Optimal Lloyd-Max quantization of LPC speech parameters

This paper introduces an approach to scalar quantization of LPC reflection coefficients which outperforms typical scalar quantization schemes using a squared-error distortion measure. As in vector quantization, an iterative algorithm is used to generate the source codebook. Scalar quantization produces a greater distortion, for a given number of bits- than vector quantization but reduced computation time and storage requirements make the Lloyd-Max approach economically attractive for implementation in real-time hardware. The algorithm is developed and a comparison with another scalar quantization scheme is made.