PERCELP - Perceptually Enhanced Random Codebook Excited Linear Prediction

The method of exciting the vocal tract filter with vectors chosen from fixed and adaptive codebooks has become the dominant technique in present day low bit rate speech coders. It has long been recognised that to improve the perceptual quality of these essentially voice production models, it is necessary to consider psychoacoustic properties of the human ear. The weighting filter traditionally used for this purpose is sub-optimal as it doesn't explicitly evaluate auditory characteristics. In this paper we replace the weighting filter with an auditory model which enables the search for the optimum stochastic codevector in the psychoacoustic domain. The resulting speech quality is considerably better and the computational overhead low enough to warrant the use of the technique in contemporary speech coders.