Audibility of "typical" Digital Filters in a Hi-Fi Playback
Reply #543 – 2014-11-23 23:43:31
I didn't provide an extended response on AVS forum at the time when the criticism was first made, but after having had time to analyse the files I am sorry but I disagree. A null test with the files I initially provided is in fact quite successful at a moderate listening volume . My impression of the noise is far higher than you and mzil. Maybe we are subjectively different when it comes to perception of it. At the start of the track it sounded like a white noise generator next to your ear while your stereo played .On reflection, I don't think the conversion was "broken", merely that the dither that the old version of Audacity had created was somewhat noisy. That was my point. Unskilled people go and look up signal processing algorithms and implement them. Without critical listening skills, they think the job is done and ship it. This yet again demonstrates that there are two types of listeners: those that can hear non-linear artifacts and those who can't. Speaking of that, I downloaded your latest tracks. At 50.9 to 51.1 I thought the "s" in street sounded distinctly different. There was more lisping in one than the other. In the trial mode I was able to consistently tell the difference for a good sequence of trials. But when I ran with the test without feedback, I think I got down to 30% probably of chance or some such thing. I am just not motivated to try again and don't remember where the critical segment was in my original testing. So if you like to declare this a loss for me, you can . I am just too lazy to try harder and see if I can pass it. Despite the lack of corroboration of the AVS forum member's claims by way of ABX software, I took this subjective comment as a vindication of my efforts. Certainly for my ears the AVS files had a difference in "tone" in addition to the misalignment of approximately 10mS. I think these factors were enough to invalidate the AVS Forum exercise, i.e. it was too easy to tell the files apart (as I had done with a formal ABX). Yet, hardly anyone has managed to pass that test, with ease or not! Ultimately what I like people to take to the bank is just that: that we don't hear the same. The data is 100% compelling in that regard. And very easy to verify. Anyone who disagrees should try to duplicate our results. They can try to find the flaws just the same. But I suspect with full knowledge of all of that, they don't have critical enough listening skills to find and hear the difference in DBT ABX tool. And given that proof, you cannot extrapolate from your hearing to the rest of the people. Just because you don't hear the difference doesn't mean it is inaudible and sufficient justification for you to go around and accuse people are being wrong to observe otherwise.I am unable to explain how the tonal difference arose. Why AVS Forum could not simply have provided time-aligned files capable of being audibly nulled at a moderate listening gain (by inverting one of the files) is a little surprising. I guess it was due in part to lack of familiarity with this type of exercise, and the importance of matching the files. Not only were the files not time aligned, but even when time aligned and with one file inverted, they would not null to silence at a moderate gain setting. Actually they duplicated a real-life situation. Took a professional audio workstation tool, Sonic Solutions, and converted the files to 16/44.1. Precisely how real music is produced. That in double blind tests we could tell the difference it means that what people observe in the wild can very much be true. That transparency is not there. It matters not that the timing difference may be the reason. What matters in this context is that we are wrong. We are wrong to say they are imagining things when they compare the high res to 16/44.1. We are making idealized assumptions that are not true in reality. So I think it is good that we discovered how professional resamplers are not the animals we think they are. There is no visibility into their design as I have mentioned. I imagine hardly any music is produced using Sox resampler. Engineers use the professional tools. And if those tools produce non-transparent results, then we should get the masters and not be subject to this. These are 100% logical and defensible conclusions we can draw. Anyone who doesn't want to go there has fingers in the ears and in denial.