Windows 10, Audacity 2.3.3
I’m trying to compress a track, then take the difference between the compressed track and the original, in order to isolate the compression artifacts. Like in this video (volume warning): https://www.youtube.com/watch?v=DwpS7gOt554
I’ve found a few older forum posts that mention using Effect->Invert on the original track to subtract it from the other, but it doesn’t have the effect that I’m looking for. It creates a very subtle change, and doesn’t wind up cancelling any sound.
For reference, these are the posts:
I noticed that there are two macros that have related names: Difference and Subtract. But I don’t know if they’re intended for this purpose, or how to use them correctly.
Any suggestions are greatly appreciated.
Are you using MP3?
It creates a very subtle change, and doesn’t wind up cancelling any sound.
Try making an identical copy and subtract that from the original. If you do it correctly it should cancel to dead-silence. Once you’ve got it working you can try the compressed copy.
Did you make the compressed track so you know your original is “the original” for the compressed track? A “similar” recording is not good enough… If you record yourself saying “hello” twice, they are different recordings and subtraction sounds exactly like addition (regular mixing).
MP3 adds some silence to the beginning and end. That delay at the beginning will mess-up the subtraction. Audacity might remove the added silence in the beginning, but zoom-in until you can see the individual samples so you can align them “exactly”.
The levels also have to match or it won’t subtract to silence. However MP3 makes some peaks higher and some lower so that’s one of the “artifacts”. (You can’t hear those slight peak differences until you subtract.) If you made the MP3 yourself the overall loudness should match and you shouldn’t adjust the volume.
Note that “the difference IN sound” is NOT the same as “the sound OF the difference”. Subtraction can reveal a difference in ''data" when there is no difference in sound. Just for example - If you add a few milliseconds of silence to one copy to slightly delay it, it still sounds identical. But when you subtract you get a HUGE-LOUD difference file with a comb filter effect. If you have experience with this stuff you know the comb filtering is a result of combining the delayed and un-delayed audio but it doesn’t sound like delay.
However, if you get silence that DOES prove that there’s no difference.
Also, the difference isn’t necessarily “artifacts” (added sounds). MP3 tries to throw-away details that are masked (drowned-out) by louder sounds. So the “difference” also includes those sounds that were thrown-away because you probably couldn’t hear then anyway (until the louder sounds are removed, “unmasking” those details).
Since lossy compression can make changes that you can’t hear until you subtract, the BEST WAY to evaluate compression is with a blind, level-matched [u]ABX Test[/u].
Thank you so much Doug, I finally got it working from your advice.
Specifically zooming in until the individual samples were visible and aligning the waves exactly made it work. I didn’t realize you could zoom in that far.
To answer your other questions, yes, I made the compressed track myself. But the compression algorithm must do something to the exact timing, as you mentioned about MP3. It seems like other formats add or remove more delay than MP3 does, so it isn’t a consistent amount. Kind of interesting.
Thanks again for the help, I didn’t respect just how precisely the tracks have to be aligned.