XLD is app on Mac OS X that uses CoreAudio to encode AAC. Basically that's what iTunes was using.

My parameter for XLD was:
True VBR
Encoder quality: Max
Target quality: ~111
Resulted bitrate is around 256kbps, sometimes over 300kbps

Nero 1.3.3.0 was used for my older encodings
And I used Foobar2000 to encode and choose the best settings I could. And resulted in ~200kbps file.

I also had some files encoded with iTunes 7.x and they look like Nero.

What does this mean?
I know eventually for SQ we need listen test. But these graph suggest dramatic improvement of compression? Or not.
If it is indeed improvement, I probably have to ditch all my nero encoded files...And redo all the rip and such. Painful!

Is this a bug in the encoder?

Also I'd like to add that, iTunes Plus from iTunes Store has a clear cut off line at 22KHz.

Update: I tried to convert a few Apple lossless files to iTunes Plus with iTunes. And did not see any obvious cut off above 22KHz.

Here we go again...

Please read thread:

http://www.hydrogenaudio.org/forums/index....showtopic=73970

from beginning to end. At the end you will see that THERE IS NO POINT IN CHECKING SPECTRUM PLOTS. Only a listening test shows if something sounds closer or more distant from the original.

I agree with the OP, actually.

Did you read the thread that I pointed to? Did you try listening to those files?
There you can see that files with lower cutoff and lower bitrate sound closer to originals.

I would care about that thread if it was relevant. But it isn't. Did you read the OP?

He was clearly not ascribing a preference one way or another - he was simply pointing out that the highpass with the new encoder was much higher than he was expecting, was much higher than what he say with Nero, understood that such highpass behavior has been correlated with poor quality in the past, and wanted some sort of justification as to what was going on. You haven't really clarified any of that.

I share the same curiosity, even though I freely admit that such speculation does not really mean anything in the context of encoding decisions. But it can certainly suggest strong/weak spots in such testing.

I would say a bug in encoder, but who knows...

iTunes Plus = Constrained VBR, 256 kbits, high quality. Not True VBR.



A dramatic improvement would have to be an audible phenomenon, in my opinion.

On the first look it may look like Nero was cheap. Why can't it keep the HF range even in the highest bitrate settings?

But look at it like this: even with the highest setting, the encoder still has to throw out stuff. Else you would need > 600kbps rates and get lossless.

Preserving very high frequencies takes considerably more storage space for considerably lower (to non existing) audibility. The graphs might just look exactly like that, when a good psymodel is doing its job*, even with a very high quality setting. If you find someone who can actually hear a difference, that would of course be another case.

* Well, the ideal (not in a practical but ideal, theoretical sense) psymodel would probably not have this straight cutoff but would fluctuate more freely at the upper end). But still, we should discuss this with samples, not pictures.

The thread that I pointed to is relevant. It points out that that introducing higher cutoff doesn't bring improved quality. It also points out that there is no 'hard' cut off in Nero 1.3.3, but that only quiet parts of high frequencies are removed. And it points out that the upcoming version of Nero will reintroduce quiet parts at high frequencies only at very high bitrates.

Yes iTunes Plus is not true VBR. But true VBR has been implemented in the latest CoreAudio.

Of course, I am not saying the new iTunes encoder will produce better sounding files. As I said, listening test is needed. Maybe some other people who's done a listening test and design of those test can do something, but I have no experience in listening test.

I am just extremely curious what is going on. I definitely expect it to cut off HF but it doesn't look like it, so I don't know what it is cutting off to actually compress the file. So I also suspected it may be a bug?

Please don't be offended that I said do I need to discard my nero encoded files if the above charts indicate improvements.
But to anyone who's interested in audio encoding, I believe this is a curious case. It is probably never seen in any lossy encoder?

True VBR has been available in CoreAudio and Quicktime for years.

It is not hard to conduct a personal listening test. You can get ABC/HR here:
http://ff123.net/abchr/abchr.html

It is not necessary to remove something to get higher compression. It is in most cases small modification of values that does the trick. These small modifications are hard to spot on a spectrum plot.