Audibility of "typical" Digital Filters in a Hi-Fi Playback
Reply #922 – 2014-12-03 18:50:48
If we are objectivists, we better be objectivists all the time. Good. Then can we agree on these things? Or not? 1. The BS paper reports a listening test where the filtering was detected 10% better than chance (which, in the number of trials performed, is statistically significant) 2. The different scenarios tested all gave broadly similar results overall. Agree.3. Some differences between scenarios became larger when the data was post-sorted and re-analysed, but the rigorous way of investigating this is to re-test these apparently better conditions, not post-process the data looking for the better conditions after the fact. It is routine to perform such post-process analysis. Excluding Simpson's paradox is a good thing, not bad.4. There was no re-test, so at this stage we shouldn't claim differences between filter types, dither, etc. But we can say that the claim that "no double blind test has shown differences of this type" or that "the distortions are below JND" have been falsified.5. The theories as to why/how the filtered samples were differentiated from the original samples are just that: theories. It would be possible to design listening tests to probe these theories, but this has not been done. At this stage, we shouldn't claim the filtering was audible because of X, Y or Z. Correct.6. No one has done a listening test to probe the audibility of ultrasonic ringing itself. (AFAIK - do you know different?) This may be indicative of that.7. Audible IMD from ultrasonic signals is readily demonstrable under certain circumstances. In double blind tests with music as opposed to near clipping tones?8. The worst audible IMD from ultrasonic signals is far more easily audible than the worst effects of ultrasonic ringing from the "worst" "CD" sinc filter. There is no data on record to represent this. The testing in Stuart's paper used music. The only test provided for IM distortion has near clipping ultrasonic tones which has no music profile that matches it.9. Measurable IMD from ultrasonic signals reduces with ultrasonic signal level, often almost proportional to some positive integer power of the signal level, rather than linearly. (i.e. it falls away surprisingly quickly). Correct. And given the fact that the level of ultrasonics also drop sharply on their own, makes the IM distortion an unlikely cause.My objective conclusion from all the evidence is that neither mechanism should be readily audible, but in a test where people were just about able to hear some kind of difference, that difference could be down to one, the other, both, or neither - pending further investigation. Agree.