Computationally efficient compression of audio signals by means of RIQ-DPCM

Abstract

The need to transmit large amounts of data over limited bandwidth channels has resulted in many methods for digital data compression. The common approach is to identify and remove redundancy from the input data stream using knowledge of the source characteristics. In the case of signals intended for human observers (speech, music, pictures, etc.) it is also useful to consider the strengths and weaknesses of the human sensory systems in order to achieve a greater degree of data compression. Unfortunately, achieving perceptually transparent compression requires considerable computational resources. For situations requiring extremely low computational complexity without strictly transparent coding, such as multimedia applications on personal computer platforms, a new adaptive differential pulse code modulation (DPCM) data compression scheme is proposed. Although standard DPCM structures are widely used in single-talker speech coding systems, the models and statistical assumptions well-known for speech signals are not applicable to arbitrary audio signals such as music. The new DPCM formulation presented includes a recursively indexed quantizer (RIQ) to eliminate the problem of overload distortion, a simple predictor structure to take advantage of the short-term correlation present in wideband audio signals, and an adaptation strategy to optimize the system to the local statistics of the input signal. Thus, the new RIQ-DPCM formulation is presented as a computationally efficient means of wideband audio compression.<>