Nonlinear systems identification is a widespread topic of interest, particularly within the audio industry, as these techniques are employed to synthesize black box models of nonlinear audio effects. Given the myriad approaches to black box modeling, questions arise as to whether an “optimal” approach exists, or one that achieves valid subjective results as a model with minimal computational expense. This thesis uses ABX listening tests to compare black box models of three hardware audio effects using two popular nonlinear implementations, along with two proposed modified implementations. Models were constructed in the Hammerstein form using sine sweeps and a novel measurement technique for the filters and nonlinearities, respectively. Testing revolved around null hypotheses assuming no change in model identification regardless of the device modeled, implementation used, or program material of the model stimulus. Results provide clear evidence of an effect on all of these accounts, and support a full rejection of the null hypotheses. Outcomes demonstrate a preferable implementation out of the algorithms tested, and suggest the removal of certain implementations as valid approaches altogether.
Wesley A. Bulla
Entertainment and Music Business, Mike Curb College of
Master of Science in Audio Engineering (MSAE)
audio; engineering; algorithm; Hammerstein; modeling; coding; psychoacoustics; distortion; non-linear; effect
Mayo, P. G. (2018). "A Perceptual Comparison of “Black Box” Modeling Algorithms for Nonlinear Audio Systems." Master of Science in Audio Engineering (MSAE) thesis, Belmont University, Nashville, TN. 1. https://repository.belmont.edu/msaetheses/1