Evaluating a model of auditory masking for applications in audio coding

Abstract

This investigation evaluates the adequacy of an excitation pattern (EP) model of auditory-masking to account for a masking condition commonly observed in the digital coding of audio signals: namely, the masking of quantization noise by either speech or musical sounds. This condition is modeled as a case of masking of a spectrally complex target by a multi-component complex masker (either speech or musical sounds). The evaluation focuses on the extent to which the EP model accounts for the 10-15 dB excess masking observed when multiple maskers occur simultaneously. Excitation-pattern predictions of the masking effects of two-component maskers were compared with predictions of masking-additivity models which closely predict experimental data. The EP model underestimates the combined effects of the two maskers both when the masker frequency components fall within the signal auditory-filter bandwidth, and when they fall outside this bandwidth. A correction for masking additivity may be needed prior to applying the EP model to audio-coding situations.