For how much music does AAC just not cut it?

Posted:
in iPod + iTunes + AppleTV edited January 2014
I found an interesting listening test here:



MPEG-4 AAC vs MP3 vs Ogg



I tried doing the compressing myself, and got the same results. While other listening tests I've read, and my own experience, put AAC ahead of MP3 most of the time, this particular material gives AAC a much harder time than MP3. These edgy synthesizer notes stay much clearer and cleaner in the MP3 encodings, whereas even at 192K, the AAC results sound gritty and noisy by comparison.



While AAC is pretty good in most cases, there's obviously still some room for improvement in Quicktime's AAC encoder. If LAME MP3 can do better than AAC on material like this, it seems that it should be possible to modify the AAC algorithm to make different choices about bit allocations when presented with material such as this.

Comments

  • Reply 1 of 13
    telomartelomar Posts: 1,804member
    Just compare the AAC to the original and I tend to find the AAC comes out closer in this instance (as it happens I own the CD that's from so had it in 3 of those formats already). That said there are times when AAC tends to come out a touch tinny.
  • Reply 2 of 13
    I wish someone would compare



    AAC file from iTMS

    AAC ripped from CD using iTunes/Quicktime/etc.

    MP3s ripped from CD using iTunes/LAME/etc.



    Most comparisons omit the first type of AAC file. The encoder Apple uses seems to not be the same one in Quicktime (high or low quality)
  • Reply 3 of 13
    I don't have any music from iTMS, so I can;t check FOR YOU.



    But pick a song from iTMS, ?-I for the info window .... this tells you the codec that was used ... compare it to a song you ripped yourself (using the QT codec) and see if they use the same codec for Quicktime and the songs on iTMS.



    I've always assumed they were the same
  • Reply 4 of 13
    Then there's the issue of level matching when comparing a song from the iTMS (i.e. you have to make sure they are the same, even a 0.1dB difference is audible as a change in "quality") since you can't be sure they generated the song from the same raw digital audio as is on the CD.



    I haven't actually ever bought a song from the iTMS, I'm not a singles oriented guy, and buy mostly albums, so for that kind of $ I want a physical CD.
  • Reply 5 of 13
    shetlineshetline Posts: 4,695member
    Quote:

    Originally posted by nguyenhm16

    Then there's the issue of level matching when comparing a song from the iTMS (i.e. you have to make sure they are the same, even a 0.1dB difference is audible as a change in "quality") since you can't be sure they generated the song from the same raw digital audio as is on the CD.



    I know how important level matching can be in listening tests, but the compression artifacts that can be heard on this one particular music sample aren't that subtle. Small level differences are often interpreted as levels of "detail", more/less bass, treble, midrange... but a 0.1 db difference isn't going to sound like a grungy noise floor going up about 10-20 dB.



    Unfortunately, I can't get the listening test web site's "AIFF" file to work... I say "AIFF" in quotes, because it doesn't seem to be a proper AIFF file, and nothing I have will open it.



    However, if I take the LAME encoded version of the file that's been encoded at a high average bit rate (I think it averages to around 224K), expand that and recompress it (yes, I know that's not an ideal way to test) with 128K AAC, the results I get sound just like the iTMS sample of the same song, with the same artifacts.



    I think that this sample from Fischerspooner's "Emerge" really is pointing out a weakness in Apple's AAC encoding, be it what you can do with iTunes or Quicktime yourself, or whatever Apple does behind the scenes to encode music for iTMS.
  • Reply 6 of 13
    The guy's obviously got a chip on his shoulder about Apple and AAC. Drawing conclusions from one sample (when there are thousands of different 'problem samples' for each of the different encoders) is either incredibly stupid or malicious.



    Quote:

    iTunes MP3 at 160kbps is better than AAC at 128kbps if you can afford the extra file size.



    And that recommendation is just bizarre. Why would you compare mp3 at 160kbps and AAC at 128kbps instead of at the same bitrate?



    More "problem" samples:

    http://www.hydrogenaudio.org/index.php?showtopic=4601&



    Bottom line:



    If your looking for "CD quality" store your music as WAV, AIFF or FLAC files.



    If you want to save a little bit of space without sacrificing too much quality use mp3's encoded as lame -aps (i.e. alt preset standard) for roughly 200kbps depending on music type.



    If you want your music around 128kbps (or lower) then use AAC as it will kick mp3's arse.



    For info, see the test results here:

    http://audio.ciara.us/test/index.html



    And remember there's no law against having certain songs encoded in different encoders and or bitrates if you do notice a problem with a track.
  • Reply 7 of 13
    shetlineshetline Posts: 4,695member
    Quote:

    Originally posted by stupider...likeafox

    The guy's obviously got a chip on his shoulder about Apple and AAC. Drawing conclusions from one sample (when there are thousands of different 'problem samples' for each of the different encoders) is either incredibly stupid or malicious.



    I'm not saying this guy's overall conclusions are a fair assessment of AAC (he does at least qualify what he says by saying "for the kind of music I listen to"), I'm merely pointing out that his samples shows that there's not-insignficant room for improvement in Apple's AAC codecs.



    If you have:



    Codec A: Sounds good with music X, bad with music Y.

    Codec B: Sounds good with music Y, bad with music X.



