Multiple-description coding (MDC) of speech with an invertible auditory model

Network signal processing aspects dominate in speech and audio coding applications such as Internet telephony or packet radio networks. We demonstrate that our approach to speech coding in a perceptual domain provides an implicit forward error concealment mechanism to handle random erasures of the channel. To this end, the individual acoustic subchannels of our auditory model are grouped into different transport subchannels or packets. Due to the strongly overlapping, redundant filterbank structure of the model, reconstruction of speech without audible degradation becomes possible even if a significant percentage of channels is erased (e.g., up to 40% in a 50-channel auditory model for narrowband speech). We discuss this result both from a hearing-physiology and a frame-theoretic perspective.