A new auditory masking model for speech and audio coders

This paper defines a new auditory masking model based on auditory and signal detection theories. The model attempts to provide a generic and quantitative way of detecting whether a probe signal can be perceived in the presence of a masker signal. The model can be applied to any masker/probe signal and does not require heuristic methods to account for the asymmetry of noise and tone masking. Thus a signal does not need to be decomposed or classified into tonal and noise components to compute a masking threshold. The unified and quantitative model makes it conducive for use in speech and audio coding algorithms. Despite the fact that existing auditory masking models are not well defined and are heuristic in nature, they are widely used in almost all modern audio-coding algorithms. Therefore, it is realistic to assume that a better masking model, which models the actual physiological mechanisms and neurology of the hearing system, will make an important contribution towards better quality and/or lower bit rate speech and audio coders.