Background info: https://www.sounddevices.com/32-bit-flo ... explained/

I was talking to my Sweetwater "engineer" who is extremely knowledgeable and a progrock/jazz musician guitar player. I was interested in getting a stand-alone digital recording device (I eventually got the Zoom H4n Pro, by the way, which his awesome for Sticks. Two inputs on it, that take XLR or guitar plugs, and also will output 48 v phantom power, and also have two onboard mics for recording. Will do 96 kHz and 24-bit recording.)

He suggested I get the ZOOM F6, which is also a nice bit of gear, but it runs $650 and I don't need 6 mic inputs, and no guitar jack inputs. I'd have to have adapters and it also has no onboard mics, and I wanted those for recording sound effects in the world, and of course, piano performances with the kids, and orchestra, and other uses. It's also a bit bigger than I wanted to lug around.

But to the point of the article, 32-bit float recording, my engineer mentioned that the F6 could do that, and why that was a big deal: you don't need to set a level. You literally can't clip the signal unless you physically crush the mic with too much sound pressure. With the sound floor range and dynamic range that 32-bit float recording offers, you can just pull up any level to match the level you want without any noise or distortion. I've seen some ridiculously low levels pulled up to regular normalization without any noise, and more importantly, tracks that would normally be ruined by some clipping or have to be patched and repaired and punched in, are easily fixed by just pulling down the level in post. See the article for examples, and this video shows it very well.



But then I got to thinking--wait, my Focusrite Scarlett i2, for around $200 or less, has 96 kHz (actually 128 kHz now!) and will do 32-bit float recording. I even found a 6-year old project file that coincidentally I'm redoing the vocals, and saw that I had recorded the project in 96 kHz, 32-bit float. Not knowing what the heck it was, I just set the project at the max settings. And some of the tracks had some clipped vocals, that I thought were ruined. I just pulled them down and they were perfectly fine! I pulled up some other tracks that were too low and I had tried to normalize some of them, but the really low tracks I didn't even bother with at the time. I pulled them up a huge amount and they were just fine.

So, 32-bit float recording? How about that! Please share your own thoughts and experiences on that. I went to school for digital sound synthesis back in '92 and this was not even a thought at the time that I recall ever hearing about. Now I feel like I'm on the cutting edge of tech for video, and fairly ignorant of audio tech, even though I went to school for that. But I've never had shame in my own ignorance, because I'm always learning and happy to learn more!

And I'm gonna live in 32-bit float land for a bit! (I already hear you on "Set the damn levels right in the beginning!" and actually don't have a great response to that very clever observation, whoever makes it first!)

Adds: here's a nice roundup of the F6 and that 32-bit float recording.

Heh! This sounds like the same debate you see in photography land with people who shoot RAW vs those who shoot JPEG.

I think it’s about the dynamic range of the device, not the number of bits per se. A digital circuit usually has an analogue front-end, right? The number of bits and the integer or float encoding are still governed by the headroom of the analogue pre-amp and how it’s aligned to the ADC (likewise the DAC, on output).

It’s true in theory that float encoding allows the capture of very fast transients, but so does 24- or 32-bit integer. Float encoding trades accuracy in the bigger signals to gain greater range. To my thinking, the benefit of more bits is what happens when you mix signals together.

Rather than bits, these days I tend to check the spec sheets of the gadgets and look at the headroom, dynamic range, distortion etc.

For me, I suspect I’m mostly chasing specs I can’t hear. In my twenties I could tell the difference between 48 khz and 44.1 kHz reverb tails in a blindfold test. A few years later I couldn’t pick the difference. So I’m not sure what I’m doing with these 96 kHz converters apart from taking up more storage!

And if I’m listening to compressed muzak in the car, or on earbuds... it doesn’t matter, I reckon.

Interesting subject.

Interesting - it looks like the F6 and the MixPre II use multiple fixed point ADCs to cover different dynamic ranges (to widen what would be possible with just one ADC), and somehow blend between them to calculate 32-bit floats for the host. I can see how that could be useful, to cover the entire dynamic range of a microphone for example and not worry so much about signal level.

