The self dither argument is not as obvious as it may appear. To be effective at dithering, the noise has to be at the right level of course but also should be white and temporally constant. The noise floors present in music data normally come from the self noise of the analog components used in recording and are composites of a number of noise PDFs. For example, a graph in a second paper by the same group (cited below if wanted) shows spectra of the measured noise floors from around a dozen recordings. The noise spectra are composites with the lower frequencies clearly 1/f noise and the upper frequencies summing closer to flat. Whether composite noise of this sort is both temporally continuous and white enough to be relied on for dither needs to be shown; it's been shown under at least some circumstances (not in these papers) that a truncation distortion spectrum can be produced and measured when signals are truncated to 24b.
I'm not saying the self dither argument is necessarily wrong; but it needs verification as to when and where it is reliably valid. If 24b truncation turns out to be demonstrably audible in an AB/X, then the self dither idea clearly needs to be rethought. Vicki Melchior (graph mentioned is fig 8 in this paper: http://www.aes.org/e-lib/browse.cfm?elib=17501) On Feb 6, 2015, at 2:20 PM, Nigel Redmon wrote: > First, if there is enough noise in the signal before truncation, then it’s > dithered by default—no correlation. -- dupswapdrop -- the music-dsp mailing list and website: subscription info, FAQ, source code archive, list archive, book reviews, dsp links http://music.columbia.edu/cmc/music-dsp http://music.columbia.edu/mailman/listinfo/music-dsp