Optimized error correction of MELP speech parameters via maximum a posteriori (MAP) techniques

The U.S. Government has developed and adopted a new Federal standard vocoder which operates at 2400 bps and is called MELP-mixed excitation linear prediction. This algorithm has quite good voice quality under benign error channel conditions. However, when subjected to high error conditions as may be experienced in vehicular applications, correction techniques may be employed which utilize the underlying inter-frame residual redundancy of the MELP parameters. This paper describes experiments conducted on the MELP algorithm when combined with Viterbi convolutional error decoding, and enhanced with maximum a posteriori techniques which capitalize on the redundancy statistics. Both hard and soft Viterbi decoding situations are investigated.