The first part of this paper surveys and classifies the best performing currently available lossless compression algorithms for stereo-CD-quality digital audio signals sampled at 44.1 kHz and quantized to 16 bits. This study finds that these algorithms appear to have reached a limit in compression that is very modest compared to what can be achieved with lossy audio coding technology. With this limit as a target, we designed a computationally efficient algorithm for losslesss audio compression (which we call AudioPaK). This new lossless compression algorithm uses only a small number of integer arithmetic operations and performs as well, or better than most state-of-the-art lossless compression algorithms. The main operations of the algorithm are prediction with integer coefficients and Golomb-Rice coding. The second part of the paper presents the complete architecture of AudioPaK including suggestions on parallelizing parts of the algorithm using the MMX instruction set. 37 Pages