skippy
New member
Too busy working for a living, settling in after the move, and slamming the studio back together in a major hurry to mix and master a location-recorded a capella CD that has to be ready for sale by the first of December. And with two more in the chute- and _zero_ time to do the acoustic treatment of the room in the new place.... Ain't we got fun?
Anyway, the guy who is the real dither ace is Sjoko2, not me. My dither knowledge is pretty dated. But the noise shaping that you can do by using dither properly really can go a long way towards moving the quantization noise out of the "hey, isn't that kinda grainy?" realm and into the "sounds more analog" realm. Reverb tails are the place where you'll hear it: try taking something at 24 bit, and then listen to it straight up, truncated to 16, dithered to 16, then dithered _twice_.
Bottom line for me is to dither only once, when mastering to 16-bit release media, and to keep everything else in the 24-bit depth for work-in-progress.
I did some fun research into the psychoacoustics of noise back in my misspent youth, about 1981- one of my first products was a super-low-cost wavetable subtractive synth chip that was intended to be included on game carts for the Atari 7800 (my first chip design was the graphics processor for that box, in fact). One thing you really need is noise, either white or pink, which you can then filter with a formant generator and do other useful things with. In the analog domain, getting usable noise is relatively easy- junction noise from a diode, suitably amplified, is a good source. But it's a cast-iron bitch to get good noise in the *digital* domain, and the semiconductor process we had to work with really wouldn't support any of the normal analog noise sources (it was hard enough to get the sound off the chip after the D/A: truly, game-quality only).
Digitally-sourced noise (pseudorandom number sequences) almost always sounds pitched. Local correlation in the bit stream tricks your brain into thinking that there's signal down in that noise, like your primordial mama calliing you back to the cave in a blizzard. We spent months running simulations trying to come up with convincing digital noise- the best we did was to mung together the results of 4 separate pseudorandom sequences or different lengths, with a really gnarly chunk of logic. Both my co-developer and I could still hear a little pitched whistle way down in there, though, even with our best effort (given schedule constraints). The, we realized that we were each hearing _a different pitch_- and we knew we were at the point of diminishing returns.
Anyway, the quality of the noise is absolutely key to the results of the dither operation. And at that, I'll let Sjoko2 take over with the current state of the art, and go back to just being a nerd.
Anyway, the guy who is the real dither ace is Sjoko2, not me. My dither knowledge is pretty dated. But the noise shaping that you can do by using dither properly really can go a long way towards moving the quantization noise out of the "hey, isn't that kinda grainy?" realm and into the "sounds more analog" realm. Reverb tails are the place where you'll hear it: try taking something at 24 bit, and then listen to it straight up, truncated to 16, dithered to 16, then dithered _twice_.
Bottom line for me is to dither only once, when mastering to 16-bit release media, and to keep everything else in the 24-bit depth for work-in-progress.
I did some fun research into the psychoacoustics of noise back in my misspent youth, about 1981- one of my first products was a super-low-cost wavetable subtractive synth chip that was intended to be included on game carts for the Atari 7800 (my first chip design was the graphics processor for that box, in fact). One thing you really need is noise, either white or pink, which you can then filter with a formant generator and do other useful things with. In the analog domain, getting usable noise is relatively easy- junction noise from a diode, suitably amplified, is a good source. But it's a cast-iron bitch to get good noise in the *digital* domain, and the semiconductor process we had to work with really wouldn't support any of the normal analog noise sources (it was hard enough to get the sound off the chip after the D/A: truly, game-quality only).
Digitally-sourced noise (pseudorandom number sequences) almost always sounds pitched. Local correlation in the bit stream tricks your brain into thinking that there's signal down in that noise, like your primordial mama calliing you back to the cave in a blizzard. We spent months running simulations trying to come up with convincing digital noise- the best we did was to mung together the results of 4 separate pseudorandom sequences or different lengths, with a really gnarly chunk of logic. Both my co-developer and I could still hear a little pitched whistle way down in there, though, even with our best effort (given schedule constraints). The, we realized that we were each hearing _a different pitch_- and we knew we were at the point of diminishing returns.
Anyway, the quality of the noise is absolutely key to the results of the dither operation. And at that, I'll let Sjoko2 take over with the current state of the art, and go back to just being a nerd.
This can be a difficult concept to wrap the ol' brain around.