Is there a bitrate where all people tested can not distinguish AAC from an uncompressed file? Has a test of this kind been done?

I'd been leaning towards lossless archiving, but given that I want at least one if not two backups I'm thinking about archiving two or three copies of AAC. Should be supported by apple for a long time, so it seems stable enough, and with the money I'd save on HD space I could buy an SACD player and more media. I listen through a very high quality system (ibook-->metric halo ad/da-->audio note soro SE integrated-->audio note an/e-sp speakers) and I don't want to compromise quality but I don't want to use more space than required either.

Thanks very much for any thoughts on this.

Such a test is extremely difficult. One can generally find clips that are not transparent at 160kbps without too much problems, so 170-200kbps is a good guess. This is assuming a good encoder (Nero or iTunes).

Everyone's hearing is different, so you'll have to conduct tests yourself to find the correct bitrate. For Apple's AAC encoder, 192kbit is probably a good place to start your testing.

However, given your level of paranoia (probably similar to mine ) and equipment, you'll probably find yourself unsatisifed at some point unless you archive to lossless. BTW, if you're planning on moving to SACD, be aware that you can't (easily) rip those to your computer.

Theoretically, you can't say any lossy format at any bitrate is sufficient to provide "universal" transparency. As long as something is lossy, there is always some chance that someone will hear an artifact.

If someone is so bothered about the chance of hearing artifacts they really should stick with lossless...this way they are guaranteed 1=1.

Yeah, paranoid is a reasonable word, but if it were the case that no one on any system has been able to distinguish 224 or even 320 AAC from WAV then I might be willing to go that way.

I thought it was *totally* impossible to rip SACD: has someone found a way around this already?

My vision of my software archive would be: ALL redbook ripped to HD, some vinyl discs, and some SACD discs.

Thanks again.

You can connect a SACD player to your soundcard's line-in and record from there.

Well, yeah, but I'm talking about an SACD copy.

Would you agree that this figure is surprisingly similar for most audio codecs?

With the caveat that some have many more unfixable (i.e. not transparent at any almost bitrate) samples than others (e.g. mp3 vs MusePack), it seems you need a setting which averages about 200kbps to deliver almost always transparent results?

While the bitrate for "very good" results has fallen dramatically between mp3 and HE AAC, the bitrate for "transparent" results has changed less it seems.

Or are we just becoming more critical? :-)

Cheers,
David.

P.S. is there a list of AAC problem samples?

Because storage limits are always increasing over time, I expect transparency freaks to stick with lossless (averaging 800-900kbps) whereas everybody else will prefer moderate-or-low-bitrate (<128kbps) "good enough" lossy. Like our economic system, the middle will slowly disappear. 192-256kbps lossy will not offer enough improvement over the "good enough" bitrates to get those users to "trade up" while the paranoid folks will always be critical of lossy codecs.

Personally, I am not interested in efforts to get lossy sounding transparent at higher bitrates...I instead want 64kbps to sound great.

...but so is my music collection!




x3 for surround, x2 for "hi res" etc etc!




You could apply that argument to almost anything in life, but it rarely happens. It's like saying only junk and audiophile equipment sales should survive and cheap equipment gets better - yet millions still buy "mid-hi" (however you define that).


When is the 1000GB mini iPod due? 6 years? So until then, I have a use for transparent lossy.

When it arrives, you'll be paying extra for the HDTV output this device will almost certainly have (what else are normal people doing to do with 1000GB?!). I suspect good affordable audio-centric devices (for as long as they continue to exist, which may not be long!) will mainly be based around lossy audio, so there's great merrit to taking formats which give "very good but not transparent" audio to everyone, and using them to give transparent results.

Maybe the kind of debates we have at HA will seem less important, but they'll still be literally millions of people using low bitrate audio codecs at comparatively high bitrates to reach transparency. It's just so easy for people - it's an extension of what they know.


Of course you know the killer argument against what you've said? Even before mp3 was invented, people were already saying something similar! "Transport and storage costs will plummet, while capacity will rocket - why are you bothering with this lossy stuff? By the time you get it working, they'll be no need for it!" - yet the MPEG patent holders have somehow managed to make a profit on their pointless investment.

I predict this will continue. There's as much reason to have small "transparent" encoding, as there is to have smaller "very good" - your argument should do away with both, but I don't think it will kill either.

Still, I could easily be wrong - when do you want to meet back here and check? 6 years?


Now, back to the original question...

Cheers,
David.

100GB ipod in two years max.
1TB ipod...5 years?

So are there no listening tests you all can recommend to determine the trasparency threshold for AAC?

Why don't you try your own. Use what's transparent for you.

for me with a cheap headphone and soundcard, 160kbps is indistinguishable from uncompressed.
For bitrate above 192kbps, they shouldn't have annoying artifacts............it won't sound to bad if you've lost your original uncompressed file

It is possible to prove hearing of artifact that is adding sound energy that wasn't there before.
It is possible to prove NOT hearing artifact that is removing sound energy that wasn't audible anyway.

