Abstract
Voice transmissions over packet networks have attracted attention recently with the quick growth of the Internet. In real-time data transmission over unreliable networks, packet losses are inevitable and the delayed packets further increase the packet loss rate. The lost and delayed packets degrade the playback quality. The degradation of the quality is more perceptible if a low bit-rate speech coder is adopted. Networks are more and more heterogeneous recently. Therefore, the scalable coding is important in network environments. In this thesis, we proposed a robust and scalable scheme based on the MDC schemes for low bit-rate speech coders. We use a split-band approach that decomposes the residual signal into two frequency bands, low-band and high-band. The two bands use different methods to generate low-band and high-band enhancement layers. Both of the sample-based multiple description coding (MDC) scheme and the LSP-based MDC scheme are used for concealing the lost packets. We adopt the sample-based MDC scheme for the high-band enhancement layer and the LSP-based MDC scheme for the core and low-band enhancement layers. We have implemented the proposed scheme on ITU-T G.723.1 standard. The experimental results show that the proposed scheme can provide good quality and scalable capability under different conditions of bit rates and packet loss rates.