Novel multiple description scalable speech codec based on sinusoidal model

A novel multiple description scalable codec is proposed, based on sinusoidal and equivalent rectangular bandwidth (ERB) noise model. It divides the parameters of sinusoidal and noise model into different descriptions and layers. There are three kinds of rates in a description. It combines the advantages of scalable coding and multiple description coding (MDC), and can provide relatively high transmission reliability and rate adaptation. The performance of the proposed codec under memory and memoryless channel was evaluated. Experimental results show that it outperforms other codecs in the same bit rate when packet losses exit.