32 Bit Float Audio BLEW MY MIND!!

I this video I show you the magic of 32bit float audio, how to work with it and how it can save you!

Recorders that do 32 bit Float

Sound Devices Pre-Mix Series

.......................................................................................................

[YeahWhiplash]

When I was interning at a Orlando production house one of the audio gurus was mentioning 32 bit audio and how it's the raw of video. Seeing what you've done with it has made me a believer, I hope this becomes a standard for audio production!

[MichaelScrip]

I have a Sound Devices MixPre-6 II and I love 32-bit Float! I am a one-man crew and sometimes I can't monitor audio. I shoot dance recitals and other events. Now I can do a bag-drop and leave my recorder backstage plugged into their soundboard. Of course I will do a sound check before the show... but I don't have to worry if the levels get too loud (or soft). It's a great insurance policy! Then I sync the audio and video in post and all is well! My next trick is to get into timecode. My MixPre has built-in timecode and I want to get a Tentacle Sync for my camera. Then it'll be even easier to sync!

[PubRunner]

I just brought a Zoom F2-Pro and it has solved all my audio issues having the lav mic in my helmet for Motovlogs . I use Davinci Resolve and it just works, in my helmet I can overpower the mic if I shout. I love it.
BTW I have the same speakers you have on your other PC, I like them too.

[gwpaulsen]

Wow! Thanks for sharing this (and working on the edit, not just yakking). To have such a big window of dynamic range to work with is just INCREDIBLE. Agreed: BRAIN EXPLODED. New best friend to every one-man-band out there.

[niltonmedeiros]

The power of 32 Bit Float is in its calculation when processing information. If you use 24bit, the calculation truncates information during processing. In 32 Bit float, it retains more of the information. I think of it this way; when using division, some calculations had remainder numbers and others did not. 24 Bit audio is the number with a remainder and simply ignores it - truncate that information - gone for ever. 32Bit Float is the calculation that has the remainder numbers which it retains and uses.

[SpeederXL]

That seems awesome and saw just recently that there are recorders that can record 32 bit float now. I'm thinking about to buy one my self. There still has to be an analog maximum level where it clips though (because if not - that means it will handle infinitely high input voltage - wich is impossible). But, I'm pretty shure the thing is that the recorder always record full range up to the maximum signal level (wich is about the same as the line input of a normal recorder, when the gain set to minimum). Any gain is just an adjustment of the "floating" 0 dBfs point.
For low level recordings, there is also a limit in the form of the "analog" noise floor. If you record at low enough level, that noise floor will get above the self noise of your microphones - which makes your recordings noisier than they would be with more gain.
Even if there is no practical lower limit when bit depth artifacts become a problem, you may still need external gain with a good preamp, if you have microphones with low self noise that also has low sensitivity. But indeed, in most cases, very little gain can always be used (even when recording very quiet sounds - since the volume can just be increases in post production, without any quality loss) - wich really minimizes the risk of clipping, if a sudden unexpected peak apphears. It's usually mostly when recording those quiet sounds (for example ambient sounds) with a lot of gain the risk of a sudden unexpected peak, is at it's greatest (just simple things, like someone dropping something on the floor/ground or suddenly shout or close a door, can cause a heavy clipping in those situations, haha)

[PowersBenzoCoaching]

Excellent video. How did you bring up the audio without raising the noise floor! And can this be accomplished in Logic Pro or Final cut?

[tylerlukkasson]

Great video, bro! Thanks for your insight into 32-bit. Never really understood it before now, and your video helped clarify it for me.

[ShaiYammanee]

That blew my mind!
I need to check this out even more.
This would be such a useful upgrade to my gear

[ShaunTrillo]

just gonna like off the bat for cracking a beer in the beginning

[jettlifeproductions]

😂😂🤣 Brian preparing to play a clipped file like he was waiting for a bomb to go off 😂🤣🤣🤣

[chicagowebsiteservice]

I tried this by recording some video on my iphone 12. The properties in premiere pro confirmed it was recorded in 32-bit float and my project was also 23-bit float. But when I edited the audio in adobe audition and lowered the volume like you did, the audio was still clipped just at a lower volume. What am I doing wrong?

[stroker]

32 bit float audio is still 24 bit integer audio. 32 bit integer audio is even better, but not the same as 32 bit float, so don't mistake them. Even though your recorder can record in 32 bit float, your AD/DA is only a 24 bit converter.

[randomgeocacher]

Translating the float into more humanly understandable terms might help understanding why it works; 32bit float audio stores a “perfect” 24 bit integer audio along with a 8bit “loudness” value. The loudness part is extremely efficient and because it’s value exposed to the power of 2, allowing it to go from -2^128 to +2^128 (which is an insanely large number, effectively infinity). So presuming all of the real world physics/electrics works out, in theory you’ll always have a perfectly resolved audio signal regardless if the loudness is almost zero or almost infinity.

On the actual implementation side, there’s a number of different FpADC designs. Most FpADC rely on reading the same electrical signal at very different amplification levels, and then cleverly determining the best value from one or more integer ADCs (just like HDR algorithms retrieve more/better resolution from multiple images), and then converting that number to float format. So essentially you can build an FpADC with a few op-amps and a couple of intADCs and some tiny processor for the number-crunching.

[Digital.Done.Right.]

thanks for this easy example. i have set my MixPre to this and plan on trying it out. Just will say, not all DAW can handle the format

[PeterMossUkulele]

Great video man - I only found out about this myself today and you explanation was very clear and concise, so thank you. Have a question You have sold me on the concept but like RAW video/photo files you were referring to earlier, they take a lot more room. Is there a calculation to the amount of extra data required on a SD card, say comparing a recording a sound file of the same duration at 44.1k 24 bit to a 32 bit float file?

[BarefootMediaTV]

if that doesnt work is it too far gone? I just did a test to see how to repair my 32bit float audio but i guess i had it too loud?

[Reticuli]

Wait, are you sure real floating-point math has anything to do with the ADC stage? My understanding is 32bit float ADC is marketing gibberish by either having a huge amount of extra (usually unused) headroom or having multiple ADCs listening to the same signals and switching between them based on levels or summing them. The technique of having plenty of extra, fixed, hidden headroom is not new, as even 24bit did that sometimes. This works out fine because useable thermal dynamic range is really like only 18-20bits, anyway, so you put the mixer's input clip at like 4 to 6 bits equivalent-dB above the top meter LED. Real floating point I believe is a DSP-domain math medium, only, not something that actually can be done at the ADC, DAC, or even usually stream transmission stages (SPDIF, AES/EBU, etc). It's one of the reasons there's not much utility in float in a live environment where ADC, DAC, and transmission stream are all happening on-the-fly and you have to watch the levels, anyway. Technically, float loses out with low-level accuracy near the zero crossing compared to the same bit depth fixed point, and if you have to prevent the output from clipping, anyway, there's no point in using a lossy math technique. Float has become ubiquitous because it's used far outside the audio realm, like RF, telecom, etc, so you don't have to write this code by scratch. There are libraries of this stuff available to use for a given chip you're using

[MorelliMedia]

I scanned a few comments and didn't see this. Can you make the fix in the NLE, I'm in Premier, without going to an outside program? Thanks.

[mj]

hmm...what am I doing wrong here? I am using a Tascam x6 and get similarly distorted audio, but when I edit in audition (as shown here) and bring the DB down, nothing is recovered. The audio simply gets shaved off where it clips and the whole thing just gets quieter...does that make sense?