6. Conclusion
The research suggests that the hybrid
algorithm was suitably transparent for some of the test subjects, but not for
all, and not across the average of the 20 subjects tested. However, using an
algorithm that uses a convolution scheme to create distinct early echoes and a
recursive filter network to create a diffuse reverb tail is a viable artificial
reverberation design strategy.
There is room for further research into many of the topics considered in this thesis. An adaptation of the decimation technique for the equalization component can be explored instead of the subjective shelving architecture. The use of non-radix-2 DFTs could provide the sample lengths needed for smooth and tonally balance block convolution. The addition of this type of quasi-adaptive voicing will greatly enhance the robustness of the algorithm, eliminating the need for subjective equalization design. The algorithm would be capable of evaluating its own voicing needs based on any discrepancies between the full and truncated impulse responses.
A limiting utility of reverberation network architecture is the lack of adequate time and frequency density measurement tools. Currently, only a mathematical approach for a single and parallel comb network have been utilized. Correlation of the ISO3382 time and clarity measurements could be adapted into accurate Time and Frequency Density measurement values. These measurements could be compared to the Schroeder and Jot specified values for accurate reverberation.
With processor performance as a major limiting factor, Moorer suggests varying the bit depth of the impulse response to ease the burden of computation. Instead of encoding every sample with 16 bits or greater, the rapid decay can be exploited by encoding the first 30% at over 12 bits, the next 30% between 8 and 11 bits, the next 30% between 4 and 8 bits and the final 10% with less than 4 bits. A typical monaural impulse response of 2 seconds, encoded at CD quality (16 bit, 44.1 kHz) will contain 1.4 Mb (176 kB), however, by employing this bit ratio, the impulse response could contain between 961 kb (120 kB) and 661 kb (83 kB).
This hybrid model is able to capture the general reverberation and frequency character of the acoustic spaces examined. When using an impulse response of an auditorium, room, hall, etc. the implementation of this algorithm produces a quick, quasi real-time reverberant output that can easily be used as an alternative to a convolution system.