To dither, or not to dither?
- KVRian
- 773 posts since 23 Apr, 2002 from audio/hamburg/germany/earth/space/unkown!
My 2 cents:
Depending on the DA converter, truncation noise, when coming from higher bit depth signals (64bit) can be quite strong. (I suspect that's what I'm hearing) I mix in headphones, and when I do not dither the signal at the output of my DAW, depth perception is quite off.
Since any dither imparts a sound, it may actually a side effect I'm hearing, nevertheless, the difference is there.
Depending on the DA converter, truncation noise, when coming from higher bit depth signals (64bit) can be quite strong. (I suspect that's what I'm hearing) I mix in headphones, and when I do not dither the signal at the output of my DAW, depth perception is quite off.
Since any dither imparts a sound, it may actually a side effect I'm hearing, nevertheless, the difference is there.
-
- KVRAF
- 2590 posts since 19 Mar, 2008 from germany
Hmm, interesting!
What kind of music? What dynamics? Is the signal or instrument fading
out into the silence? Has the track been normalized before?
Have you made an anonymous A-B comparison?
free mp3s + info: andy-enroe.de songs + weird stuff: enroe.de
- KVRian
- 773 posts since 23 Apr, 2002 from audio/hamburg/germany/earth/space/unkown!
I'll have to a/b, thought (would have to create a dedicated setup, my A/B plugin will revert the signal back to 64bit) it is quite obvious when mixing. Without dither, I tend to overcompensate for sharpness, e.g. my signal will become way too peaky. It is very apparent with transient heavy material, like hip hop drums. The mix will sound very transparent, but boring and empty. The dither basically removes a certain amount of blur, wich would makes sense, if it is the truncation noise that is going away. Anyhow, I'm very aware that audio is a psychologically very tricky, but I'm pretty sure I'm already eliminating that variable to some extent. I'm using an airwindows dither, and those tend to be on the "colorful" side (very obvious with some of them), although I'm using a cleaner one, meant for monitoring (monitoring plugin 24bit).
- KVRist
- Topic Starter
- 316 posts since 9 Jan, 2023
- KVRist
- Topic Starter
- 316 posts since 9 Jan, 2023
One more last (I hope) question: Do you dither only when exporting the final master, or do you tend to do it (if you're going to dither) throughout a project (mid-project mixes, exported tracks using a lot of plugins, etc.)?
- KVRAF
- 10540 posts since 20 Nov, 2003 from Lost and Spaced
- KVRAF
- 9453 posts since 17 Sep, 2002 from Gothenburg Sweden
It's by and large completely irrelevant. Have it on all the time if you believe it makes a difference and off if not. Worry about other stuff instead, that's using your time more efficiently. Don't worry about it. Test it if you're curious and hear for yourself.
- KVRAF
- 5512 posts since 2 Sep, 2019
You are only going to notice quantization error very near -inf.
Let’s look at what dithering does:
It adds extremely low level noise to the high resolution (24-bit) mix. It’s so low that it would get quantized to -inf on its own when converted to 16-bit. But it may on occasion flip the LSB from 0 to 1 in the 16-bit conversion, if it just missed the rounding up cutoff, because sound mixing is additive.
The difference is pretty arbitrary unless it pushed an inaudible bit just into the audible range. This has the psychoacoustic effect of seeming to bring the ghost of sub-audible sound to the surface through partial representation. It results in slightly smoother fades to silence in the very quietest signals.
Let’s look at what dithering does:
It adds extremely low level noise to the high resolution (24-bit) mix. It’s so low that it would get quantized to -inf on its own when converted to 16-bit. But it may on occasion flip the LSB from 0 to 1 in the 16-bit conversion, if it just missed the rounding up cutoff, because sound mixing is additive.
The difference is pretty arbitrary unless it pushed an inaudible bit just into the audible range. This has the psychoacoustic effect of seeming to bring the ghost of sub-audible sound to the surface through partial representation. It results in slightly smoother fades to silence in the very quietest signals.
THIS MUSIC HAS BEEN MIXED TO BE PLAYED LOUD SO TURN IT UP
- KVRist
- Topic Starter
- 316 posts since 9 Jan, 2023
Hmm...okay.jamcat wrote: ↑Fri Feb 16, 2024 7:23 pm You are only going to notice quantization error very near -inf.
Let’s look at what dithering does:
It adds extremely low level noise to the high resolution (24-bit) mix. It’s so low that it would get quantized to -inf on its own when converted to 16-bit. But it may on occasion flip the LSB from 0 to 1 in the 16-bit conversion, if it just missed the rounding up cutoff, because sound mixing is additive.
The difference is pretty arbitrary unless it pushed an inaudible bit just into the audible range. This has the psychoacoustic effect of seeming to bring the ghost of sub-audible sound to the surface through partial representation. It results in slightly smoother fades to silence in the very quietest signals.
Sounds like it could be quite useful, then.
But again, just curious if, after knowing that, if it's something that should be done simply when exporting the master, or throughout the project (mixes, submixes, etc.)?
- KVRAF
- 15277 posts since 8 Mar, 2005 from Utrecht, Holland
Do it only when bouncing down to 16bits, so that should logically only be at final export of the master.
