New directions in speech and audio coding

Abstract

Rapid progress has been made in recent years in synthesizing high-quality natural sounding speech at low bit rates. The progress in speech coding together with advances in VLSI circuits has sparked great interest in using low-bit-rate speech coding technology for practical applications. and new standards are being developed for transmitting digital speech at 4.8, 8, and 16 kbits/s. New algorithms have brought the bit rate for coding of CD-quality audio signals down to 128 kbits/s. The rapidly increasing demand for communication over digital channels (cellular radio and ISDN) is providing a strong impetus for decreasing the bit rates even further—to 4 kbits/s or lower for speech and to 64 kbits/s or lower for CD-quality audio. Present coding algorithms make use of concepts such as linear prediction, transform coding, vector quantization, and subjective error criteria for lowering the bit rate. As important as these concepts are, just incremental improvements of the present coders cannot bring the bit rate down further without degrading the signal quality. A critical review of the present state of the art in speech coding will be provided and new results from recent work will be presented.