mp3 vs aiff
Results 1 to 9 of 9

Thread: mp3 vs aiff

  1. #1

    Default mp3 vs aiff

    what is the difference between mp3 and aiff?
    quality?levels?ect...

  2. #2
    DJTT Moderator Dude Jester's Avatar
    Join Date
    Feb 2009
    Location
    New Zealand
    Posts
    12,370

    Default

    aiff is similar to wav as in it is a lossless format, whereas mp3 is a compressed (lossy) format. check this out

    http://www.differencebetween.net/tec...-aiff-and-mp3/
    Acer E5 i7 2TB 16GB ~ 512GB SSD ~ WIN 10 ~ TSP 2.11 ~ AUDIO 6 ~ X1 ~ DN-X1600 ~ X1 ~ SPECTRA ~ TWISTER ~ ATH-PRO500 MK2 ~ AT2020 MIC

    "This aggression will not stand, man."

  3. #3
    Tech Mentor
    Join Date
    Oct 2011
    Posts
    277

    Default

    aiff is an uncompressed file format which stores audio data and associated metadata. it was developed by apple. (for those more technically inclined, i believe the central difference to WAV--besides the file header/the way metadata is stored--is endianness.)

    mp3 is an audio format that uses lossy compression. at the bitrates which you get in online stores nowadays, it can be considered fully transparent (i.e., perceptually indistinguishable from the uncompressed original).

    choosing a lossless format such as AIFF, WAV, FLAC, or ALAC for archival purposes can still make sense. It allows you to transcode to any lossy format you desire without generation loss.
    Last edited by rgtb; 02-06-2012 at 06:30 AM. Reason: typo

  4. #4
    Tech Guru lethal_pizzle's Avatar
    Join Date
    Feb 2009
    Location
    Maida Vegas, London
    Posts
    2,816

    Default

    Quote Originally Posted by rgtb View Post
    perceptually indistinguishable from the uncompressed original
    Not everyone would agree with that statement (although I do - years of walkman abuse has seen to that)
    DJTT Nu Disco Mix Train Vol 1
    beats and balearic bobs in north-west london
    iTunes podcast
    soundcloud

  5. #5
    Tech Mentor
    Join Date
    Oct 2011
    Posts
    277

    Default

    Quote Originally Posted by lethal_pizzle View Post
    Not everyone would agree with that statement (although I do - years of walkman abuse has seen to that)
    i know. those are usually people who do sighted tests.

    however, show me the scientifically sound empirical test where MP3 (V0 VBR or 320 CBR) can be discerned from the original. imo, sighted tests (which don't properly control for biases) are completely useless garbage.

    EDIT: this is not to step on anybody's toes. there might be exceptions. people with extremely good hearing and extremely good equipment might ABX a track with statistical significance now and again. particularly if you picked audio material that is known to pose difficulties to lossy compression algorithms. but most of it is audiophile voodoo. people claim to hear a difference but cannot hear the difference once we do an ABX.

    2nd EDIT: lastly, scientific tests nowadays don't do the testing on high bitrate stuff. they test bitrates below 160kbps or so because they know with high confidence (in the statistical sense) that high-bitrate encoded stuff is transparent.
    Last edited by rgtb; 02-06-2012 at 06:46 AM.

  6. #6
    Tech Guru mostapha's Avatar
    Join Date
    May 2010
    Location
    Atlanta
    Posts
    4,748

    Default

    I've achieved .05 significance double blind ABXing 320kbps mp3s against wav files on my HD25s and on KRK RP5s in an untreated room.

    I can't hear the difference on earbuds or in a club. I could on my decent Etymotic IEMs. And I'm convinced I'll be able to when I buy shures.

    I can also hear above 19kHz in blind listening tests at reasonable volumes. The last time I went to a club, concert, or venue without earplugs was in 2004. It was also the second time.

    Because I'm bored and procrastinating, I just did a semi-blind non-ABX "which is better" version of the test. Picked 5 tracks out of my "newish" playlist and created a playlist for each. For each, I encoded a 320kbps version and hid all metadata from iTunes view…repeat + shuffle.

    5 out of 5.

