A little while ago, an mp3 vs. wav debate cropped up on DJTechTools forum. A lot of people saying wav was the only way to go…others saying that no one could hear a difference…others saying it didn't matter at club volumes. You get the idea.
So, I got bored and compiled an ABX test and put it up as an audio file. Had a few people respond so far, and the results are interesting. Most of the songs are modern(ish) dance leaning towards house and techno because that's what I had available in lossless formats. If you're interested, read on.
Okay, so the test is online at http://www.prism.gatech.edu/~gtg054r/soundtest.wav
. It's a ~500MB wav file, so bear that in mind if you have to pay for bandwidth. I'll probably create a FLAC or ALAC copy as well (which will be the same, but .flac or .m4a instead of .wav) if people ask for it.
It's about 45 minutes long. You'll be listening to 140 audio clips, plus some explanation of what's going on (yay…i found an sm58
on the floor…don't judge the crappy recording), plus a section of pink noise to set your volume to a comfortable level. Please avoid changing your volume during the test (certainly only even think about changing it between songs, not between individual trials). Every clip was normalized to the same volume and played with just a bit of headroom. So you shouldn't have to touch your volume control after you start.
After a citation (artist, title, etc.) you hear 4 audio clips in this order: wav, mp3, wav, mp3 to orient you to what you're hearing. Some of them are introduced, others aren't. I just forgot to copypasta all the regions and was too lazy to fix it. If it bothers anyone, I'll fix it.
And no, they're not full songs. They're 15-second clips. If I used full songs, the test would take around 12 hours to complete.
Please format your responses thusly:
1[tab]1[tab][guess, m for mp3, w for wav]
1[tab]2[tab][guess, m for mp3, w for wav]
1[tab]3[tab][guess, m for mp3, w for wav]
1[tab]4[tab][guess, m for mp3, w for wav]
2[tab]1[tab][guess, m for mp3, w for wav]
2[tab]1[tab][guess, m for mp3, w for wav]
Doing so means I can easily use a spreadsheet to do the analysis instead of doing it by hand and potentially making a transcription/computation error. Any text editor will work (or you could use google docs or excel or numbers…whatever) just make sure that you send me just that text and not other formatting information if you can.
Please PM me your results
. I'll do the analysis and keep the actual answers secret (though I'll respond to you with them once you've done it) to avoid people cheating. After a few people have done it, I'll post a description of the results (without revealing anyone's identity…so there's no potential for embarassment if your ears aren't as good as you think they are……except for me……'cuz I'm a boss…and like to be proven wrong). I'll be publishing my results right after I get enough results to make me happy.
I know it's hard to trust, but I assure you that the test is double-blind. I created wav and mp3 versions (then converted the mp3 back to a wav) and in addition to naming them 1wav.wav and 1mp3.wav, they were copied to 1A.wav and 1B.wav…then, using a shell script, I gave each file a crap ton of chances to switch names (back and forth) based on random number generation…and kept track of it in a text file that I didn't read. So, when I was flipping coins to decide whether each trial was going to be the A file or B file, I didn't know which was the wav and which was the mp3 (they were both referred to wavs inside Logic). And yes, I know how to swap things (using a temp file) so they're not the same file. If my results are different, it's likely more so because the songs came out of my collection than golden ears or anything like that. I've heard the lossless versions on the same headphones I used for the test (HD-25s) dozens of times, at least.
I look forward to seeing the results and hope I didn't waste all that time.
In addition to thinking this is interesting, a few questions have come up about some weird results I'm getting. If people think this is interesting, I'll happily discuss them as well. Hopefully they don't show some glaring fault in the test.
Edit: oh yeah…if you've done the test or are going to, please let me know whether or not you think you know what to listen for (regardless of whether or not you think you did well) and what you did the test on (what headphones/monitors/etc.) so I can play with statistics some more.
I like statistics.