The success of a perceptual audio coder depends strongly on the joint efficiency of three of its main building blocks: the analysis/synthesis filter bank, the psychoacoustic model and the quantizer. This paper focuses on the combined choice of an appropriate filter bank and an accurate psychoacoustic model. A few statistical results concerning stationarity and harmonicity are used to discuss the objective and subjective coding gain, the necessity to address multiresolution, and the pertinence of including aspects of pitch and timbre perception in perceptual modeling. Specific solutions are proposed.
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.