It is different enough to significantly outweigh the small differences between 16 bit vs. 24 bit, and 44.1 kHz vs. 96 kHz.
Normalizing the peak amplitude does not guarantee equal loudness, especially with the above mentioned randomization that also adds randomness to the peak level. You would need to use something like ReplayGain for more accurate matching.
However, the best approach is to generate only one file with the software synthesizer, at the highest possible quality (i.e. 192 kHz/32-bit), leaving a couple dB of headroom so that the sample rate conversions will not result in clipping, and converting that file to all the other (lower quality) formats to be tested. Finally, to make sure that the DAC does not introduce any differences, convert the files back to 192 kHz/32-bit, and compare those with the foobar2000 ABX comparator. If the software you use for the conversions is well written, you should not need to apply any level matching or synchronization.
Ok, I went ahead and generated a new file at the optimized settings and converted it to 16/44.1 and back again. I also exported multiple files from the same Live project as I did before. I also ABX'ed. I have determined the following:
-I cannot tell the difference between any of the derivate files of the original 192/32 master file. You were correct.
-When using midi, none of the audio has been captured prior to exportation. The initial settings for exporting the master file greatly affect the audio quality. Exporting at 192khz vs 96khz and so on produces a marked difference in quality, which likewise applies to the bit rate. These different export settings can ABX'ed at 100% accuracy. This has been the source of my confusion. However, interconverting the files post-export does not seem to result in any noticeable difference.
Edited by ctoth666 - 9/24/12 at 11:46am