A SWITCHED-ADAPTIVE QUANTIZATION TECHNIQUE USING μ-LAW QUANTIZERS

Abstract

In this paper, we propose a switched-adaptive quantization technique suitable for high quality quantization of signals which, as well as speech signals, have short-term statistics modeled by Gaussian probability density function. Particularly, from the set of the 2k disposable quantizers, the technique we propose performs a two-stage selection of the quantizer which is designed for the signal statistic nearest to the one of the signal to be processed. In the first selection stage, the choice of the quantizer from the set of k disposable quantizers is managed by the switched technique. Further, in the second selection stage, the choice is made between the restricted and the unrestricted quantizers, regarding the fact whether the maximum amplitude of the samples within the current frame is lower or higher than the upper support region threshold of the restricted quantizer, respectively. In order to properly define the frame length, in this paper we are introducing the rate-quality compromise criterion, along which we are studying the gain in the signal to quantization noise ratio and in the compression, that are achievable with the proposed technique in reference to the G.711 recommendation and another technique we have recently proposed.