The results presented in this chapter indicate that for non-individualized, anechoic HRTFs, significant data reduction can be achieved without perceptual consequences. The applied lossy HRTF representation method is based on known limitations of the human hearing system, and specifically exploits limitations in its frequency resolution.
It has been demonstrated that using noise stimuli, the absolute phase spectrum of the HRTF can be replaced by a linear phase curve, with an inter-aural delay that matches the average delay of the low-frequency part of an HRTF pair.
For the magnitude spectra, one magnitude value per ERB seems to be both a sufficient as well as a required resolution to result in a transparent HRTF representation. A coarser frequency resolution can result in audible differences for certain combinations of a specific HRTF pair and a subject. Although some HRTF pairs seem to be less sensitive to reductions in the number of parameters, comparisons between various HRTF sets and analysis of inter-subject results indicated that this can not be generalized.
In the next chapter, the proposed parametric method will be extended to multiple simultaneous sound sources, the incorporation of this method in MPEG Surround will be outlined, and listening test results using a variety of (multi-channel) audio content will be discussed.