A new perceptual model for audio coding based on spectro-temporal masking

In psychoacoustics, considerable advances have been made recently in developing computational models that can predict the discriminability of two sounds taking into account spectro-temporal masking effects. These models operate as artificial observers by making predictions about the discriminability of arbitrary signals [e.g. Dau et al. J. Acoust. Soc. Am. 99, Vol. 36(15), 1996]. Therefore, such models can be applied in the context of a perceptual audio coder. A drawback, however, is the computational complexity of such advanced models, especially because the model needs to evaluate each quantization option separately. In this contribution a model is introduced and evaluated that is a computationally lighter version of the Dau model but maintains its essential spectro-temporal masking predictions. Listening test results in a transform coder setting show that the proposed model outperforms a conventional purely spectral masking model and the original model proposed by Dau.