Quick question: Will this work?

[wordmunger]

I'm doing a quick test to see if people can tell the difference between various MP3 sampling rates (I know, it's been done a million times, but I want to see if musical training has any impact on the results).

So, I imported the song into iTunes at 256 kbps. Then I put it into Garageband to edit the clip down to about 30 seconds in length. Then I exported to iTunes again, and used the "convert to MP3" function to make a 256 kbps, 128 kbps, and 64 kbps version.

So, for all you audiophiles out there, is this a good way to create the sample files, or might something be lost in all those file conversions?

 

[AviationFan]

In general, multiple stages of compression (file size, not dynamic range) should be avoided unless it is lossless (which MP3 is not). For best results, compression can be fine at the very end of the process (such as AC-3 or DTS encoding for DVDs), where you create the final media for distribution, but not at the beginning or in between.

Therefore, for your experiment I recommend you start with an uncompressed audio file and generate from it the various compressed versions at different bitrates, using just one single encoding for each file. Otherwise, your files may not accurately reflect the true potential of the audio encoding that was used.

- Martin

 

[wordmunger]

Yeah, I thought so. I redid them starting with Apple Lossless. But does using Garageband introduce any other problems?

 

[Avatar74]

I'm doing a quick test to see if people can tell the difference between various MP3 sampling rates (I know, it's been done a million times, but I want to see if musical training has any impact on the results).

So, I imported the song into iTunes at 256 kbps. Then I put it into Garageband to edit the clip down to about 30 seconds in length. Then I exported to iTunes again, and used the "convert to MP3" function to make a 256 kbps, 128 kbps, and 64 kbps version.

So, for all you audiophiles out there, is this a good way to create the sample files, or might something be lost in all those file conversions?

Generally you want to do a bounce from an uncompressed source.... and preferably a high resolution master (e.g. 24-bit). You also want to use a good encoder that is capable of a high resolution bounce... that is, it's not a slapdash encoding but an actual realtime resampling taking place... I've found ProTools LE and its MP3 plug in to be good for this.

With MP3 I think you're going to find that even the average listener can discern artifacts at 128Kbps and below. But if you want to do a real test, the conditions have to be double-blinded... That is, neither the person administering the test nor the subject taking the test can know which sample is being tested at which time.

A control factor you might want to put into place is to test the results when you mislabel the samples, to see what kind of psychological effect there is when people are under the false impression that the sample they're listening to is a higher bitrate than it is.

Musical training is not necessarily going to produce someone of better listening ability. Sound engineering training perhaps... but so-called "audiophiles" are the last people you want to ask. These individuals are biased towards more cumbersome solutions because of confirmation bias... they rationalize purchases of expensive equipment so as to be blinded by their own bias toward status symbols when it comes to correct evaluation of relative quality between samples. Not one ABX test in a controlled, double-blind condition has been able to demonstrate that anyone could consistently discern (75% or better accuracy) between MP3, MPEG-4 AAC and the 16-bit Linear PCM uncompressed source.

Again, there is a difference between asking the participant which sounds "better" versus asking them to correctly identify which sample is in what format. The latter is the better litmus test.

In the case of MPEG-4 AAC (the default format of iTunes), 128Kbps AAC is perceptibly indistinguishable from a 16-bit Linear PCM source. These findings are consistent with that of the Audio Engineering Society, the foremost organizatioin of experts in the field of audio engineering, mastering, compression and encoding.

 

[wordmunger]

A control factor you might want to put into place is to test the results when you mislabel the samples, to see what kind of psychological effect there is when people are under the false impression that the sample they're listening to is a higher bitrate than it is.

That's a neat idea -- this is really just a quick test, though, so I didn't want to burden the study with too many different examples.

Again, there is a difference between asking the participant which sounds "better" versus asking them to correctly identify which sample is in what format. The latter is the better litmus test.

I'm not sure I buy that. Isn't the point to discern what sounds better, i.e. which format is the best?

In the case of MPEG-4 AAC (the default format of iTunes), 128Kbps AAC is perceptibly indistinguishable from a 16-bit Linear PCM source. These findings are consistent with that of the Audio Engineering Society, the foremost organizatioin of experts in the field of audio engineering, mastering, compression and encoding.

Cool -- do you have a link or a reference?

Anyway, my little informal study is up. It's not ideal, but I think it's a lot better than most of the online studies you see out there. We usually get around 500 or so responses to these things, so I'll try to remember to update this thread when we get our results. If you want to participate, here's the link:

http://scienceblogs.com/cognitivedaily/2007/11/casual_fridays_are_all_mp3s_eq.php