Channel-encoded transmission of MELP-compressed speech

A mixed-excitation linear predictive (MELP) vocoder was previously selected as the U.S. Federal Standard for 2400 bps speech compression. This paper examines novel methods of error control for the new speech coder. In particular, we describe methods for exploiting the residual redundancy in the MELP bitstream; this is done by effecting maximum a posteriori decoding of the channel-encoded MELP bitstream. Examples include a variety of rate-1/2 convolutional codes and suitably-modified "turbo" codes.