Joint pre-echo control and frame erasure concealment for VoIP audio codecs

Pre-echo artifacts are a common problem of transform audio codecs owing to their comparatively large block lengths. If such codecs are used in Voice-over-IP (VoIP) applications, the concealment of lost audio packets poses another major issue. These two problems are usually solved by two separate algorithmic components: A mechanism for pre-echo control (PEC) and a frame erasure concealment (FEC) algorithm. In this contribution we propose a combined PEC/FEC approach which elegantly solves both problems based on a single side information stream. In particular, our proposal is suitable for (subband) transform audio codecs that are based on a frequency transform with half-overlapped windowing (such as the popular MDCT). The described method is part of the Huawei/ETRI candidate for the future ITU-T super-wideband extensions of Rec. G.729.1 and G.718 where it has been successfully applied in the context of high frequency (8-14 kHz) encoding.