An accurate speech signal modeling for communication applications on IP networks

The article deals with speech coding, which is particularly crucial for modern telecommunication systems. Since the end of the 1930s, much work has been done in the speech coding domain. But in spite of the numerous norms and products that have been developed, challenges exist and are more than ever a topical question. We have focused on the modeling of the speech signal. Hence we have studied a spectral modeling based on the MBE (multiband excitation) model, as the INMARSAT coder normalization (INMARSAT-M voice codec specifications, 1991) demonstrates the maturity of this approach. This coder gives a good quality/bit rate compromise; however it presents some limitations in terms of performance. In a first step, our study has consisted in proposing some improvements. This has led us to develop a coder modeling partly based on an improved MBE modeling, and on new modeling technologies.