Wide-band speech coding based on bandwidth extension and sparse linear prediction

Abstract

A novel wide-band speech coding scheme based on bandwidth extension and sparse linear prediction is proposed in this paper. Bandwidth extension utilizes the correlation that exists between the low and high frequency parts of the wide-band speech signal to regenerate, at the decoder, the high frequency components using the decoded low band signal. This leads to a wide-band speech coding scheme built on a narrow-band codec at the extra cost of a small increase in the bite rate. Two successful narrow-band coding structures are considered here for this purpose, namely Multi-Pulse Excitation (MPE) and Algebraic Code Excited Linear Prediction (ACELP), in conjunction with the recently introduced sparse linear prediction. Our tests confirm the usefulness of this new prediction technique for narrow-band speech coding. Moreover, a well-posed formulation is derived that significantly decreases the computational complexity and makes it possible to solve the sparse linear prediction problem within a reasonable complexity.