For those of you who are really interested, the forum below hosts many very good discussions on audio compression and the various formats. I learned a great deal from there when I started digitizing my CD collection.

http://www.hydrogenaudio.org/forums/index.php?act=idx

Regards,
Dan

bitflipper


Have you guys tried this MP3 bit-rate blind ABX? It's an eye-opener.

I was feeling pretty cocky after getting 5 out of 5 correct, even if that wasn't statistically significant. By the time I'd done 10 tests, my percentage was down to 83%, and after 15 tests it was 71%. It depended heavily on the style of music being A/B'd. I could nail the Jazz samples every time, but with distorted guitars and wailing vocals it was a total guess.

This is great! And I beat you :P After 15 tests I got 73% haha! Just like you, the Jazz and the Latin style acoustic percussion thing, I could pick it every time. The Rock one was the only one I made any mistakes on. For me, it was the acoustic guitar on the first that gave it away. The brightness and detail was lost in the 128kbps. It was the ride cymbal on the jazz that gave it away - same reasons as the acoustic. But without any detailed high end instruments in the 3rd distorted guitar rock piece, it was much more difficult to assess. It's good to challenge yourself too and just listen to it outright without listening to the test tracks.


Pretty impressive encoding though. I'm amazed that's 128. Sounds more like what I'm used to from 192. It has improved a lot over the years. I'd like to know more about the encoding they used.

 
 These compressed file formats were a necessary evil and have made a lot of money for Amazon , Apple and others......billions of dollars. Anyone who has seriously wanted to have a big web exposure as a musician has probably made the compromise.

 Back in the days of tape some liabilities were tape hiss and those annoying cracks and pops from pressed records and the potential for scratches. The CD came along and the audio improved, then as I see it we needed a band aid approach to make things work in the age of computer music. They did the best they could with what they had to work with. The continued improvement of these formats and the more powerful computer will hopefully soon close the gap between what we had then and what we need now. Some may say the gap has already been closed.

  IMO the companies who make the big money online need to take more drastic measures to adopt the better formats. Mp3 seems to have become the de facto term for an online compressed file. Most people refer to their portable audio devices as Mp3 players. I hope we can get away from that. I think there is concern in some circles that an artist making a file for mass online distribution needs to insure cross compatibility by continuing to use Mp3.

 Converting to something else may loose customers who still play Mp3 exclusively on their portables even if it sounds better.

The correct but widely unreported answer is, "the frequencies between 4220 and 4225 suffer the most"

P.S. It's widely unreported for a reason...

bitflipper
Here's how you can get an idea of what the MP3 encoding is removing. 

Put a single hi-hat hit at the front of your song in the count-in area. This will be used later for alignment. Export the project as a normal wave, then create an MP3 from that wave. Now open a new project and import both the wave file and the MP3. Using the hi-hat hit, visually line the two files up. Match their levels to within a fraction of a db (this will take some trial and error). Now invert the polarity of either of them and bounce the two to a third track. If you've matched levels well, this third track will contain pretty much what the MP3 encoder has removed. 

I should point out, though, that this is purely an intellectual exercise and has no practical value. The whole concept behind perceptual encoding is that what's being removed is stuff you can't hear anyway. So just because you can hear it in isolation doesn't mean it was audible in the original mix.

Dave,
 
I've stated this on the forums many many times.
 
The (partial) fundamental flaw with your outline above is that you forgot to remove the ~72 samples that mp3's add in front of the file (I've proved it over and over, and even Beagle admits they are there). That alone will (also) skew the so called null test.
 
I ask you: would you purposely put ~72 samples in front of one of two .WAV files you suspected were not the same and null test them? No? Then why would you line up a .WAV and .mp3 without removing the dead air mp3's add at the beginning of the file?
 
And oh, by the way.... this is not a purely intellectual exercise.