Is re-dither needed here?

[IchiOtoSasayaki]

Hi all,
I have a 24bit wav file of a long track (~30 minutes) that was already mastered and I need to split it to few tracks.
I will apply tiny fade in and out to every track just in case.
The original long track already have dither on it, when I split it in cubase, should I apply another dither before exporting the new tracks?
I will export them to the same bit depth of 24bit wav.

btw, the original dither amount was set to 16bit in case the track would be converted to mp3/aac in the future and some systems read those as 16 bit.

Thank you for your help!

[James Perrett]

Since you aren't changing the level, the only parts that would theoretically need to be re-dithered are the fades. If these are short fades of a few milliseconds where there is no sound then I probably wouldn't bother with re-dithering. If the fades are longer then it may be better to re-dither although, given that you have 16 bit dither on there already, the signal would have to be attenuated by more than around 40dB before that dither became ineffective.

[IchiOtoSasayaki]

Perfect! Thank you James!

[Hugh Robjohns]

Whaaaaat?

The track is dithered at 16 bits but you've saved it as 24 bit???? Why? What is this madness?

To answer the original question, dithering needs to be applied whenever the word length is reduced. That's it. That's the rule.

So if you import a 24 bit file into a DAW it's moved into a 32 bit floating point format but that maintains a 24 bit mantissa so nothing has actually changed in terms of the audio wordlength. If you did nothing to it but exported it again you don't need to redither because the wordlength hasn't actually changed.

However, if you do anything to it, like adjust the level, introduce fades, etc (and most DAWs introduce fast fades at cut points) the wordlength becomes longer, and so the rule says it must be redithered to provide a 24 bit output with a constant noisefloor and no quantisation distortion during the fades.

James' point is that if your cuts or fades are performed during silence at the start or end of a track there is no signal to distort and therefore redithering won't make any difference and is unnecessary and that's undoubtedly true in practice.

Bur there is some nuance in this argument... Some mastering engineers put absolute silence — so-called digital black — between tracks, while some leave dither noise throughout.

So maybe those cuts/fades are working on dither noise which will become distorted through the unfiltered fade.... Then again, since we're talking about dither noise at -141dBFS I doubt anyone will notice.

In practice, I doubt most people would notice that a claimed 24 bit file is actually a 16 bit file either, since the acoustic or production noise floor of the recording may well be higher anyway.

But it is rather disingenuous, not to say completely pointless and wasteful of data storage space to save/distribute a 16 bit dithered file as a 24 bit one. The bottom eight bits are just solid noise!

[RichardT]

I agree - though it won’t make a significant difference if you send it to streamers, if it’s going to be downloaded I think you should really distribute it at 16 bits.

[IchiOtoSasayaki]

Thanks for answering Hugh and Richard. This is becoming interesting

I just watched again Dan Worrall's video "WTF is Dither" (at 16:00) to see if I remembered correctly.
He saw that mp3 files can sometimes be read as 16 files by some systems.

So we have exported 24bit wav file as the master that can later be converted to other formats, the idea was that when converted to mp3 from the wav it will have a 16bit already there just in case it's decoded in some player this way.

But admittedly Dan was was referring to converting mp3 from 16bit wav, or exporting mp3 with 16bit dither.

He does not say to render 24bit with 16bit dither like we did!
So what have we done?

I promise Hugh, it will never happen again

So maybe those cuts/fades are working on dither noise which will become distorted through the unfiltered fade.... Then again, since we're talking about dither noise at -141dBFS I doubt anyone will notice.

Yes, there is dither throughout the track so the cut and few milliseconds fades are indeed probably distorting it... although I can't hear it.

But it is rather disingenuous, not to say completely pointless and wasteful of data storage space to save/distribute a 16 bit dithered file as a 24 bit one. The bottom eight bits are just solid noise!

So just to understand further, does this mean that we have lost low level dynamic content due to the 16 bit dither?

Thanks!

[RichardT]

Your noise floor has been raised by the 16 bit dither being applied - dither is effectively added noise.

Normally the noise arising from the recording process outweighs the intrinsic noise floor of a 16 bit dithered signal.

[Hugh Robjohns]

He saw that mp3 files can sometimes be read as 16 files by some systems.

I've not seen the video, but I expect he was saying that some mp3 encoders only encode a 16bit wordlength.

So if you input a 24bit dithered file into such a device it will truncate it and thus lose the dither, theoretically resulting in a distorted mp3 file (but only during extremely quiet sections and fades!).

So we have exported 24bit wav file as the master that can later be converted to other formats...

You have taken a 24 bit file, laid a hulking great slab of dither noise at -93dBFS, and re-saved as a really noisy 24 bit file....

You've still got a CD quality recording, though... but its taking up 1.5× the storage space it needs to!

I promise Hugh, it will never happen again

My work here is done!

So just to understand further, does this mean that we have lost low level dynamic content due to the 16 bit dither?

Strictly speaking, no. The whole point of dither is to retain low level content that would otherwise disappear through truncation. So by dithering it you haven't actually lost it... but you have smothered that low level content in dither noise.

That said, a 16 bit flat (TPDF) dither lays a noise floor at around -93dBFS.... Do your mixes actually have any relevant content at-93dBFS? (Obviously reverb tails etc produce signal that low, but they're fading into noise anyway).

Historically, most commercial music has a dynamic range of 60dB or less — and much less in the case of most popular music genres.

So if your material peaks above -10dBFS or so the 16 bit dither noise will be the best part of 20dB below the quietest recorded sound, so the 0chances are that 16 bit is more than sufficient for a perfect consumer rendition of your musical arts.

It's only if your material really does have huge dynamic range — 100dB or more— or if you need to retain a 20dB headroom margin to cope with the unexpected, that 24bit becomes necessary.

In your case your mastered material is clearly perfectly well contained in a 16 bit format.... as is almost everybody's!

[IchiOtoSasayaki]

Thanks for explaining!
I learned a lot

[Hugh Robjohns]