Converting 24 bit files to 32 bit floating point.

[Nibiru]

Hey all,

Just wondering, is this not recommended? I want to see if there's any advantage in working with 32 bfp, and was wondering if there's any downside in doing so? Forgive my ignorance. I'm gonna assume there wouldn't be a downside aside from disk space, but does the audio suffer from any side affects?

[NazRat]

Most likely your audio interface is 24 bit so storing in 32 bit float just takes up more disk space. As for the processing aspect, I don't know. DP (and most other DAWS and plugs) use 32 bit float internally. Is there any advantage to not having to up convert to 32 float when loading a file? Don't know . . . maybe less overhead, but taking up more disk space . . . With the horse power and terabyte disks available maybe it's a non-issue.

[David Polich]

I think this has been the subject of a number of threads. Honestly,
I see zero value in recording in 32-bit floating point. There are no
32-bit D-to-A converters available in commercial audio products,
and CD audio is still 16-bit 44.1khz.

[toodamnhip]

I think this has been the subject of a number of threads. Honestly,
I see zero value in recording in 32-bit floating point. There are no
32-bit D-to-A converters available in commercial audio products,
and CD audio is still 16-bit 44.1khz.

Simple question, off the top of my head. Wouldn;t converting from 32 bit to 16 bit be a better form of math for conversion programs like waves and iZotope? It would be a more “even” math equation wouldn’t it?

[daniel.sneed]

[...] DP (and most other DAWS and plugs) use 32 bit float internally. [...]

Of course all DP plugs are 32 bit float. But I wish all third party plugs would be 32bit float, but they are not presently. Many are still 64 bit fixed instead.

In this regard, my main concern is about Izotope plugs, which I use in quite every projects.

As it's been stated many times on this board, 32bit float plugs is much straightforward when it comes to workflow. Just figure this: You only check levels at AD and DA converter stages. That is to say DAW input and output only. No level-related distortion in between to take care of.

[Nibiru]

I think this has been the subject of a number of threads. Honestly,
I see zero value in recording in 32-bit floating point. There are no
32-bit D-to-A converters available in commercial audio products,
and CD audio is still 16-bit 44.1khz.

I understand there are no 32 bit converters out yet. My question comes after reading much of the threads about 32 bit float, here at motunation. It seems headroom and errors are the reasons why anyone would operate in 32bfp. I guess my question was merely to ask if there is a downside in terms of weird artifacts or something. Which there shouldn't be, I know. I'm no genius when it comes to math, so I figured I'd ask around here because there are clearly math people in here.

[bayswater]

Simple question, off the top of my head. Wouldn;t converting from 32 bit to 16 bit be a better form of math for conversion programs like waves and iZotope? It would be a more “even” math equation wouldn’t it?

That seemed like an obvious thing to me too, but when I asked someone who would know (a hardware engineer who builds such things as converters for industrial use) I was told, no, a proper conversion process resamples and rebuilds the audio sample, rather than just selectively dropping or adding data points. He also told me a 32 bit converter would be a waste of effort -- somewhere around bit 27, general background noise would take over.

[Phil O]

...I guess my question was merely to ask if there is a downside in terms of weird artifacts or something...

I've been recording all of my projects at 32 bit floating point for a while now. It's true that there's no advantage to tracking at 32 bits since converters are only 24 fixed, but there are some advantages while editing/mixing. I've had zero problems with weird artifacts, etc. HD's are big enough these days so the hit to storage space is not even noticed. Is there really an audible difference? I don't know, but recording this way gives me a warm fuzzy feeling and that's all that matters.

Phil

[Kubi]

A few important observations, if I may:

32bit float is actually 24bit, plus 8 extra bits to infinitely increase the headroom (think of it as a 24bit audio that can slide up and down a scale.)

Why would you do this? Because you don't lose resolution as you reduce level. And so you don't hit your head as you increase the level. The word stays 24bits long, and the extra 8 bits tell it to "slide" up or down the volume scale, as needed.

So, you can convert 24bit audio to 32bit float and it won't change a damn thing. In fact DP does this all the time - you record 24bit audio, then you mix it. what does DP do? It converts it on the fly to 32bit float, does all the mixing and leveling and compressing and EQing you ask it to do, and then writes it back to disk when you record.

If you record to 24bits, like most of us do, it will write down the 24bit result. If the level is really really low because you didn't pay attention, your 24bit audio may end up being 20 bits long... or 16 bits if your gain staging is really pathetic. Really really pathetic.

If you record to 32bit float, it will write down the audio in full 24bit resolution, and the remaining 8 bits will say, "btw, this is really really low. This guy should watch his gain staging... but we'll help him out."

Now, there *may* be 32bit fixed audio out there (I know there is 48bit fixed, the old ProTools TDM standard) that's an entirely different story.

But as far as 32bit float is concerned, think of it as 24bit audio that's really really forgiving when it comes to levels, both in terms of clipping as well as in terms of losing resolution when printing too low. That's all.

[Kubi]

Ah, one more thing. While it sounds tempting, there is unfortunately zero mathematical logic to thinking that reducing 32bit to 16 bits is somehow cleaner than ie. 24bits to 16 bits. Number one, because bit depth reduction doesn't happen by simply omission. It'd sound awful. Number two, because the 16bit format is fixed, the 32bit format is float, so 32bit float is actually 24bit audio to begin with.

Ditto btw for SR conversion. Zero advantage to go from 88.2 to 44.1 khz, instead of say, 96 to 44.1 kHz. Zero. The algorithms aren't that simple. Thank goodness, they'd sound awful if they were.

[Nibiru]

A BIG thank you for that explanation, Kubi. That's pretty much what I've been reading for the past few days, just needed a "real" voice of confirmation. ok, so it's 32bfp from now on for me!