A Novel Multiple Description Scalable Speech Codec Based on Sinusoidal Model

Robust and flexible speech codecs are required more and more by speech communication over unreliable and heterogeneous channels such as the Internet. In this paper, a novel multiple description scalable codec is proposed, which is based on the sinusoidal model and equivalent rectangular bandwidth (ERB) noise model. It combines the advantages of scalable coding and multiple descriptions coding (MDC), and provides relatively high transmission reliability and rate adaptation. The lost packet does not affect the state recovery of this state-less codec. The performance of the proposed codec was evaluated under memory and memoryless channels. Analytical and experimental results show that it is suitable to unreliable and band limited channel such as Internet.