Enhancing vocoder performance for music signals

Abstract

Low bit rate standards involving the code excited linear predictive coder (CELP) are typically optimized for telephone speech. The performance of CELP with non-speech signals, and particularly with music, degrades considerably. It is of interest these days to improve the robustness of these coders by adding pre- and post-processing stages to the core compression algorithm. In this paper, we propose a frequency-domain technique involving time-varying filters to enhance the audio quality with a small number of overhead information bits. Preliminary results have shown both qualitative and quantitative improvements in the sound quality.