A novel computing structure for 3‐D audio rendering with multiple sources

A novel computing structure has been implemented for efficient 3‐D audio production. Head‐related transfer function (HRTF) on a sphere is represented as a linearly weighted combination of M eigenfunctions (EFs). Each EF is a function of frequency (or time) and each weight (of M) is a function of azimuth and evaluation. With this representation the 3‐D positioned binaural signals of any single sound source can be obtained by first multiplying the dry signal with M weights and then convolving them with M EFs. Proper delay corresponding to ITD is introduced into binaural signals for the listener. Similarly, multiple sound source 3‐D positioning is achieved by weighting each source with its corresponding weights and by filtering them with the same set of EFs. For every source added only one more set of weight has to be used to multiply the source without increasing the time‐consuming EF filtering. It results in an order of magnitude or more MIPS saving compared with direct convolution, especially when the num...