Quote:
Originally Posted by
Moondog007
➡️
OK.... can we summarize again for those (like me) who are slow to understand all this. Can we make an easy check list of best practices around dither? I'll start based on my understanding but am sure to be wrong, so please correct/add/delete these points.
1) Always dither when you are bouncing?
2) Always dither to the sample/bit rate of the bounce
3) Always dither when sending stuff round trip out to analog gear.
4) Be sure to use stereo dither? (not sure which company does/don't do this)
I'm very unsure about which type of dither to use where in the process. Noise based dither versus? mmm MAybe some best practices around that as well would be great.
Rules
The basic rule is dither on word (sample) length reduction (truncation, including rounding for this discussion).
The unwritten rule is that it doesn't matter if the lowest bit level is not reproducible (or hearable). Now, before people lose their minds, every plugin and DAW you use or have ever used churns out huge quantities of truncations, continuously. Most of these don't propagate to your DAC, not even the 24th bit, so it doesn't matter.
Practical interpretation
Because we rarely generate audio streams or files other than 16-bit, 24-bit, or 32-bit float, and because your DAW is almost certainly working in float, we can summarize as:
Dither when you save as 16-bit.
Dither when you save as 24-bit. (Me: Not that it matters, since it's not hearable at ~-140 dB, and your DAC is unable to reproduce it anyway. The unavoidable error due to electrons and the world is ~3 bits, so fiddling with a half-bit maximum error is silly. But do it anyway—why not? It's just a setting.)
The idea of trying to dither float is silly, because the error inherently floats too, so stop there.
Noise shaping
I prefer plain old TPDF, but at 16-bit you can play around if it's important to you to get the apparent noise floor below ~-90 dB. Kind of silly shaping 24-bit, though.
Your questions
Quote:
1) Always dither when you are bouncing?
it's not the bounce, it's the truncation. So, if you're bouncing to 32-bit float, no. The others, yes.
Quote:
2) Always dither to the sample/bit rate of the bounce
I don't get this question. Dither is not about rate, it's about reduction of work length.
Quote:
3) Always dither when sending stuff round trip out to analog gear.
If it makes you feel good.
Quote:
4) Be sure to use stereo dither? (not sure which company does/don't do this)
Probably different opinions on the importance, but it boils does to this: If you can hear the hiss (pretty loud monitoring of quiet passages, if you do, and you better have an excellent amp), then the difference will be whether the hiss is mono, or you perceive it as more spacious stereo. This is another thing that is of zero consequence for 24-bit, because the dither is too far below the Johnson-Nyquist noise of even perfect audio gear. The hiss will
always be multichannel for 24-bit.
Me:
I personally don't see the point of 16-bit anymore, but recognize there is still 16-bit distribution for people who do that. Basically, concern for any of the items you mention disappear for 24-bit audio. Even if you still want to dither, the details like noise shaping and per-channel dither no longer make sense.