A novel method for binaural rendering based on parametric representations has been outlined. In contrast to conventional, convolution-based methods, head-related impulse responses are transformed to the parameter domain and combined with parameters that describe the statistical properties of the various signals that are radiated by virtual sources. From the combination of statistical properties between virtual source signals and HRTF parameters, binaural parameters were derived that describe the relevant perceptual properties at the level of the eardrums. A binaural rendering stage subsequently reinstates the binaural parameters obtained to various time/frequency tiles of a down-mix of the various virtual source signals.
The proposed method can be integrated with parametric multi-channel audio coders that rely on inter-channel cues such as level differences and inter-channel correlations. As such, it is currently an integral part of the MPEG Surround specification, including extensions to stereo down-mixes and support for anechoic HRTFs (BRIRs). Results of a listening test revealed that the proposed method outperforms conventional, convolution-based methods in terms of perceived quality and computational complexity. These properties, combined with the unsurpassed compression efficiency of MPEG Surround, make this MPEG Surround extension very suitable for mobile applications.