An adaptive multirate (AMR) speech codec proposed for the GSM is described. By dynamically splitting the gross bit rates between source (speech) and channel coding according to the channel quality, almost wireline speech quality even for poor channel conditions, and higher speech quality for relatively good conditions can be achieved. The CELP (code-excited linear prediction) based speech codec and the RSC (recursive systematic convolutional) code based channel codec were designed as homogenous/hierarchical as possible for a robust transmission. Novel algorithms for mode bit coding and error concealment were used. One of our design philosophy employed was that the source and channel codecs are jointly instead of individually optimized.