Quality Enhancement of Low Bit Rate Speech Coder with Nonlinear Prediction

Abstract

Toll quality speech codec design with a low bit rate is really a challenging task in modern communication because of the drastic increase in end-users in social networks. Most of the low bit rate speech codecs are based on linear prediction. The code-excited linear prediction codec (CELP) gives good quality decoded speech at a lower bit rate of 4.8 Kbps. But, it neglects the natural nonlinear effects present in speech production process. So, some adaptive techniques are to be used to make the system nonlinear to perform better than linear prediction speech codecs. An adaptive technique with nonlinear prediction of speech, based on truncated Volterra series, is used to generate the nonlinear prediction coefficients. The generated nonlinear prediction coefficients are implemented in G723.1 CELP codec to introduce code-excited nonlinear prediction (CENLP) codec. Advancements in the performance are evaluated using subjective and objective quality measures and compared with the normal G723.1 CELP codec.