A re-investigation of scalar quantization for mobile speech transmission

Speech, audio, image or video signals are highly correlated and source coding algorithms are used to reduce the redundancy. Due to complexity and delay constraints the coded signal contains redundancy both inside one frame and also from frame to frame (time correlation). There has been increasing interest in algorithms that exploit the a priori knowledge within the channel or the source decoder. We present a unified view of source and channel decoding supported by a priori knowledge. The complexity of our decoding algorithm depends exponentially on the number of bits representing a quantized symbol. Therefore we investigate source encoding by scalar quantization. With this concept more bits per dimension are required for source encoding because we have to compensate for the loss in coding gain introduced by the lower dimensionality. As a consequence, less redundancy can be added by a channel code if the gross bit rate is kept constant. However the lower dimensionality allows the efficient use of unequal error protection (UEP) and our proposed decoding techniques. We show the application of our joint source-channel (de-)coding concepts to the ANSI-136 speech codec