Dithering from 32 Float to 24 Bit

[Bob Olhsson]

Actually I was testing going from mix files written to 32 float and then dithered to 16 vs. mix files dithered to 24 and then dithered to 16. I agree much of the damage has probably already been done by truncating to a 32 float file but I wouldn't assume 32 float and not higher precision is what is being dithered to 24.

Really the best test is to simply listen for yourself.

[DJ Hellfire]

Originally Posted by Zarabozo

If neither your recordings or your mix are clipping anywhere, you don't need to do anything.

What most people seems to not understand about the 32 bit float on PT10, is that it's an invisible feature. Your metters still work the same way as if you were working at 24 bit. You can actually switch the session parameters from 24 bit to 32 bit float without actual changes to the files.

The reason for this is that 32 bit float is the same as 24 bit, except it has a lot more headroom. If you clip a recording in 24 bit for 3 dbs, the file gets clipped. If you clip by 3 dbs while using 32 bit float, the mixer metter shows your recording clipping, but the file supports that clipping with that extra headroom (meaning that you can lower the clip gain and have no distortion at all).

If you convert your 32 bit float session to 24 bit, you don't need any dither because the only thing getting truncated is the extra headroom. Any wave information on the file bellow that extra headroom is perfectly compatible with 24 bit and show the same levels in the mixer.

The keyword here is "float". It's not 32 bit vs 24 bit. It's 32 bit float vs 24 bit. 32 bit float "peaks" at the same exact point as a 24 bit file. The difference is that it has the space to store more bits over that peak point, while 24 bit simply hits the roof and distorts.

HTH

Francisco

Thanks a lot for this explaination! This is the only thing I've read on the subject that makes me fully understand it! You also answered all the questions I had about 32-bit float!

[daeron80]

Has everyone forgotten that PT does it's processing and bussing at 32 float? Unless you have HDX or a Native card (in which case there's a mix of 32 and 64-float going on), every time you play back a 24-bit PT session, you're hearing a 32-float stream being converted to 24-bit fixed so it can be sent to your converters. That's exactly the same thing as playing back 32-float files. Whether the session format is 24-bit or 32-float, if you have any mixing, plug-ins, or even gain change going on anywhere in the session, the bit stream at the master bus is taking advantage of 32-float math, and that 32-float signal is being converted to 24-bit fixed so that you can hear it, because there's no such thing as a 32-float DAC.

I have found that a 24-bit fixed file converted from a 32-float file tends to sound better if 24-bit TPDF is added prior to conversion. I understand that the low level detail is then being modulated by signal amplitude. But somehow it seems to make it hold up better to further processing down the line.

[groundcontrol]

:-$
Cha Ching!

[mu-tron-kid]

Originally Posted by groundcontrol

THAT is a good question.
Also did you check your null test with a bit scope?
I'm not sure residual products (such as dither noise) at lsb level would be detectable by ear unless you have an incredible monitoring chain and control room... Not sure but wondering?

Cheers!

Right! To clarify, the null test produces no Residual MUSICAL leftovers.
The only thing left IS the dither noise. (sorry, didn't do a bit scope yet)
The original 24 bit file,
a POW-r Dithered 16 bit Bounce,
an UNDITHERED 16 bit bounce,
and an internally bussed then exported 16 bit file (which gets the default Digi Dither):
when ANY two files are played against each other and one is inverted, sonically identical very low level "Noise" is all that is left (@ full volume on the mains in the studio), even against the 24 bit original.

I did this with an average level 24bit acoustic ballad mix that had been processed with audiosuite for "mastering".

Since the files were INVERTED in the 24 bit session with audiosuite, the default audiosuite 24 bit dither is applied.
But there should be something more definitively different between the noise leftovers given the pages and pages of back and forth over the different Noise Shaping options within POW-r Dither alone.

[drumzalicious]

I believe there is some misinformation here about Bounce To Disk. I've been using BTD when bouncing down ideas and I always tell PT to bounce as a wav file then under it I switch it from 24 bit to 16bit and the file comes out as a 16 bit wav file.

