A stereo audio coder with a nearly constant signal-to-noise ratio.

摘要

By exploiting advances in psychoacoustic research, state-of-the-art audio coders can achieve compression ratios of 10:1 or more without loss of perceptual quality. However, these coders are high-complexity coders. An enormous amount of computation is required to determine the masking properties and to calculate the masking thresholds.; This thesis develops a low-complexity audio coder based on the simple noise-shaping strategy developed by Paillard et al. The Paillard strategy is to shape the noise spectrum in such a way that the signal-to-noise ratio is 13 dB everywhere; therefore, no computation is required to calculate the masking thresholds. This thesis modifies the strategy and devises a hierarchical quantization scheme to achieve nearly constant signal-to-noise ratios for the most part of the spectrum. This thesis further modifies the strategy by considering the mechanism of masking. Informal listening tests have shown that this low-complexity audio coder performs very well at the fixed bit rate of 192 kbps, which means a compression ratio of 7.35:1. The low-complexity design at the expense of a lower compression ratio is an inherent trade-off between complexity and bit rates.