Towards a new perceptual coding paradigm for audio signals

Rate-Distortion Allocation for Time-Frequency Dependent Audio Coding

“…An informal subjective listening test revealed that, for all files, and at all the prescribed rates, the quality of ED coding was higher than that of NMR. The quality improvement of ED over NMR increased as the rate dropped, corroborating the observations contained in [1] concerning the inadequacies of NMR for non-transparent-quality audio coding.…”

“…We shall compare coding results for two different distortion measures (d in (2)): (1) the standard NMR measure, and (2) a distortion measure posed entirely in the perceptual domain, following [1], [2]. For the latter, define the Excitation Distortion (ED) to be the maximum dB-difference between reference and coded excitation patterns; in symbols:…”

Vectorial Spectral Quantization for Audio Coding

“…Besides the constraint imposed by the limited amount of bits, an additional constraint is added during the quantization process, which avoids the setting to zero of a whole sub-band, to counteract some of the drawbacks of the perceptual model. As also observed in [5], the setting to zero of an entire sub-band, even if the overall (average or maximum) noise to mask ratio is below 1, may engender perceptual artifacts because the masking effects have been calculated on the original signal. These problems can in principle be alleviated through the introduction of a loop to consider the perceptual characteristics of the quantized signal, but it would increase significantly the complexity of the encoding.…”

“…These problems can in principle be alleviated through the introduction of a loop to consider the perceptual characteristics of the quantized signal, but it would increase significantly the complexity of the encoding. Another approach has been considered in [5] where the minimization of the error is done in the loudness domain. Although good results have been obtained, the combination with the use of the MDCT is still problematic.…”

A prototype system for object coding of musical audio

“…Finally, following recent results in perceptual audio coding [14], the prior for the residual is designed so that the quantitative importance of the signal in each auditory band is roughly proportional to its loudness. We define the excitation power of the signal in the auditory band centered at frequency f on frame t by E tf =…”