    All of them from the first 5 or so seconds of a modern dance track purchased at beatport or traxsource. I didn't really have a question in my mind which is which, but it usually took a total of ~5 listens through the first few seconds of the tracks. Actually doing the ABX double-blind would have made it easier because I could just listen for something specific that either isn't in the mp3 or sounds worse in it (certain types of transients and reverb tails are the easiest).
    Last edited by mostapha; 02-06-2012 at 10:16 AM.

  7. #7
    Tech Mentor
    Join Date
    Oct 2011
    Posts
    277

    Default

    so, i hope it's all right with the OP that this thread goes a bit OT--we've answered his questions after all.

    three quick suggestions for others who are new to this but are interested in doing some testing of their own:

    1.) a good software to do ABX tests is foobar2000. you need to add the ABX component to it.

    2.) you might wanna post logs of your tests. claims about hearing are similar to claims about one's ability at driving: like 90-95% of people think of themselves are above-average drivers which, ofc, is impossible.

    3.) it's advisable to do about 15 trials per test. for example, if you do 16 and get just 3 wrong, you can be fairly certain that you can differentiate between A and B. (for those more technically inclined, 3 errors in 16 trials means you can reject the null hypothesis of no perceptible difference between A and B at the 1% level.)

    (edit: just to clarify: 5/5 is significant as well, as mostapha posted. ofc, the number of trials should be set prior to starting the test.)
    Last edited by rgtb; 02-06-2012 at 11:30 AM.

  8. #8
    Tech Guru mostapha's Avatar
    Join Date
    May 2010
    Location
    Atlanta
    Posts
    4,748

    Default

    Right. The test wasn't purely scientific. I just had to get up with my gf at 6 and was bored.

    Also, 90% of drivers could be above average if some portion of the othe 10% REALLY suck. But proving whether or not driving performance follows a normal distribution is beyond the scope of a DJ forum and not relate to audio ABX tests.

    Spoiler: with driving, girls follow a normal distribution, guys probably follow a bimodal distribution that is probably still symmetrical, depending on the measure used. Part of the confusion in tests like these is that the relationship of girls->normal, guys->bimodal hold for a LOT of measures (general population ends up vaguely normal) and the reasons may be cultural rather than biological because it also only seems to hold in historically patriarchal societies without extreme competition for survival.

    Also, major props for rgtb for knowing something about statistics. It's something that's important an also lost on a lot of people.

  9. #9
    Tech Mentor
    Join Date
    Oct 2011
    Posts
    277

    Default

    Quote Originally Posted by mostapha View Post
    Also, 90% of drivers could be above average if some portion of the othe 10% REALLY suck.
    i mean, the problem is that "driving aptitude" or whatever you wanna call it is not something that you can really measure on an interval scale. so, when i used the term "average," i was speaking colloquially. you'd first have to operationalize driving aptitude before you could actually make the statement i made. but i digress. and, certainly, i see the tongue-in-cheek of your last.

    but, to get back to the listening test thing. i really think it would be an interesting thing to do here in the DJTT forums. you know, there are many public tests of lossy compression algorithms. many let the subjects download the files onto their own computers. and probably many of the subjects have really bad audio equipment at home. sure, the RP5 and headphones you have are not top-of-the-line. they're certainly better, though, than what the average consumer has. therefore, i'd really think it would be cool to do some ABX testing with the DJTT crowd.

    idk, maybe take some netlabel track that's creative-commons licensed or some track produced by a DJTT forum member. anything where we don't run into copyright issues. and then somebody takes the WAV and calls it A, compresses it as 320kbps MP3 with LAME or whatever and calls the result B. then randomly produces five files X1, X2, ..., X5 which are randomly either A or B. uploads A, B, and X1-X5 as a zip to a one-click hoster. sends the answer key of what are X1-X5 to a mod. then posts the link to the zip on the forum. and everybody can put in their opinion as to what X1-X5 are. and, some time later, the mod posts the answer key and we evaluate to what extent people were able to identify A, the original WAV.

    ofc, it still wouldn't be rigorous as people would be able to see what answers were already given. but i'd still enjoy doing it...

    ---

    EDIT: obviously it would be easy to see what is MP3 based on the file size. Currently thinking about this. Maybe one could encode as MP3 and decode right away, thus distributing B in WAV format as well.

    But even then, people could cheat by looking at the power spectrum of A and B. I gotta find a better way. Perhaps use one of the testing sites out there...
    Last edited by rgtb; 02-06-2012 at 11:27 PM.

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •