Some of the most advanced surround sound formats are based on heavy numerical signal processing, which is required at the production stage or in the reproduction system. In both cases, convolution with very long impulse responses is a typical requirement, for creation of realistic reverberation. Solid spatial imaging requires independent convolution of many channels with proper sets of impulse responses.
Typical cases addressed here are binaural/transaural systems based on convolution with HRTFs and cross-talk cancellation at the reproduction stage, and Ambisonics-like processing based on convolution with B-format impulse responses and proper spatial decoding for feeding multiple-speaker arrays.
Recently it was shown that coupling the better-working parts of these two very different approaches a third class or surround sound systems can be obtained [1]: the new hybrid method, which now relies on stable physical and psychoacoustical bases, resembles closely the original Ambiophonics approach pioneered by Ralph Glasgal [2].
In this paper it is shown how the heavy processing task required for implementing an Ambiophonics system in real-time can be accomplished by use of a very cheap general purpose computer, by means of a clever multiple convolution algorithm, based on the partitioned-impulse-response approach initially suggested by Stockham in 1966 [3].
This algorithms provides efficient convolution in the frequency domain, with little input/output latency, and overall performance increased over standard overlap-and-save algorithm, thanks to the better correspondence with the hardware structure of modern CPUs.
