Error concealment by near optimum MMSE-estimation of source codec parameters

In digital communications source coding is indispensable to achieve a high bandwidth efficiency in applications where bandwidth is a limited resource. Usually these source coding algorithms determine speech or audio parameters which are highly sensitive to transmission errors. This paper deals with an error concealment technique that benefits from residual redundancy remaining after source coding. Due to delay and complexity constraints a certain amount of inter-frame as well as intra-frame correlation between source codec parameters remains which might be exploited to enhance the subjective audio quality. In this contribution we propose an algorithm which is able to utilize both types of redundancy if the source parameters exhibit a two-dimensional Markov property. Actually, this approach is nearly optimal in the minimum mean square error (MMSE) sense but much less complex compared to the optimal estimator.