PDF optimized parametric vector quantization of speech line spectral frequencies

Abstract

A computationally efficient, high quality, vector quantization scheme based on a parametric probability density function (PDF) is developed for encoding speech line spectral frequencies (LSF). For this purpose, speech LSFs are modeled as i.i.d realizations of a multivariate normal mixture density. The mixture model parameters are efficiently estimated from the training data using the expectation maximization (EM) algorithm. The estimated density is suitably quantized using transform coding and bit-allocation techniques for both fixed rate and variable rate systems. Source encoding using the resultant codebook involves no searches and its computational complexity is minimal and independent of the rate of the system. Experimental results show that the proposed scheme provides 2-3 bits gain over conventional MSVQ schemes. The proposed memoryless quantizer is enhanced to form a quantizer with memory. The quantizer with memory provides transparent quality speech at 20 bits/frame.