I was curious to recreate the calculations on this page
http://www.underbit.com/resources/mpeg/audio/compliance/
So I set up some tools to test and was able to repeat the test.

I have updated stats for Winamp, and calculated stats for Foobar2000. Dither was disabled. I only tested Layer 3.


Winamp @16bit (These stats are identical to the Underbit page)
RMS: 8.602x10^-6 (-101.31dB)
Peak: 1.526x10^-5 (-96.33dB)

Winamp @24bit
RMS: 8.278x10^-8 (-141.64dB)
Peak: 2.384x10^-7 (-132.45dB)

Foobar @16bit (skipped first 529 samples as I was unable to disable decoder delay compensation)
RMS: 8.611 x 10^-6 (-101.30dB)
Peak: 1.538 x 10^-5 (-96.26dB)

Foobar @24bit (skipped first 529 samples as I was unable to disable decoder delay compensation)
RMS: 4.156x10^-8 (-147.63dB)
Peak: 2.384x10^-7 (-132.45dB)

Thanks for the science, benski.

Re-ran the tests using the in_mad Winamp plugin. Disabled all options. I think 24->16bit dither might be hardcoded, which would explain the poor results for 16bit.

MAD @16bit
RMS: 1.932 x 10^-5 (-94.28dB) - FAILS Maximum allowable difference -101.1dB
Peak: 5.805 x 10^-5 (-84.72dB)

MAD @24bit
RMS: 9.389 x 10^-8 (-140.55dB)
Peak: 4.768 x 10^-7 (-126.43dB)

benski, could you post the results for the latest Winamp and the latest foobar2000?

Thanks!