    Then you should be able to create a new Codec C that works more like A with X and more like B with Y, and then sounds good for both X and Y.
  • Reply 8 of 13
    Quote:

    Originally posted by shetline

    I'm merely pointing out that his samples shows that there's not-insignficant room for improvement in Apple's AAC codecs.





    If you're saying AAC is a bad perceptual codec, then you're wrong, plain and simple. This appears to be what the guy on the other end of that link is saying, and not only is he wrong, his argument is incredibly weak.



    If you're saying that perceptual codecs in general need to be continually tweaked as new 'bugs' and corner-cases are discovered, and AAC can be improved then that's correct, but hardly news, and certainly not something you can use as a stick to beat AAC with, particularly compared with mp3 and vorbis, which both have similar issues for the very same reasons (and on the same genre of music in many cases).



    Remember that perceptual codecs are a numbers game, where you try to fool most of the people most of the time. Given this fact (and the existance of lossless formats) you may find that Apple (or whoever they licence the codec from) will never consider this a high enough priority to fix as they are concentrating (along with the Vorbis developers) on much lower bitratets in the 32 -128 kbps range, where sounding pleasant is often more important than sounding exactly like the original.



    What is probably much more likely to solve this problem is a good VBR mode, so that the bits saved on easy passages can be saved up and used to encode the tricky parts.
  • Reply 9 of 13
    shetlineshetline Posts: 4,695member
    Quote:

    Originally posted by stupider...likeafox

    If you're saying AAC is a bad perceptual codec



    I've never even came close to saying that. My first post contains the words "While AAC is pretty good in most cases...". Was that not glowing enough to offset your notion that I'm on a rampage against AAC?



    Would it help to know that I've ripped nearly 2500 songs from my own CDs as 192K AAC, many of them re-ripped after first importing them as MP3 files with iTunes 3? Would it help to know that I've bought nearly 300 128K AAC songs from iTMS?



    If you're saying that perceptual codecs in general need to be continually tweaked as new 'bugs' and corner-cases are discovered, and AAC can be improved then that's correct, but hardly news



    No, it isn't news. Recall if you will the title of this thread: "For how much music does AAC just not cut it?" I'm asking this question because I haven't found any AAC music before the above example that sounds so far off to me, so it makes me wonder if other people here have run into much music that they feel hasn't survived AAC encoding very well.



    iTunes AAC does surprisingly well on the "fatboy.wav" sample, for instance, which is very tricky for a lot of encoders. Since fatboy.wav (to my ears -- not based on spectral analysis!) seems to have the "buzzy" kind of sound I associate square and sawtooth waves -- part of what Mathew says is tricky about his "Emerge" sample -- I was surprised AAC didn't do better with Emerge.



    and certainly not something you can use as a stick to beat AAC with



    There is no stick, Neo. The stick is only in your mind.



    Remember that perceptual codecs are a numbers game, where you try to fool most of the people most of the time.



    Ideally, however, you'll stay within a JND (Just Noticeable Difference) of the original for a high percentage of listeners, say 95%. For the remaining 5%, you shouldn't be too far off. Maybe that's just not possible at data rates like 128K or 192K.



    Given this fact (and the existance of lossless formats) you may find that Apple (or whoever they licence the codec from) will never consider this a high enough priority to fix as they are concentrating (along with the Vorbis developers) on much lower bitratets in the 32 -128 kbps range, where sounding pleasant is often more important than sounding exactly like the original.



    1) Bandwidths and data capacities, while increasing, are not increasing fast enough to make lossless formats (even FLAC) practical for most people.



    2) Bandwidths and data capacities are, however, increasing, not decreasing, so I don't see that much pressure to keep pushing for lower data rates -- maybe for putting music on solid-state players or radio-type services, but certainly not for the material sold through iTMS.



    I think Apple would create a lot of unhappy customers if they tried to scrimp by selling us less than 128Kbps, whereas they'd at least make a few customers happier if they improve the quality of what they can do with the 128Kbps we're already getting.



    3) I'm sure you know that, unfortunately, not all compression errors are pleasant errors.



    What is probably much more likely to solve this problem is a good VBR mode, so that the bits saved on easy passages can be saved up and used to encode the tricky parts.



    Actually, QuickTime's AAC codec already does this. It's still not true VBR, however, because only "saved up" bits are used for tricky passages, without the option of saying "okay, the saved up bits still aren't enough, so let's just up the bit count for a while".
  • Reply 10 of 13
    Quote:

    Originally posted by shetline

    I found an interesting listening test here:



    MPEG-4 AAC vs MP3 vs Ogg




    Frankly I wouldn't trust any *** who likes Fischerspooner to tell me about music quality.
  • Reply 11 of 13
    shetlineshetline Posts: 4,695member
    Quote:

    Originally posted by Chopper3

    Frankly I wouldn't trust any cunt who likes Fischerspooner to tell me about music quality.



    I'm not going by the guy's word or his taste in music... I'm listening to the audio samples for myself, and hearing what there is to hear.
  • Reply 12 of 13
    addisonaddison Posts: 1,185member
    THAT LANGUAGE IS OFFENSIVE
  • Reply 13 of 13
    ryaxnbryaxnb Posts: 583member
    this thread is going nowhere fast...
Sign In or Register to comment.