Start here: http://www.gearslutz.com/board/mastering-forum/702605-mastered-itunes-guidlines-apple-2.html and read the stuff by j_j who is this guy: James Johnston. And he has a blog.
the audio production community went anti-science rather before the republicans.
Regarding the Wikipedia article on Joint Stereo:
To say the least, it’s incomplete, and completely lacking in understanding of human hearing.
Joint stereo, done correctly ( which is impossible in MP3 due to the standard), allows the encoder to correct for a number of otherwise obvious imaging and noise-audibility problems.
Further down, post #164
“CD quality”, to me, is when the subject can not reliably recognize the reference.
Also, many tests were in headphones, and there are very different artifacts one hears in speakers vs. headphones.
There are also different classes of artifacts that different people hear. Some (like me) can not stand imaging artifacts. Some people are hypersensitive to pre-echo, but don’t give two whits about imaging artifacts. The test listeners were by and large people who dislike pre-echo, not people who notice imaging artifacts.
Don’t blame me, I was the one with the codec that didn’t have imaging artifacts.
[from Cheebs Goat] If the encoder detects information that is center-panned, it only keeps one “channel” of it and “tags” it as center data for the decoder instead of keeping identical information in both the left and right channel. [end Cheebs Goat]
That’s what the encoder for MPEG-1 Layer 2 does. (MP2)
MP3 does M/S in addition, but not in a good way. Either you do the whole spectrum as M/S or you do it all as L/R. This does not deal with time delay or more than one stereo image in the music. ’nuff said?
AAC does M/S, intensity (i.e. what is described above) both, and can do so differently in different frequency bands. It’s the minimum necessary to avoid imaging artifacts at high rates. The encoder required to drive it is a bit more complex, but would do 8 channels in real time on an old 1GHz pentium.
Hmm, can one attach .wav files? Yeah. If I can get the time this afternoon I’ll post a couple of wav files to the thread (not music, just noise and tones) that show why you must have joint stereo coding. Basically, it’s the Suzanne Vega problem, known in the science as “Binaural Masking Level Depression”.
Post #171 has some test tones (and code to generate them) to test phase issues and masking, post #190 explains what’s in them.
The missing fundamental problem, post #334
Originally Posted by SweetLossy
[j_j]: This does work for lousy loudspeakers, it’s not the same as the real thing at all, but first you have to have the original bass frequencies in order to do this. If somebody has removed them already .
Post #391 has buz.wav test tone to test a codec for various things. Not for listening.
Post #452 has buz.wav as run through various encoders and decoded back to .wav
#458 and 459 have stereo yech.wav and various encode/decode
And from another thread:
I should add: In AAC, the problems with “joint stereo coding” are corrected, the encoder does whatever requires less bits/sample, always, taking into account the relevant part of M/S psychoacoustics, called “Binaural Masking Level Depression”, also known as the ‘Suzanne Vega Effect’ to some. (it’s only one of the things going on with Vega, though)
A killer sample:
If you find yourself confronted with it, take Track 11 of Ry Cooder’s album “Jazz” and put the first 30 or so seconds through most any audio encoder.
Tough clip, that.