quantize the transform coefficients.With the motivation of developing high quality speech coders at low bit rate, this thesis investigates two new speech coding algorithms with the goal of obtaining high quality synthetic speech at the rate 2.4 kbps.Based on the VTQ system, the Vector Transform Quantization with Coefficient Ranking (VTQ-CR) system and its enhanced version, Vector Transform Quantization with Coefficient Ranking and Adaptive Linear Prediction (VTQ-CR-ALP), are developed.Coefficient ranking technique and adaptive transform domain linear prediction analysis are proposed to improve the performance of conventional VTQ coders.