In Lincoln's paper, he quotes the energy of a the +/-1 (in the least
significant bit) wave as -93db. In LAME, this is
-86db (short block) and -92db (long), so that's pretty good
agreement. That's the peak energy in a sfb, the average energy
is about 10db lower (when the input is a single sine wave).
Lincoln uses a ATH = ATH_formula - 114, so I'll go ahead and make
that the default. It is close enough to the value of 110
suggested by macik's tests.
Mark
--
MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
