Classification and spectral extrapolation based packet reconstruction for low-delay speech coding

A common aspect of speech transmission through packetized networks is the need to consider discarded (missing) packets as a result of error detection or network overload. The missing packets and the possible mistracking that results in the speech decoder lead to significant quality degradation. In this paper, we examine recovery techniques based on speech classification and spectral extrapolation. The recovery system extrapolates independently the excitation signal and the short-term synthesis filter using an extrapolation strategy based on speech classification (voiced, unvoiced, transition, silence). The extrapolation of the short-term filter uses a least-squares fading memory polynomial filter applied to reflection coefficients. Objective and subjective quality evaluations of the recovery system applied to the LD-CELP G.728 standard for random and burst frame erasures are presented. The results indicate that the system is robust up to a frame erasure rate of 10%. Very little degradation in quality was observed at erasure rates up to 3% for random frame erasures.