Not only does it show as 16 bits when viewing it in various programs but it plays on my ipod and as some of you know older iPods will only play wav files that are 16 bits.

[gandlz]

just another question:

why (according to the manual) is dither applied when exporting from 32bit to 32bit?
it is also mentioned, that no dither is applied when exporting from 24bit to 24bit.

That sounds strange to me!

And: What dither (16, 18, 20bit) is used when exporting to 24 or 32bit?

Thanks,
Gandlz

[Darryl Ramm]

Originally Posted by daeron80

Has everyone forgotten that PT does it's processing and bussing at 32 float? Unless you have HDX or a Native card (in which case there's a mix of 32 and 64-float going on), every time you play back a 24-bit PT session, you're hearing a 32-float stream being converted to 24-bit fixed so it can be sent to your converters. That's exactly the same thing as playing back 32-float files. Whether the session format is 24-bit or 32-float, if you have any mixing, plug-ins, or even gain change going on anywhere in the session, the bit stream at the master bus is taking advantage of 32-float math, and that 32-float signal is being converted to 24-bit fixed so that you can hear it, because there's no such thing as a 32-float DAC.

You are trying to make some valid points but being let down by multiple inaccuracies. Non-legacy TDM Pro Tools uses a 64-bit mixer, and plugins use a combination of 32-bit and 64-bit floating point processing. That's true of Pro Tools standard, HD Native, HDX etc. Legacy TDM Pro Tools (e.g. Pro Tools 10 HD with a TDM hardware) uses 48 bit fixed point mixer.

Session file formats are all to do with dynamic range in the content file, they affect nothing to do with how the data is processed in the mixer or plugins.

---

If somebody believed that dithering a signal at 24 bits was important then absolutely the fact that the source for that happens to be 32-bit float is kind of irrelevent. It's just a source able to represent signals lower than the 24th fixed-point bit, and what do you do with that 24th bit and beyond signal as you lose that information? Sure if you believed dithering helps there you would dither it.

The issue is by the the you are down to 24 bits you are pretty much lost in the converter and analog noise floor anyhow--unlike dithering at 16bits for CD. So just don't worry about it.

[Darryl Ramm]

Originally Posted by gandlz

just another question:

why (according to the manual) is dither applied when exporting from 32bit to 32bit?
it is also mentioned, that no dither is applied when exporting from 24bit to 24bit.
That sounds strange to me!

Not dithering 24-bit fixed to 24-bit fixed makes perfect sense

Dithering 32-bit float to 32-bit float makes less sense. And what do they mean by dithering there? It's more complicated with a floating point format. It's a bit of a mystery at least from the documentation. If the belief was the file is heading to a playback system that was going to effectively truncate the exponent off and treat it as 24-bit fixed, then sure dithering makes sense. Worse case the dithering is happening down at the LSB of the Mantissa and likely just irrelevant to most practical uses.

Anyhow something that it may be possible to look at with a very low level signal and some careful playing around if anybody really cared.

And: What dither (16, 18, 20bit) is used when exporting to 24 or 32bit?

For 24-bit fixed at least... Pro Tools won't automatically dither at less than the bit depth. It would be throwing away data for no reason, if you are exporting data at a certain bit depth presumably you want those bits. The dither happens in the least significant bit (which yes might involve changing the next bit up as well, but that's it).

[gandlz]

No dither when bitrate is not changed makes also sense to me. That’s why it sounds strange to me as they dither 32bit export to 32bit.
But your idea about it makes sense. In the end you can only play through a 24bit DAC so dither would be needed.

I just got confused by the long processing time when exporting 32bit files. So I now go the session folder and just copy the desired file.
Also I usually bounce my masters to a new track and have dither plugin at the last insert, so I can choose the one I like and also check for levels or inner sample peaks while bouncing. After this I also have to copy the final master from the session folder because otherwise additional dither would be applied in every case (exporting to 32, 24 or 16bit). So I also have to use additional software to make my 16bit CD-Files.

The question about “What dither (16, 18, 20bit) is used…” is because Avid dither plugin does not offer 24bit dither. So I think it would be really stupid to apply 16bit dither when exporting to 32 or 24bit.