An analysis of the weaknesses of the MBE coding scheme

This paper explores the deficiencies in the multiband-excitation (MBE) model (proposed by Griffin and Lim (1988) for very-low-bitrate speech coding, and incorporated in the INMARSAT-M standard (1991)) that restricts its application in commercial telephone and mobile applications. The pitch frequency has been found to be the most crucial parameter in the MBE scenario. The errors occurring in the estimation of this parameter get reflected in the estimation of other parameters such as voiced/unvoiced decision and the spectral shape, since the estimation techniques of these parameters implicitly assume accurate knowledge of the pitch frequency for the spectral harmonic matching techniques they employ. Thus the output quality of 8 kHz-sampled speech suffers from a set of audible distortions which result in the degradation of the decoded speech quality. This paper is an attempt to analyse the weaknesses of this coding scheme under telephone-channel conditions