It has been proven times and again that humans may hear artefacts that aren't there at all.
It has been proven enough that humans may hear addition artefacts that are really there.

Why post this? Not to argue your point, because thats the state of affairs today. But it is conceivable that some day there will be solid proof of lossy codec that provides universal transparency. For that you'd need solid proof of artefacts that are inaudible physiologically.

But, the real reason I chimed in, is that your post made me thinking about difference between artefacts that ADD to the signal and artefacts that REMOVE from the signal. There is an important difference between these.

Human hearing has developed to survive, and for that detection of faintest sounds was important. Brain processing and nonlinearity of hearing is getting so sharp that it becomes ridiculous - we can hear things that aren't there.
On other hand, very sharp detection of missing sounds isn't important to survival, and this ability is undeveloped. Thus, quite often, when artefact of lossy codec is removing signal, not only do we have hard times detecting it missing physiologically, but our brain would fill in the gap with our expectations.
Thus, lossy codecs have much better chance of reaching universal transparency than say analog chains that ADD distortion products.

Do we also see things that are invisible?

No, seeing things that are invisible would be a paradox

we do, however, see things that arent there --> optical illusions.

What about AAC at 448 KBps? Or Psytel at 512 KBps?

Edit: OK, that was dumb. Who ever does AAC at 448. That's for multi-channel right?
It would be smarter to use a lossless codec at ~768 kbps than 448, even though it's almost twice as big, if you're an audiophile.

although maybe someone could compare 448KBps AAC to Flac?

It's difficult enough to conduct high-bitrate (say 192 and higher) listening tests as is, since most people should be unable to distinguish these from the original music. Going up to 448 is just.. well, I reckon there's a reason why extremely-high-bitrate codec profiles are dubbed 'insane' or 'braindead' I can't even think of a word to describe 448 in this manner

You already realized that it was dumb, this was just to point out that it would be extremely difficult for anyone to distinguish such high-bitrate files from the original (except perhaps on hypothetical problem samples which would be known to completely break the psychoacoustic model).

At bitrates that high, switching to WavPack lossy or the likes should produce better results than using a psychouacoustic-model-based lossy codec.

How does it come that i have now read in multiple occassions that "_Advanced_ Audio Codec" is transparent for everyday usage at 170-190kbit, with LAME 3.96.1 being transparent at about 190kbit?

I mean, WTF? I trade in the massive popularity and compatibility of MP3, builtin DRM-capabilities, and what do i get for that? Transparency are 20kbit less average bitrate??

MPC did that already long ago. Someway, this sounds similiar to Vorbis - the only real advantage is in narrowband-bitrates. Maybe this is the reason why AAC is so uninteresting for so many people:

want compatibility? MP3 delivers
want transparency at lower bitrates? MPC delivers
want good quality at narrowband? Vorbis delivers

AAC seems to be only mediocre at the first two, good at the third(narrowbad) while being the only codec which is ruled by proprietary encoders.

- Lyx

It will be interesting to see the result of the next AAC listening test. QuickTime 7 should bring some exciting improvements, as will Nero 3..

mp3 wasn't very good in the early days either, you know

AAC compatibility is improving, since it's got the biggest online music store behind it pushing hard.

Also, AAC is a relatively new format. Or rather, it's been around a while but no serious effort has been put into AAC encoders until Nero started using it and then iTunes came along. The point I'm making is that MP3 encoders weren't great when they first came out and several years later you get something like LAME. The capability of AAC is greater, even if current encoders don't reach the full potential of it.

So, the main lobby when it comes to hardware is the people who have an interest in DRM.


In other words, it isn't ready yet to compete with the other audioformats, because it doesn't have something superior to offer _right now_.

So the only advantage it offers right now is: DRM, hope and very good narrowband-performance.
Sorry, thats not enough for me.

Hmm.. I would say it is already quite well-tuned and has been for over a year. It's just that QT doesn't offer a VBR mode yet, so many people select 192kbps to be "safe".

So what? You can create DRM free AAC's using any of the AAC encoders. Don't like DRM, don't purchase AAC with DRM.

Don't confuse DRM with the format itself. AAC sits in MPEG4 containers. There's no built in DRM in the spec. Apple's Fairplay DRM is their own extension to the container, and it's not there in AAC's you encode yourself. AAC is just as DRM free as MP3 is. If somebody wanted to come along and make a proprietary extension to wrap up MP3 in a DRM format, they could do it just as well as Apple has done it with AAC.

Edit: And frankly, I don't see why DRM is such a big deal to you. The Fairplay DRM is well and truly cracked and they ain't repairing it anytime soon. And more to the point, the very concept of DRM is fundamentally flawed anyway. It will always be cracked. It cannot be made uncrackable. So BFD, let 'em wrap DRM around the music, it doesn't affect me any. I'll just strip the DRM back off and get on with my life.


But it is superior _right now_. On most samples, I get transparency (for me) at a much lower bitrate than I do with MP3. Simple as that, really. It also works on all my equipment (iPod, stereo, PC, etc) and I'm quite happy with it.

