Quantitative Characterisation Of Perceptually Relevant Artifacts Of Synthetic Reverberation Using The Earwig Distribution
Synthetic reverberation is generally derived by using filter-based algorithms to mimic the types of impulse responses which are generated in natural room responses. Standard approaches would include Schroeder parallel comb filter, feedback delay networks and nested all-pass structures. One of the difficulties in specifying a useable reverb algorithm is that sonic quality is sensitive to the parameter choices used. For example, inappropriate choice of gain and delay in a Schroeder reverberator leads to what is referred to as ‘graineyness’ or ‘fluttering’. These are subjective descriptions of perceived aural characteristics associated with reverb response. Because of the qualitative nature of these artifacts, parameter specification for any new reverb structure generally involves subjective assessment of sonic quality - a ‘tweak-and-listen’ approach. This paper attempts to take some initial steps toward objective assessment of synthetic reverberation by identifying possible quantitative measures which correlate with these perceptual features of reverberation response. These quantitative measures are derived from a novel joint time frequency distribution which incorporates accurate ear masking effects. This has been developed by generating ear-response based smoothing kernels for the Wigner Distribution leading to its designation as the EarWig Distribution (EWD). Thus, the EWD of an audio signal provides us with a representation which highlights perceptually relevant signal features only. From the EWD of reverb responses it is shown that it is possible to identify some objective signal characteristics which contribute to perceived graineyness and fluttering. This provides the basis for the establishment of associated quantitative measures which would be a significant contribution in synthetic reverberation design.