Using 32bit Float Point

[DonaldM]

Originally Posted by Darryl Ramm

No--as others here have said. You cannot invent information/data out of thin air. If fixed point data is clipped, it is clipped, data gone/not there. You cannot magically bring back missing data. But 32 bit float formats help avoid you losing data when things would otherwise clip during digital processing inside Pro Tools (or just about every other DAW).

Okay, thanks!

[JFreak]

Originally Posted by Darryl Ramm

Fighting semantics here, but that is really in the analog domain (or when truncating to fixed point). Outputting in the "digital domain" with 32 bit float would be sending somebody a 32 bit float file. And that can contain values that represent more than 0dbFS.

Yes and no.

Yes the 32bit-float won't clip the audio, but no, the 32-bit float still cannot exceed 0dBFS.

In the digital domain, there is this absolute of zero dBFS which cannot be exceeded. Period. If you fail to understand that, you shouldn't post what you just did.

What you output to 32bit-float can never exceed 0dBFS but once you hit that limit the mantissa will take care of the nasty thing that 24bit-fixed would just clip. But it will never go above 0dBFS.

[feliponk]

Originally Posted by JFreak

Yes and no.

Yes the 32bit-float won't clip the audio, but no, the 32-bit float still cannot exceed 0dBFS.

In the digital domain, there is this absolute of zero dBFS which cannot be exceeded. Period. If you fail to understand that, you shouldn't post what you just did.

What you output to 32bit-float can never exceed 0dBFS but once you hit that limit the mantissa will take care of the nasty thing that 24bit-fixed would just clip. But it will never go above 0dBFS.

Sometime us, the audio engineers, know a little bit about this, right?

[JFreak]

Originally Posted by Darryl Ramm

Fighting semantics here (...) a 32 bit float file (...) can contain values that represent more than 0dbFS.

No.

0dBFS is the absolute maximum, and whether you use 8bit or 16bit or 24bit or 32bit or (...) 256bit word length, you won't be able to exceed 0dBFS (the absolute) -- on the contrary, you will be able to represent values much lower than that, giving you more dynamics BELOW the absolute.

[feliponk]

Originally Posted by kdarbyshirebryant

Output yes. Manipulation ie mixing and storage no.

NO, you have more margin BELOW but 0 dBFS is always the max.

[feliponk]

Originally Posted by kdarbyshirebryant

By the way I really don't like the insinuation that I'm not an audio engineer.

If I offended you, I apologize.

Enough for me on this matter, 0 dBFS is still the max, at least for all that I know after 24 years.

[Darryl Ramm]

And here is McDSPs take on AAX plugins and how great 32 bit float is because it allows > 0dB. http://mcdsp.com/2013/08/06/can-you-spell-aax-3/

You need to move past the old (true of fixed point) claims that 0dBFS represents the maximum value the system can represent. That is simply no longer true when considering how the systems use floating point values *internally*. And is immediately obvious to anybody who looks at how these float values are technically encoded/represented.

[feliponk]

Originally Posted by Darryl Ramm

It would take a complete idiot of a software developer when converting from fixed to float format to peg the maximum float number representation at 0dB. Fixing this clipping issue during data processing is *the* reason float data formats was developed/is used. And you really think every DAW developers are that stupid?

So if you are correct, why not a single DAW, including Pro Tools 11, shows levels above 0 dBFS when using digital audio scales?

dBFS dB Full Scale

0 dBFS represents the highest possible level in digital gear. All other measurements expressed in terms of dBFS will always be less than 0 dB (negative numbers). 0 dBFS indicates the digital number with all digits ="1", the highest possible sample.

[Darryl Ramm]

I do not know how many different ways I can keep saying the same thing and have you not listen/get it. 0dBFS means full scale as realized at an input/output. Internally a floating point format system can (and does) handle values above this. If those values were ever output as is they would clip. But internally they represent >0dB and can be processed without internal clipping, and then scaled back before output if needed.

That is why everybody cares about/wants a 32 bit float format/is the exact reason DAWs support it to begin with.

Did you read the McDSP article link? You want to believe they don't know what they are talking about?

[dr_daw]

0dBFS is exactly that, FULL - SCALE, meaning that it is absolute. Regardless of YOUR belief system, as previously mentioned 8bit, 16bit etc...0dBFS is the digital maximum of data being captured, anything outside of that is NOT present and physically impossible to recover.

Are you using an internal signal generator with this experiment? I guarantee you that if you use an external tone generator and clip your input AD you will NEVER recreate the lost signal. Remember, there are only a handful of AD converters that are actually 24bit. The vast majority of converters in our interfaces are 20bit. Do the math, find out for yourself. 1 bit = 6db, so if your interface has a dynamic range of 109dB of Dynamic range, you're dealing with an 18.2bit (likely 20bit) converter.

Simple wiki search,

24-bit digital audio has a theoretical maximum SNR of 144 dB, compared to 96 dB for 16-bit; however, as of 2007 digital audio converter technology is limited to a SNR of about 124 dB (21-bit)[12] because of real-world limitations in integrated circuit design. Still, this approximately matches the performance of the human auditory system.[13][14]