Multiple-description coding of speech using forward error correction codes

A flexible framework is presented which performs multiple-description coding of speech signals with two or more channels. The use of forward error correction codes together with a layered speech codec permits encoding into more than two descriptions without excessive increase in complexity. Results of a formal MOS listening test reveal considerable improvements in robustness as long as base layer quality and the number of descriptions are chosen appropriately. A modification of the original encoding scheme allows trading off bit rate savings against robustness to extreme channel conditions. Different coding schemes can easily be compared using a real-time demonstrator software.