We are the KVR collective. Resistance is futile. You will be assimilated.
My MusicCalc is served over https!!
My MusicCalc is served over https!!
- KVRAF
- 5512 posts since 2 Sep, 2019
Dithering should only be applied during a downward conversion/export that reduces the original bit depth. You can actually add dithering with just a specially prepared white noise audio file mixed at just the right volume, no need for special plugins or converters.irpacynot wrote: ↑Sat Feb 17, 2024 10:02 amHmm...okay.jamcat wrote: ↑Fri Feb 16, 2024 7:23 pm You are only going to notice quantization error very near -inf.
Let’s look at what dithering does:
It adds extremely low level noise to the high resolution (24-bit) mix. It’s so low that it would get quantized to -inf on its own when converted to 16-bit. But it may on occasion flip the LSB from 0 to 1 in the 16-bit conversion, if it just missed the rounding up cutoff, because sound mixing is additive.
The difference is pretty arbitrary unless it pushed an inaudible bit just into the audible range. This has the psychoacoustic effect of seeming to bring the ghost of sub-audible sound to the surface through partial representation. It results in slightly smoother fades to silence in the very quietest signals.
Sounds like it could be quite useful, then.
But again, just curious if, after knowing that, if it's something that should be done simply when exporting the master, or throughout the project (mixes, submixes, etc.)?
But dithering is pretty pointless when only going down to 16-bit. In actual practice you’d have to have dog ears to hear a difference. It becomes far more useful the lower the bit depth is that you’re converting too, but at the same time, the more distinctly its effect sounds like a bed of white noise masking the tails.
And in actual practice, there is a near zero chance that you would be exporting below 16-bit. So is dithering still all that useful or necessary anymore? Probably not really. Using it or not is nothing to lose sleep over.
THIS MUSIC HAS BEEN MIXED TO BE PLAYED LOUD SO TURN IT UP
- KVRAF
- 10540 posts since 20 Nov, 2003 from Lost and Spaced
That's why I use it. I bounce down at 24 bit (48k), but the distribution wants 16 bit 44.1k. The tricky bit is loudness. I notice the LUFS jumps a little higher on the 16 bit dithered one.
- KVRist
- Topic Starter
- 316 posts since 9 Jan, 2023
-
- KVRian
- 1097 posts since 28 May, 2010 from Finland
I guess the true answer is:
You need to listen to both and then decide whether the non-dithered quantization error sounds worse or better than the applied dither. The dither is meant to mask large errors in the quantization error.
However, if the algorithm to apply the dither is sensical, then it's expected that the dithered signal could in most cases sound better than the non-dithered one. Because it's meant to round out the quantization error (so it should not sound worse).
But you might also notice that more advanced forms of dither are more than just an on/off button:
https://external-content.duckduckgo.com ... ipo=images
So based on that, maybe you need to try out the switches and see which one fits best.
I am not sure whether the "Off" mode is for just already dithered files though. Could be. Which also reminds me of:
IMPORTANT: Dither should be applied only once. Otherwise you're mixing noise with noise and you might get phasing effects and/or a growing noise floor.
You need to listen to both and then decide whether the non-dithered quantization error sounds worse or better than the applied dither. The dither is meant to mask large errors in the quantization error.
However, if the algorithm to apply the dither is sensical, then it's expected that the dithered signal could in most cases sound better than the non-dithered one. Because it's meant to round out the quantization error (so it should not sound worse).
But you might also notice that more advanced forms of dither are more than just an on/off button:
https://external-content.duckduckgo.com ... ipo=images
So based on that, maybe you need to try out the switches and see which one fits best.
I am not sure whether the "Off" mode is for just already dithered files though. Could be. Which also reminds me of:
IMPORTANT: Dither should be applied only once. Otherwise you're mixing noise with noise and you might get phasing effects and/or a growing noise floor.
Last edited by soundmodel on Thu Feb 22, 2024 12:29 pm, edited 2 times in total.
- KVRist
- Topic Starter
- 316 posts since 9 Jan, 2023
Hmm...well, I'm keeping everything at 48khz throughout the entire project, individual tracks rendered, the mix, the master. However, most of my tunes are going onto Youtube, and it's my understanding they do 44.1. I'm not sure if I should export my master down to 44.1 with dithering or not.soundmodel wrote: ↑Thu Feb 22, 2024 7:10 am I guess the true answer is:
You need to listen to both and then decide whether the non-dithered quantization error sounds worse or better than the applied dither. The dither is meant to mask large errors in the quantization error.
However, if the algorithm to apply the dither is sensical, then it's expected that the dithered signal could in most cases sound better than the non-dithered one. Because it's meant to round out the quantization error (so it should not sound worse).
But you might also notice that more advanced forms of dither are more than just an on/off button:
https://external-content.duckduckgo.com ... ipo=images
So based on that, maybe you need to try out the switches and see which one fits best.
I am not sure whether the "Off" mode is for just already dithered files though. Could be.
And I'll be honest, I've tried to compare, but I couldn't tell you for sure if I truly noticed a difference (with dithering on or off). I seem highly susceptible to fooling myself with regards to such subtle changes.