Simulated Gaussian noise with too few components gives atypical masking results

Evidence

Gaussian noise can be simulated by a large number of low-level tonal components with random phase. Such approximations have to be adopted with care when used as maskers, because of the non-Gaussian amplitude distribution produced in the case of too few lines and/or non-arbitrary phase conditions. The same holds for noises produced by “statistically” chosen sequences of “0” sand “1” s. Such noises often produce masking effects different from those produced by Gaussian noise because of the different amplitude distribution and the different temporal and spectral pattern within narrowfrequency bands. Such patterns are important for the excitation level versus critical-band rate versus time pattern, the basis of masking. — Professor Dr.-Ing. Hugo Fastl, Professor Dr.-Ing. Eberhard, Psychoacoustics- Facts and Models, 616

Categorization

• QUESTION: This is a practical warning for anyone constructing experimental noise stimuli — worth confirming exact line counts and phase rules needed for faithful Gaussian approximation.