Hello,
I have just posted the results of my MP3 decoder test (http://blog.szynalski.com/2011/03/16/itunes-10-2-1-fails-to-decode-mp3-files-properly/), covering iTunes, Winamp, Windows Media Player and Foobar2000.
The TL;DR is that all MP3 decoders are NOT created equal. Specifically, iTunes produces significantly different output streams than the other players.
Best,
Tomasz
Spare your time, guys. The guy concludes "distortion" just because it doesn't match the output of an old Fraunhofer decoder exactly and the difference file sounds "distorted" .
Spare your time, guys. The guy concludes "distortion" just because it doesn't match the output of an old Fraunhofer decoder exactly and the difference file sounds "distorted" .
This is a decoder comparison. Checking differences with a reference is correct, all decoders should produce the same output (with some minimal and specified differences allowed).
It'd be interesting to see 1) what happens under OS X, and 2) what happens with AAC.
I'll test both of those scenarios this evening when I get home, unless someone else wants to take the baton before then.
This is a decoder comparison. Checking differences with a reference is correct, all decoders should produce the same output (with some minimal and specified differences allowed).
My point is, one cannot conclude "distortion" in this case from the sound of the difference file. This might also be the effect of a limiter. I don't remember if it was Apple's AAC encoder or decoder, but it is also happening there and no audible harm.
Did some quick tests under Windows XP w/iTunes 10.1.2.17 just having iTunes decode to WAV (i.e. no loop-back):
1) I get similar difference results as tszyn when comparing against either fb2k's or LAME 3.98.4's decoded output (whereas both of those essentially null against each other).
2) Difference file isn't nearly as "busy" with CBR 256 as it is with V2.
3) iTunes is correctly removing the encoder delay and padding from the decoded MP3, but in addition, it is also cutting off exactly one frame from the end (1152 samples) - verified with a few ~10 sec. MP3s...shades of the ALAC decoding bug (http://www.hydrogenaudio.org/forums/index.php?showtopic=84101) from late last year with iTunes 10.0.
As such, when I did my inverted mix pastes, I shortened fb2k's/LAME's WAVs by 1152 samples before pasting.
@googlebot: "Distortion" is any unintended change to the signal, so in the strictest technical sense, yes, iTunes' MP3 decoder is adding distortion when compared to the output of several other decoders...*why* is the question, and one that I hope we can find the answer to, or at least bring to Apple's attention (they did, after all, address the ALAC decoding bug in the next release of iTunes, 10.1).
Here's a spectral view of the fb2k<->iTunes invert paste (audio clip is ~10 sec. from Elbow's "Grounds for Divorce"):
(http://img20.imageshack.us/img20/1803/groundsfordivorce10secf.png) (http://img20.imageshack.us/i/groundsfordivorce10secf.png/)
...and for a more visible example, 10 seconds from Ministry's "Filthpig":
(http://img23.imageshack.us/img23/5719/filthpig10secfb2kitunes.png) (http://img23.imageshack.us/i/filthpig10secfb2kitunes.png/)
When I tested this a decade ago, quite a few decoders made these small mistakes above 16kHz - including the older FhG ACM codec.
The ones with errors above 16kHz are scored "2" in this table...
http://mp3decoders.mp3-tech.org/objective.html (http://mp3decoders.mp3-tech.org/objective.html)
Strange that iTunes have re-introduced something that's like this old bug.
Cheers,
David.
What does the waveform look like below 16kHz? Could Apple be trying to interpolate missing harmonics (meaning was this a deliberate choice, with the hypothetical goal of improving the sound), or an unintended consequence of something else?
- M.
Rather looks like just a sneaky old bug to me now. If Apple would have done anything in that direction they would have let you know.