Is it superior enough for me to re-encode all my music in it? No, not yet.
Is it superior enough for me to switch to by encoding all my new music in it? Yes.

You answered to something different than what i did say. I said that AAC's main lobby for hardware-support is DRM-friendly people. And, being a sane consumer, this of course does worry me when evaluating the future. Because it will be the same people who decide the path for it in the future. They may not have done anything evil yet(of course not, that would be quite stupid in the run-up-phase), but experience gives me all kinds of reasons to mistrust the DRM-lobby and content-_industry_.


Since this statement goes contrary to the experience of most other people here, there are 3 possibilites:
1. your hearing is very different than the one of the others
2. the "much lower bitrate" is not achievable with reasonable effort, because there are still cases where 192kbit are necessary - therefore requiering manually testing every single encode to be sure that enough bits were spent. In other words, this right now is a "virtual feature".
3. everyone else was wrong and your observations are right

- Lyx

Does anyone know a good killer sample for AAC?

I tried to ABX the "castanets" with AAC @ 192 kbs with my equipment and didn't succeed (9/16). With LAME at api I can even ABX the sample with 16/16!

Yes, it is usually called a hallucination.

Still lacking hardware support and falling prices for storage seem to erode the AAC faction - AND then there also is all that DRM talk which makes people shy away... Recently, I purchased my first portable player and found out that actually the options you had for AAC playback have shrunk instead of become bigger. Add the fact that AAC's superiority over MP3 gets more impressive at lower bitrates but portables tend to feature 512+ MBs storage today, instead of the typical 128 MBs of the old days, and you run out of sane arguments for using AAC. It is a pity, the industry shot themselves in the foot. Even the sternest AAC fans, guruboolez and rjamorim have wandered off for those reasons.

Myself, I don't have logical reasons for sticking with AAC; I just trust Nero's 'extreme' preset more than LAME's 'extreme' preset and I consider lossless audio a bit of a waste of space.

Perhaps AAC support will catch up a bit when players for the successor of DVDs come out that should be able to play AAC, too. That should prompt in turn the market for AAC portables. But perhaps not, and in that case I'll just stick with transcoding AAC 'extreme' to MP3 'extreme' for use on Zen Touch player and indulge in my lunacy...

I could abx it at aac 128kbps, but not 192kbps. It was a breeze with api 3.96.1. I also tried for fun with iTunes mp3 @ 192kbps, and that sounded almost as bad as the xing sample to my ears..


edit: quickly tried "fatboy" and that seemed to break the aac encoder pretty badly at 192kbps.

I remember the main goal of AAC was CD quality @96kbps (like mp3 @128kbps ). Maybe it´was a dream I had...in 10 years perhaps it´s true

A hallucination isn't invisible.... Your mixing up the terms here...

novocane, I remember hearing on one of the G4TechTV shows (The Screen Savers maybe?) that 96kbps near-CD quality really was the goal of MPEG-4 AAC.

The way I see it, at least AAC is a slight improvement (however slight) over MP3 at similar bitrates...unlike WMA which seems to be a step backward for everything except super low bitrates (<96kbps). :-P

I find it to be a marked difference between AAC @ 96kbps and 128kbps, at least with the QT encoder. It also says in the help file that they don't recommend using less than 128kbps for music. In other words, I don't think they aim for transparency at bitrates below 128kbps.

That was a joke, btw.

With my speakers, I can tell the difference between 500kbps ogg and lossless. What's my point? Theres no such thing as real transparency, theres always someone that can tell the difference (eg: me). Universally transparent is an oxymoron. It's impossible. Besides, that's no what you want anyways. You want whats transparent to you. Do a listening test and decide for youself.

You're seriously able to ABX a 500kbps file from the source? Whoa.




Good question; I agree that 192kbps is enough for castanets. I thought long and hard about this and even tried a few. I couldn't even ABX the infamous fatboy.wav "zipper" sound at 192kbps QT AAC-LC (although it had horrible artifacting at 128kbps). Can anyone find or think of any other samples? I'm really curious too.

For some people (lossy) transparency is impossible. ABXing doesn't tell the whole story, because when you've compressed everything yourself you do know how it's been compressed, and for those perfectionists amongst us any lossy file will (seem to) sound worse, simply because they know it's lossy.

Are you talking about a placebo effect? If so, I'm not sure that's an objective statement about audio quality at all.

Yes... I'm prejudice in my blind tests, lol.

The question wasn't when is it objectively transparent, it was when it's universally transparent. My point was that universal transparency isn't really relevant to an individual choosing how to archive files - he could ABX it personally, use an extreme preset or leave it lossless, but since it's impossible to be objective while listening outside of a blind test, psychological effects are going to play a part in the listening - I don't ever do lossy compression for that very reason - I've no idea how high I can ABX mp3/ogg/aac from the original (although some time when I have a nice PC audio set up I plan on doing some tests...) but when I'm listening normally it doesn't sound as good to me, so I only use lossless formats unless I'm sending stuff to other people to listen.

I think there have been tests done, and if there are I'm sure someone else could link you.