Speech and Audio Coding — Aiming at High Quality and Low Data Rates

The historic “coding gap” between high-rate coding of narrow- and wide-band speech on the one hand, and low-rate coding of narrow-band speech on the other hand, has been bridged more and more during the past 15 years. The GSM coder of 1990 was a very important milestone in this process, as it has prompted increasing research towards better quality and higher compression. These particular research efforts, together with other relevant activities worldwide, have helped closing the gap. In the following, the concepts behind this progress will be explained. A special focus will be put on the basis of this success, namely, on the fact that, finally, a break-through could be achieved for narrow-band speech, allowing for good quality at medium-to-low rates. For wide-band speech this holds true at medium rates. The same concepts as applied to speech coding were also applied to music, yet, ending up with some noticeable conceptual differences. While for speech time-domain approaches prevail, frequency-domain coding turned out to be more successful for audio. A characteristic, there, is extensive exploitation of psycho-acoustical phenomena.