After correcting for a 0.2 dB difference in level, the native 24/96 and SRC versions of the test files are again ready for downloading.
As many of you know, I'm conducting an informal experiment with the AVS community to investigate whether or not true high-resolution audio files—with information beyond what a Redbook CD can accommodate—can be reliably distinguished from the same files that have been downconverted to CD specs. The parameters of the experiment can be found here
Shortly after I made the audio files available for download, several members reported that one of each pair of files was 0.2 dB lower in level than the other. This was determined to be the result of subjecting a copy of the native 24/96 file to two sample-rate conversions—first down to 16/44.1 and then back up to 24/96—in order to hide the fact that it has a lower resolution. It turns out that the sample-rate conversion software—Sonic Solutions Sonic Process in this case—lowers the level of the file being converted by 0.1 dB to provide some headroom so the calculations will not cause any clipping. (Most sample-rate converters do this, not just Sonic Process.) Because the "hidden" 16/44.1 version went through two conversions, it was 0.2 dB lower in level than the native 24/96 version, which is close enough to the just-noticeable difference (JND) that listeners might pick the native high-res version on that basis.
A number of solutions to this problem were proposed, including a couple of rather complicated, multistep conversion procedures. We also experimented with normalizing both files as well as simply increasing the level of the converted file by 0.2 dB and decreasing the level of the native file by 0.2 dB. In addition, we considered raising the level of the file that would be converted to CD specs by 0.2 dB before doing the conversion, but there wasn't enough headroom in two of the three files to do that without clipping.
In the end, neither Mark Waldrep (aka Dr. AIX), who provided the files, nor I believe that the procedure by which the levels are matched makes much difference for this experiment. Still, I wanted to process the files as little as possible, so I rejected the idea of normalizing all of them. Because I subscribe to the "cutting is better than boosting" school of thought, I decided that the best course of action was to decrease the level of the native 24/96 files by 0.2 dB rather than boost the level of the converted files. Yes, that decreases the dynamic range of the native file, but only by 0.2 dB, which still leaves it beyond what CD can handle in each case. I have checked all three pairs of files using foobar2000's ReplayGain Scan function, and each pair is now within 0.01 dB of each other, well below the JND of perceptibility.
Of course, some AVS members will object to this approach, but I seriously doubt there would be unanimous approval of any solution within the AVS community, so I made an executive decision to do it this way. Those who believe my solution is unreasonable are free to simply not participate and to argue their position in the comments here and in the other related threads; all I ask is that you keep it respectful and not personal. Just keep in mind that this is an informal—and, I hope, fun—exercise that invites AVS members to make their own determination about whether or not the extra information in a bona fide high-res audio file makes a perceptible difference over the same content that conforms to CD specs.
Some members have also suggested that the results of the experiment should not be limited to those with audio systems capable of reproducing the extra information in the native high-res files. It's fine with me if you want to submit your determination of which files are high-res regardless of your system's capabilities, but all submissions must include a complete list of all components used in the test, including makes and models, as well as how they are connected together. I will compile the results separately for those with truly high res-capable systems and those with systems that can reproduce no more than CDs can deliver.
The new audio files are now available by clicking on the links below. As before, we've zipped each file so all browsers will initiate a download rather than trying to play them, and we've combined all six files in a single zipped package. The "A" and "B" assignments are new, so they are labeled "A2" and "B2" to make sure there is no confusion with the previous files. To make the assignments, I used the random-number generator at random.org, which is based on atmospheric noise rather than a pseudo-random algorithm—hence, it is truly random. I set the range from 1 to 100 and let it generate a random number for each pair of files—if the number was odd, the native high-res version became A2; if it was even, the native high-res version became B2.
So let's try this again, shall we? If you have already submitted your determinations, please do so again with these new files, and be sure to include a complete description of your audio system. Remember, this is not an ABX test—simply listen to each pair of files as many times as you wish, at the same volume level for both files, and see if you can tell them apart. And please don't go poking around the files with audio software—just listen. If you can, decide which one is high-res, A2 or B2, and send me your determinations by PM with the subject "AVS AIX HRA Test 2."
Thanks to all who identified the level-difference problem and offered suggestions about how to correct it.
All in one:
AVS/AIX HRA Test Files 2
Just My Imagination A2
Just My Imagination B2
On The Street Where You Live A2
On The Street Where You Live B2
Like AVS Forum on Facebook
Follow AVS Forum on Twitter
+1 AVS Forum on Google+