Reconstruction of Multiple Descriptions by MMSE Estimation

Inhalt:
This paper shows the application of soft-decision source decoding (SDSD) to correlated speech codec parameters given by multiple description scalar quantization. In the presence of additive white Gaussian noise on the transmission link, the reconstruction quality can be largely improved by the application of SDSD. Furthermore, if the transmission link is characterized by packet losses only, the presented approach improves the reconstruction quality whenever packet losses occur. The proposed approach is compared to a well-known soft-decision approach based on cross-decoding.