Lend me your ears! An AAC two-pass listening test.
The subject of recompressing compressed audio files has come up around here a few times -- sometimes in the context of people mistakenly thinking that they might improve the sound of an MP3 file by converting it to AAC.
Extra generations of compression definitely aren't a good thing, but just how bad are they? I've put together a little listening test to explore this issue. I've got a sample that's been compressed once via iTunes 4.0.1 with AAC at 128Kbps, and a second sample that's been recompressed, again using AAC at 128Kbps, starting with the first sample as its source.
Let's see how many of you have the golden ears needed to tell one pass with AAC from two.
My interest in doubled-up compression has been tweaked recently because of a few occasions where I wanted to edit music that I've purchased from iTMS. I've needed to expand the compressed music, and then, not wanting to keep huge AIFF files around, compress the music again when saving the edited results. I feel a little iffy about doing this, but so far I don't think I can detect any obvious difference in sound quality. Of course, I've been recompressing at 192K after editing rather than 128K, so I've been more cautious than what I'm doing in the above listening test.
Extra generations of compression definitely aren't a good thing, but just how bad are they? I've put together a little listening test to explore this issue. I've got a sample that's been compressed once via iTunes 4.0.1 with AAC at 128Kbps, and a second sample that's been recompressed, again using AAC at 128Kbps, starting with the first sample as its source.
Let's see how many of you have the golden ears needed to tell one pass with AAC from two.
My interest in doubled-up compression has been tweaked recently because of a few occasions where I wanted to edit music that I've purchased from iTMS. I've needed to expand the compressed music, and then, not wanting to keep huge AIFF files around, compress the music again when saving the edited results. I feel a little iffy about doing this, but so far I don't think I can detect any obvious difference in sound quality. Of course, I've been recompressing at 192K after editing rather than 128K, so I've been more cautious than what I'm doing in the above listening test.
Comments
I'm using camino, will it not load them?
And, can you really expand .mp3s and stuff? does it return it to the original Aiff state? I thought data was lost in the compression...it sounds like it would work(more or less) but also sounds a little wonky to me.
http://www.shetline.com/music/listening_test.html
It should load in Camino.
Originally posted by Wrong Robust
And, can you really expand .mp3s and stuff? does it return it to the original Aiff state? I thought data was lost in the compression...it sounds like it would work(more or less) but also sounds a little wonky to me.
No you can't expand an MP3 to the original AIFF state. There's a reason why you get around 11:1 (at 128kbs) compression with MP3's - because data is lost. Of course lossless audio compression techniques do exist, such as FLAC. You don't get as good compression (about 2:1) but all original data is preserved.
Wesley
Originally posted by Wrong Robust
And, can you really expand .mp3s and stuff? does it return it to the original Aiff state? I thought data was lost in the compression...it sounds like it would work(more or less) but also sounds a little wonky to me.
You can expand MP3 and AAC files into AIFF files, but the results aren't identical to the original source.
Every time you play an MP3 or AAC file, it's being expanded on-the-fly into an AIFF-like form (not necessarily an AIFF file, which is just one particular file format for linear PCM audio) as part of the playback process.
Are you going to post the correct answer?
I've noticed that the sound quality of the AAC files on the iTMS largely depends on the source recording. Tracks from a lot of older albums on the iTMS sound muddier to my ears.
Originally posted by mlnjr
Are you going to post the correct answer?
Yes, I intend to post the correct answer when the poll closes in two weeks. If activity in the poll is low, I might post the answer earlier.
It would be interesting to have people who can hear clear differences describe as best they can what they're hearing. As long as no one gives away what they think the correct answer is, I have no objection to the poll being influenced by some ear-training advice.
but it was really fun playing them all at the same time, and making little musical rounds. thanks.
Originally posted by xionja
didn't understand the whole quiz thing,
but it was really fun playing them all at the same time, and making little musical rounds. thanks.
I tried to sync them all by myself.
Maybe I could use shuffle play to test myself -- write down my guesses with the iTunes display hidden, then step back through what's been played after four samples have been played to check myself.
What REALLY scared me was how bad quality the "only one" compression was. High tones from drums (high hat or whatever) and Michael Stipes voice was CLEARLY distorted. I have had problems with the sound on my iBook and thought it came from that. But when I listened to the AIFF file it went through clear.
Gotta try some different settings for my future mp3 production.
Great idea BTW.
Originally posted by 709
Cool test. I downloaded the files after voting to see if I was right, and was suprised that the difference in files size between the two wasn't very much at all. Interesting.
The difference in size should be very small, because the size difference is due only to a small amount of padding that's added to the beginning and end of the sound file by each stage of the compression process. So, if you kept decoding and re-encoding the same file at the same bit rate, not only would the sound quality drop, but the length of the file would very slowly increase. (Perhaps you were thinking twice compressed would mean smaller? Not in this case, because compressing twice always means expanding once in between.)
Since the files remain nearly the same duration within a few milliseconds, the size of the files is determined almost completely by duration * bit rate.
Originally posted by Anders
I could detect a clear difference (iBook and Sennheiser HD200). And after voting I see most people agree.
Playing over my SoundSticks, I was having a hard time clearly making out what had changed in the sound quality after a second pass of compression.
After putting on a cheap pair of headphones that happened to be handy, however, the differences became much clearer.
Two things in particular that I noticed that I can put into words were an increased "swirling" quality to the background noise, and an unsteady fluttering character with some of the drum notes, notes that were tighter and steadier after only one generation of compression, but that don't apparently survive two passes of compression very well.
I'll still try my own random test for myself, but I think I won't have much trouble succeeding now that I've heard what to listen for.
I went back and listened to some other music that I'd purchased via iTunes at 128K, expanded to AIFF so that I could do some editing (trimming and adding fade-in/outs in a couple of cases, joining two tracks in another case), and then recompressed at 192K. The edited music still sound good to me, even with my ear better tuned to the kinds of distortion to expect.
So, I think I can say for myself:
128K AAC -> AIFF -> 128K AAC: Noticeable loss in quality.
128K AAC -> AIFF -> 192K AAC: Sounds good, or at least I'll have to work harder to learn to recognize any resultant distortion.
Yes, it's 2-1-1-2.
Originally posted by Harald
Now let's do MP3 v AAC at one-pass 128k to separate the men and women from the childers ...
I haven't tried this with "normal" music yet, but even at 192K it was very easy for me to hear the difference between an MP3 encoding of "fatboy.wav" (Google and ye shall find) and an AAC encoding. MP3 via iTunes simply fell apart trying to encode the weird buzzing sounds of this particular sound sample. AAC did much, much better. It's possible that a better MP3 encoder, such as LAME, might improve on iTunes' MP3 performance.
I'm more motivated to try another round of double-AAC, but this time with the second round of AAC being at 192K instead of 128K -- this matches what I'm actually doing for the few times I've wanted to edit music that I've purchased from iTunes.