(Windows 7 SP1, Audacity 2.0.2, installed by .exe installer)
If I open a 16 bit file, which already includes dither, in 32 bit float, make a small alteration (such as adding silence to the end of the file, or applying some equalization to the bass response) and then export the result to a 16 bit file, will the dither present in the original file be preserved, and if so, should I specify “no dither” in Preferences in order to avoid cumulative dither?
Once the dither is in the imported file, Audacity does not remove it. See here https://forum.audacityteam.org/t/howto-remove-dither-noise-from-silence/15692/1 for how dither can be removed from silence.
As I understand it, a second dither adds about another 3 dB to the noise level, but subsequent dithers add very little further noise. So “no dither” prevents that amount of cumulative dither.
Any effect involving amplification (so potentially including equalization) will increase the level of the dither noise, though the default “shaped” dither is concentrated in high frequencies.
Gale
Many thanks for your helpful reply, Gale.
I should also have asked whether the old dither, not having been removed, will be equally effective in masking the new quantization noise occurring during export. It seems to me that, if all that is required is the presence of some low level random noise, then it ought to be (and I haven’t yet noticed any loss of sound quality in the files I have saved without adding new dither). However, I have seen quite a few recommendations to “add dither every time either the word length or the sampling rate is reduced” (as well as quite a few recommendations to “only add dither the last time a file is saved prior to using it for listening or publication”) and that has made me a bit unsure of my understanding of the way that dither works.
It gets quite complicated when you get into the fine detail, but the brief answer and general rule is that applying dither on the second conversion will at worst do very little harm and at best will be a significant improvement, depending on what you were doing between importing the dithered 16 bit file and exporting the 32 bit float track. As a general rule I leave dither enabled.
What is important is that the if you have done any processing of the original audio track, then the first 16 bit export should have dither applied.
Ideally, if you know in advance that you will want to re-edit the track, export in 32 bit float format and dither will (automatically) not be applied because it is not needed.
If you think that you may need to re-process the track but you still want to export in 16 bit, then “Triangle” dither is marginally better than “Shaped” dither because there is (marginally) less degrading of the sound quality if you apply Triangle dither twice than if you apply Shaped dither twice.
Provided that your tracks are normalized close to 0 dB before export, then all dither noise is at a very low level, so we are only talking here about minute differences.
Just a thought of a specific case where dither should definitely be applied on the second export, and that is if you apply a fade out or fade in. In this case, the level of the original dither will be reduced to less than 1 bit in 16 bit format, and so effectively disappears, so it is necessary to apply dither to reduce harmonic distortion in the low level part of the fade.
Many thanks for those helpful replies, Steve.
Incidentally, I have posted a report about one (24/192) recording that sounded much better to me without dither added again on export, and many others (both 16 and 24 bit) that at least sound no worse to me, at https://forum.audacityteam.org/t/howto-remove-dither-noise-from-silence/15692/10