Subject: Re: Gaussian vs uniform noise audibility From: John Hershey <jhershey(at)COGSCI.UCSD.EDU> Date: Thu, 22 Jan 2004 16:20:20 -0800> Are you really sure that hearing can always be completely understood by > means of linear treatment of an arbitrarily chosen section of time? Do you I was just trying to clarify what assumptions lead to the statements being made. However I think this statistical analysis is probably misleading, as is the interpretation in terms of the Fourier transform. It's not as if the auditory system would have to measure higher-order statistics of the signal as such over time-- one signal would just sound different from another because it would have a different spectrum at any given time, even if the means, or some other statistics of the spectrum, are the same for both signals. Likewise with respect to the Fourier transform -- one has to specify at least a window size to get meaningful hints about hearing. On the other hand, thinking about it statistically is useful for forming intuitions about signals. I otherwise wouldn't have thought about the fact that non-Gaussian i.i.d. noise can introduce higher-order statistical dependencies in the Fourier transform, even though the spectrum is the same.