From both a theoretic and perceived standpoint how does the Foobar2000 mp3 decoder compare with the Fraunofer l3dec 2.72 decoder and MAD 24 bit? I have heard from old timers that the dos Fraunofer l3dec 2.72 decoded to 24 bit hex is the most accurate decoder. Visually, at 24 bit, their waveforms all appear different to 16 bit decode if viewed in an audio editor such as Wavelab.

Foobar2000's decoder is based on mpglib and outputs in 32 bit float format. I think Sound Forge's MP3 input uses Fraunhofer's decoder, but I think that also decodes to 16 bit integer samples.

can you tell the difference between any of those two? i mean yea there's a difference mathematically and what not, but i doubt most people here the difference, it's all placebo. imo.

I agree the difference would be very very small but I am still interested anyway, especially if writing to WAV. I make animations and many hard to get sound effects are only available in Mp3. With multiple mixing and DSP in mind I would like to know the best quality decoder to 24 bit wav.

For writing to wav at 16 bit mode, I'd say that MAD 16 bit dithered output and Foobar2k 16 bit dithered output would be equivalent, unless anyone shows otherwise.

About a 24 bit decode... I don't know, there's no info about this that I know, but I think that differences are not significant.

Which dithers better, MAD or Foobar2000?

i have a hard time telling apart dither and undithered at 16bit... 8 bit is easier.

You could test this yourself in a similar to my mpc/ogg/mp3 low volume test:

To create your mp3 test file either take an mp3 and apply gain to it with mp3gain (e.g. -30dB) or take a wav file, change its resolution to 24/32bit, amplify it (e.g. -30dB) and compress it. (I would recommend the 1st possibility.) Then decode them (You might want to test XMPlay which has its own mp3 decoder and up to 32bit diskwritng output) to 24/32bit wav files and amplify the results reversely (e.g. +30dB). Now listen and try to hear differnces. Different amplification levels (or fadein/out instead of constant amplification) will be useful.

Beware: The results are exaggerated, but I think at least what sounds best in the test won't give you worse results in reality than anything else.

-edit: added link