Multiple description speech codecs applying distributed subframe interleaving

Error resilient coding is strongly demanded by voice communication over unreliable packet networks. In this paper, we propose a new multiple description scheme named distributed subframe interleaving, and employ this new scheme to create two novel multiple description codecs based on CELP. With introducing distributed subframe interleaving operated on two adjacent speech packets, parameters of LSP are interleaved into different descriptions while other parameters belong to different subframes are assigned in subframe-distributed principle. Moreover, parameter interpolation is adopted to restore the lost description without introducing any extra delay at the decoder. Experimental results demonstrate that the proposed codecs are able to provide significantly better speech quality than traditional CELP codec and suitable to unreliable packet networks.