Once inside a DAW, audio data is typically always processed in 32 or even 64-bit floating point while mixing and applying effects, where the massive data range (and also using high sample rates) is beneficial. But you’ve still got these relatively small portals to get in and out of this large sphere. While it’s physically impossible for a preamp/ADC to have the 1500dB of dynamic range provided by 32-bit float, these newer interfaces seem to use this multi-fixed point ADC approach to increase the input dynamic range to something like 130 to 142dB (which could cover the whole range of most microphones), then generate a 32-bit float from that, as opposed to the more typical 110 to 130 dB for 24 bit interfaces.

If you did manage to clip the input stage of these devices with a hot enough signal, it would be “baked in” to the audio data, and reducing the volume later wouldn’t remove the clipping. In the sounddevices.com sample file test, it looks like they record the audio first, then apply gain afterwards in software to clip the signals, then write out 32-float and 24-fixed versions of files to work with. In this case, the 32-bit version would certainly be recoverable in a DAW that reads 32-bit files.

Hi, Rob! I can verify with experiments over the weekend using my Scarlett 2i2 interface and "32-bit float" recording, that if you clip the sound on the recording, you can't just lower it later. The F6 and MixPreII have those dual ADCs to handle the wider range. My Scarlett does not apparently. If I clip it on the recording, it's clipped. Buuuuutttt! If my level is too low, then it raises easily. And if you push it past clipping through FX or whatever, you can always lower it before that stage.

So, without that fancy F6 (or the F8n for a grand!), you can't just yell and scream and whisper into any mic and be able to rescue it later. At least, if you push it into distortion on the recording. Seems to me I just back off my input a bit, or even put a limiter on things (but really, just learn this new SM7B mic I just got, and then it should be eazy-peazy from here on out), and I can still play in 32-bit float with the Scarlett.

But if I wanted to drop a grand on a device that would let you record anything anywhere and never worry about levels, it would probably be the Zoom F8n. But alas for now, I'll just play with the Zoom H4n Pro and my Scarlett 2i2 and a Mac Book Pro or some other portable interface, and be quite happy. But I hope this was something to think about and have some GAS gear lust for! And it's the wave of the future, perhaps? ("The future is already here; it's just not evenly distributed yet." William Gibson from memory.)

I have the H4n and although it is handy, I prefer my studio setup.
I use a Rode NT1A with a Behringer Mic-100 tube pre-amp.
With this setup, it can even pick up the sound of me scratching my head.
http://www.rode.com/microphones/nt1-a
https://www.behringer.com/product.html?modelCode=P0207

I also have the Rode NT1A with a Behringer Mic-100 tube pre-amp. At least, I have the pre-amp, but I've not found it to be that un-noisy, in my experiments with it. My Sweetwater "engineer" laughed at me when I mentioned that I might want to use that on my SM7B as a pre-amp. He said it would be far noisier than using the Cloudlifter, which uses the phantom power +48V to make 25 DB of clean gain without any noise or coloring (or much, anyway). I didn't A/B test them, but I've been using the Cloudlifter now and really liking it, on my new SM7B (I recorded the other night with both mics side by side and in stereo. Sounded pretty much the same to me! But the Sm7B works better in noisy environments, and the Rode picks up EVERYTHING, for better or worse.) But I do have that Behringer pre-amp and some people (obviously you) find it works well with them, and others find it noisy. I haven't used it in a while, though. And that same Sweetwater engineer was happy to sell it to me when I bought it a couple of years ago, as "a great pre-amp!" so....[shrugs]....

The Rode NT1A is one of the best studio mics that I've seen. Some people are modifying them to compete with some very expensive mics. I originally bought the Behringer Mic-100 from Axe Music in Calgary for $30CAD new. I was so impressed with this little tube pre-amp that I went back and bought two more. You are correct, when you say that it picks up everything. This is why I only use it as a studio mic. I'm very impressed with the combination. Most mics would completely miss some of the audio that it is able to pick up.

On a side note, previously I mentioned that I'm fussy about the strings that I put on my instruments and I make my own custom sets. I'm also fussy about the instrument cables and I can hear the difference between two different brand cables. Recently, I discovered something else that I didn't expect. I moved a Helix Floor preset from one setlist to another setlist as a backup. I could hear a difference between the same preset located on two different setlists. I can't understand this. It's the same preset being played thru the same device. Why would it sound different between the two setlists? It's puzzling me.