>> If you can’t hear it, or reproduce it—and at 24-bit you get both—you can’t >> hear it. > > Amplify the noise to 60dB SPL and let me me introduce my favourite test > signal. Absent subtractive dither, I can introduce audible intermodulation > products. Even if not amplified, intermodulation products tend to creep up > off the noise floor, if stable enough.
Well…in a few of these items you seem to be accepting my points about 24-bit…then say things like bring that noise floor up to 60 dB SPL, a measure to circumvent the 24-bit playing field, and shortcoming of electronics and human hearing. I’m not sure if you’re pulling my leg :-D, but this no longer has much to do with delivering music. Bring the noise floor to up 60 dB SPL—and I think we’re still talking about the noise floor being the truncation (dithered or not) noise floor for 24-bit—and where is the music? At a level few sound systems can sustain, and where you ears will be irreparably damaged in short order. Now, you do follow this up with saying you can do it with out amplification, to some degree, but clearly you wouldn’t have mentioned listening to the noise floor boosted to 60 dB SPL if it was remotely close to hearable without amplification. Anyway, I can agree with you on the technical aspects of correctness, but the practical aspects dominate. I think it’s clear that additive dither for 24-bit truncation is already fixing a problem that can’t be heard, but it’s next to free—either way you’re transporting 24 bits and can’t hear the last few, no big deal that we spent a few cpu cycles to “make it right". But I have trouble getting excited about taking an unneeded remedy to the next level. Anyway, carry on, still fun to talk about. > On Jan 8, 2022, at 12:22 AM, Sampo Syreeni <de...@iki.fi> wrote: > > On 2022-01-07, Nigel Redmon wrote: > >> I'm at a loss of how to interpret the importance of "statistically be >> inefficient for measuring purposes”. I don’t think it has much to do with >> 24-bit streaming services, which is what I was addressing. > > Of course it wouldn't do much there. > > What I'm talking about is a wholesale solution to all dither needs. Including > audio, especially of low bitwidths, but then also transmission radiotelescope > signals and whatever. > > Because subtractive dither really is the wholesale solution. So good it > mimics analogue tape at 8 bits, while making 1-bit signals from beyond > Jupiter better. (Modulo.) > >> No evidence it’s necessary, even at 16-bit, but certainly not at 24-bit >> which was specifically what I was addressing. > > Then I missed your point or overspoke. Obviously if you have 24-bit accuracy, > this stuff mostly won't be necessary. > > Yet in theory it will be. In theory, even if you have a 24-bit quantizer, > with ample noise upto the 22-34th bit (the practical limit of today's > converters), if you look hard enough, you can see the quantization > underneath. (People do this, you know, in the cryptographic circuit. The guys > who actually worry about statistical bias. Also, scientific folks, who worry > about their stochastic sampling code being spot-on, to the umpteenth moment.) > >> If you can’t hear it, or reproduce it—and at 24-bit you get both—you can’t >> hear it. > > Amplify the noise to 60dB SPL and let me me introduce my favourite test > signal. Absent subtractive dither, I can introduce audible intermodulation > products. Even if not amplified, intermodulation products tend to creep up > off the noise floor, if stable enough. > >> Ecept I have a pretty good idea of how well it would work, so not much point >> in the experiment—too many other things on the list. :-D > > Thus, each of us will probably just carry on. It's not as though I'd be able > to even commecialise thise stuff. :) > -- > Sampo Syreeni, aka decoy - de...@iki.fi, http://decoy.iki.fi/front > +358-40-3751464, 025E D175 ABE5 027C 9494 EEB0 E090 8BA9 0509 85C2
