Adaptive quantization and prediction in speech coding
Feng Chongxi, Yao Hui-Juan, Weili Yang · 2005
The problems of adaptive quantization and prediction in a DPCM system are analyzed. It is shown that while quantizing speech signal with an adaptive algorithm, the probability distribution of Piassociated with the quantization intervals is not uniquely determined by the PDF of speech, but can be controlled by the adaptive parameters (Gi) and algorithm of adaptation. Thus, speech signal can be quantized by an optimum quantization charateristic (OGC) designed to match a specific power-limited PDF in order to meet certain SNR and entropy requirements. This paper derives a PDF with the maximum SNR which is greater than that of Gaussian PDF in 0.5dB. Moreover, the calculation of SNR in an adaptive quantizer and the method for searching adaptive parameters (Gi) are discussed. Finally, a constant-increment sequential adaptive prediction algorithm is developed. It removes multiplier with a prediction gain loss less than 0.5dB.