Split Vector Quantization of Compressive Sampling Measurements for Speech Compression

Abstract

Compressed sensing (CS) have gained much interest in recent years for its advantage of simultaneously acquiring and compressing signals. The acquired signals using CS need quantization to be exploitable in digital systems for storage or transmission. Split vector quantization (SVQ) is proposed to quantize the CS measurements, and a speech compression application is performed to illustrate its usefulness. The proposed method is compared with state-of-the-art quantization techniques which are scalar quantization, differential pulse code modulation, vector quantization, and multistage vector quantization for speech compression in terms of perceptual evaluation of speech quality, signal-to-noise ratio, mean square error, and execution time. Results demonstrate that SVQ represents a competitive alternative of the studied methods for CS measurements quantization.