This thesis describes the development and real-time implementation of a hifi music coder which compresses 16-bit PCM digital audio samples to 4 bits without audible degradation. To satisfy key operational requirements such as low coding delay and low hardware complexity, the coding development work was based on sub-band ADPCM techniques. Initial optimisation of the sub-band ADPCM scheme was undertaken using high-level computer simulations. This was made possible through the construction of a high-quality audio data acquisition unit which allowed music data to be transferred to the computer for processing by the simulation programs. Following the optimisation phases, the sound quality of the coders was assessed by conducting formal listening tests. The coded audio material used for these tests was obtained by recording from the computer, via the acquisition unit, audio which had been encoded and decoded through the simulations. For the tests a panel of listeners was used to judge the similarity between the original PCM audio and the coded audio using a number of short musical extracts. The results of the listening tests indicated that sub-band ADPCM scheme provided audio which was essentially indistinguishable from the original material using as few as four bands, at a compression ratio of 16:4. To extend the scope of the subjective tests a 4-band ADPCM coder was implemented in real-time using programmable digital signal processors This hardware was interfaced to a compact disc player, enabling the digital audio coming off the disc to be coded on-line. This allowed extensive listening tests to be conducted which demonstrated that hi-fi music coded to 4 bits with the sub-band ADPCM scheme could not readily be distinguished from the original, even for the most critical signals.