NTT researchers have spent considerable effort investigating compression coding of 64 kbit/s and lower bit-rate traffic that preserves the quality of the original sound, and now this goal has been achieved with the development of TwinVQ*1 encoding. In this method, the individual bits of music data are not directly encoded, but are combined into pattern segments. These patterns are compared against standard patterns which are prepared in advance. The standard pattern which provides the closest match is selected, and the number associated with that pattern is transmitted as the compression code. Coding distortion is minimized even at low bit rates, so music and other sounds are successfully regenerated that are highly faithful to their originals. In addition, the method also lets users freely specify their own tradeoff between quality and the compression ratio, ranging from very-high-quality musical audition level to a high compression ratio that minimizes the amount of data for storage purposes. Regarding the decoder on the regeneration side, every effort has been made to minimize the amount of computation. Because real-time processing is achieved with PC software or single-chip DSP*2, we can offer a decoder system at very reasonable cost that supports music on demand and other attractive services.
Following up on this promising work, we will be applying the encoding scheme to a host of multimedia systems that support musical auditions, network- based karaoke, music-oriented storage media, and other capabilities. Meanwhile, we are also seeking to enhance the performance of the coding scheme to further improve the quality of regenerated music to that of a compact disk.
(Human Interface Laboratories)
*1 TwinVQ: Transform Domain weighted interleave Vector Quantization
*2 DSP: Digital Signal Processor