High quality speech coding based on hybrid CELPC and VELPC techniques

Abstract

Code-excited linear predictive coding (CELPC) is now the main technique used to produce good quality speech at bit rates around 4.8 kbps. However, the original CELPC is impractical in most cases due to its heavy computational load. A hybrid CELPC and voice excited linear predictive coding (VELPC) scheme is presented in the paper for speech coding with lower complexity. In the algorithm, the speech signal is firstly divided into two parts, the base-band and the high-band respectively, in frequency domain, and then the base-band and the high-band signal are coded with CELPC and VELPC techniques respectively. The test experiments showed this new coder could produce synthesized speech with good quality at bit rates around 4.0 kbps. >