128kbs or 320 kbs …

… only your mp3 encoder knows for sure.

Take the test at mp3ornot.com to see if you can tell the difference between an MP3 encoded at 128KBS and one encoded at 320kbs.  I couldn’t tell the difference (but I was listening via my  laptop speakers).  I hope the author will post statistics on how many people could tell.

  1. #1 by tam on March 12, 2009 - 3:44 am

    pretty hard (laptop speakers as well). Maybe a more exhaustive test could actually rule out any luck.

  2. #2 by wajid on March 12, 2009 - 9:43 am

    aye i spotted the difference – even on laptop speakers….ears like spock me

  3. #3 by Ben on March 12, 2009 - 10:14 am

    Granted it’s machine not human listening, but Kurt looked into the affect various lossy codec rates have on some core MIR tasks in a couple of AES convention papers you may be interested in:

    – Genre Classification: http://www.aes.org/e-lib/browse.cfm?elib=14547
    – Onset Detection: http://www.aes.org/e-lib/browse.cfm?elib=14693

    I’m the second author on the genre classification paper (it’s my classifier) and I have a copy of it sitting around if you don’t have access to someone with an AES subscription. Let me know if you want to give it a read and I’ll send it over your way.

    • #4 by plamere on March 12, 2009 - 10:20 am


      I’d love to read the paper. If you could send it to Paul@echonest.com I’d appreciate it.

  4. #5 by Ben on March 12, 2009 - 10:55 am

    done and done.

  5. #6 by Ben on March 12, 2009 - 11:07 am

    My understanding of the mp3 codec was that it was aiming to be mostly invisible for 90% of the population at 128k. Also, for what it’s worth, the stimulus on that webpage is a great example of things the mp3 codec is good at dealing with. Minimal polyphony, mostly voice, few transients. If the track was something like say a rock band (or even a solo kit), it would be vastly easier to differentiate. Really an a-b test like this needs to have a few songs to tell you anything. And for that matter ab-x is a much more rigorous listening task. (For the uninitiated, in ab-x three versions are played, a, b and x. Either A or B is the exact same as X, the goal being to select this version. The closer to 50% accuracy a group of listeners gets, the better your codec.)

  6. #7 by Ju Polimeno on March 13, 2009 - 7:15 am

    In my studio monitor was easy to find the 320kbps. Cheers!

  7. #8 by aghiff on August 17, 2009 - 6:32 am

    320 kbps sound awkward…require higher level of volume…from my experience…

  8. #9 by aghi on April 12, 2010 - 12:44 am

    i gussed it as well!!!! Clip A for me was 320kbps and i guessed it on my MBP laptop speakers.

  1. 128 or 320 kbps - can you hear the difference? « zed equals zee
%d bloggers like this: