SRC Comparisons

It would be good to have it as another user choice for render. REAPER went for 64-bit processing early-on, including SoX resampling would fit in with its ethos for audio excellence. It's ok consigning the low-level artefacts to the tinfoil-hat brigade, but that could be said about dithering too. ;) If Audacity uses SoX, I presume there would be no licensing issues with inclusion in the REAPER install? >

You need to do a proper double blind ABX test if you want to have your posts to have any relevancy.

That's not a proper double blind ABX test. There are apps for that. Even a free ABX plugin for foobar, for example.

The human auditory sense system may be unreliable, but at the same time it also is capable of detecting many things for which no test has yet been devised, and can't be explained by current ones. So it makes no sense to pooh pooh someone's diligent observations. After all, what today is called "conspiracy theory", a few years ago was known as "investigative journalism".

The problem: You sample rate converted something and you hear a difference now from the original. This is a problem because it stops you in your tracks and makes you start looking for what to fix. The deal is that if pushing a button or clicking a preference setting can fix it - this doesn't cost anything and thus you are driven to find the solution. You have come to a task in audio handling where you need to make a copy of something and you don't want to degrade anything from a wrong button push. You want this distraction solved so you can get back to work and trust your ears again. This can be difficult when the difference in sound is within your perception tolerance. (You know... that perception tolerance that lets you carefully A/B 2 eq settings when the eq has been muted the whole time?) Do this: If you want to illustrate any errors or shortcomings of your recording/conversion 'device', make multiple generational passes so any error is multiplied and well within perception. Make 100 generational passes with the sample rate conversion. Now compare the results. The one with the least artifacts and destruction after 100 passes wins. Now you know which "switch setting" on your "device" works best and you can get back to work. Aside: It's been my experience that in already compromised/damaged audio, the fidelity loss starts to 'snowball'. Sample rate conversion of full program audio that already has some generations of conversion seems to be magnified. Capturing audio with analog generation loss degradation takes another hit from lower res digital conversion vs. HD sample rates. It might seem intuitive that a lower fidelity source might be just fine with an mp3 capture (for an extreme example) but it actually works the opposite. The more damage, the more pristine capture you need to preserve what's left (as some of the content is now buried in the decimal dust with a lot of noise). The visual analogy would be the post office copy machine.

Why is that people always say they've "done the tests" but several years ago, files are now missing etc...?

The mis-information, or lack of understanding, to what the 'resultant' output of a Null test shows is the more serious issue. A complete 'null' will yield an absolute 0 resultant. Any test result that shows a 'partial' nulling does NOT mean that this difference signal is down in the 'noise floor'. [This is not dithering]. Any difference is ADDED to the subject audio file. These 'difference' may be quite small, and might not come to the level of things like harmonic saturation from transformer or tubes for example. Can the 'difference' be heard? I think that THAT is an non-issue for several reasons. Only through personal listening/testing can the significance be determined/appreciated. Whether 'you' hear it or not, does NOT really matter to anyone else. Just to keep it interesting .... remember, 'Critical Listening' is a trained skill, and is similar to a musician practicing his/her instrument. Again, to the point. The simplistic view toward a Null Test is one thing ... the misunderstanding of what it shows is quite another.