Will floating point change the way we record?
All preamps have a gain control. Or do they? Floating point recording may do away with the gain control forever.
It was 1980, or the first half of 1981, no later than that. I sometime previously had the opportunity to play around a little with a digital recorder (compact disc came out in 1982). I can't remember which model it was, and it was almost certainly a prototype system. But what's left of my memory tells me a) that it was 14-bit, compared to today's standard of 24, and b) that it sounded wonderful.
I'm comparing my wonderful 1980ish experience of digital audio to analogue tape and my experience of the Studer A80 and Nagra IV-S (the Nagra sounded better than the Studer to me). Anyone who says today that tape sounds great is obviously listening through a rose-tinted ear trumpet. Tape sounds great as an effect, but its noise, distortion, and frequency response irregularities were awful.
14-bit digital audio doesn't sound so promising compared to today's 24-bit norm. But if you consider that each bit gives us, in theory, 6 dB of signal-to-noise ratio, then 14-bit digital audio has a theoretical signal-to-noise ratio of 84 decibels. Compare that to analogue tape where you might get 65 dB if you're lucky.
But in the real rather than the theoretical world you wouldn't quite get all of that 84 dB of signal-to-noise ratio. Similarly to analogue tape it would be essential to set the level going into the recorder to get the signal peaking as close to 0 dBFS as possible, of course without clipping. For live recordings, setting the level was a challenge, and I sometimes wonder why limiters were not used more often than they were. But hey - recording engineers prided themselves on their level-setting skills just as the photographers of the day took pride in being able to manually focus quickly and accurately.
Notice I said the level going into the recorder, rather than the gain. It's just a matter of perspective. These days we often record directly via the preamp in an audio interface, so the only level-setting control available is the gain. Or if you use a separate preamp you'll connect to the line input of the interface which either doesn't have a gain control, or would normally be set to 0 dB. So it's gain we think about.
In the olden days however recordings were made through a mixing console. So you would set the gain of the internal preamps of the console to suit the console's requirements, then use the channel faders (or group faders or trim controls) to set the level going into the recorder.
OK that was a bit of a digression but I think it's interesting. But the thing is that setting the gain of the preamp is and always has been vitally important. It adapts the level of acoustic sound to the level required by the recorder, whether analogue tape, 14-bit digital audio, or modern 24-bit digital audio.
The dynamic range of human hearing is around 120 dB from the quietest sound that is audible to sound that is painfully loud, so the purpose of the gain control is to fit that into the 65 dB, 84 dB or whatever is the dynamic range of the recording medium, normally by setting the loudest sounds so that they approach 0 dBFS and letting the quietest sounds fend for themselves.
So that conversation in 1980 or 1981...
Since in those days we recorded to analogue tape, and only occasionally saw digital recorders, setting the gain and the level going into the recorder correctly was vitally important. Get it wrong and set it too high and the sound would distort. Set it too low and there would be too much noise.
But then some bright spark, I'd like to think it was me but it probably wasn't, commented that 14-bit digital audio had a dynamic range of 84 dB, so setting the gain didn't need to be as precise as with analogue tape. 16-bit recorders, which were also around at the time - I remember the Sony 1600 - would have a dynamic range approaching 96 dB. If only we could have 20-bit analogue-to-digital converters then the dynamic range would be 120 dB, which is the dynamic range of human hearing and thus of any sound we would ever want to listen to. There wouldn't need to be a gain control at all!
Well that conversation was 40 years ago, and even with 24-bit recording we somehow haven't gotten away from that gain control.
Floating point recording
You may have noticed that your DAW uses floating point digital arithmetic to achieve an internal dynamic range of 1528 dB. Yes that is one thousand, five hundred and twenty-eight decibels. That might not be entirely achievable in practice and I'm not quite sure I would know how to test it. But it's a hell of a dynamic range and you would probably have to power your studio with a nuclear fusion reactor to be able to use all of it.
That part is reasonably well-known, that the DAW has a much greater dynamic range than a 24-bit WAV file. What is a little less well known is that there is an audio file format that can handle all of this dynamic range. And it is...
A WAV file.
Wow that's a surprise. We're are so used to WAV files being 24-bit, or 16-bit which is the surprisingly still relevant CD format, that it comes as a surprise that there can be such a thing as a 32-bit floating point WAV file. We can call a normal 24-bit file 'fixed point' in comparison. I'd say that the 32-bit floating point WAV file needs a snappier name that trips off the tongue a little better. Maybe a 'float' file. Just musing.
So where can you go to hear a 32-bit floating point WAV file? A 32-bit floating point WAV file sampled at 96 kHz?
Now whether your browser can play it or not is an interesting question. If it can't then you can download the file from here.
Oddly enough, the file wasn't recorded in 32-bit floating point; it was recorded as a conventional 24-bit 96 kHz WAV file in Abbey Road Studio 2, using - if I remember correctly - an AEA A440 active ribbon microphone (costing a mere $5799 or so).
But that doesn't matter because a 32-bit floating-point file has no more resolution than a normal 24-bit file.
This shows us that unlike conventional fixed point 24-bit audio, the resolution and dynamic range are different things.
32-bit float achieves its amazing dynamic range by taking a 24-bit 'window' and scaling it up or down as needed. So if the level is high, the other eight bits scale the 24 bits to higher values. If the level is low it scales them down. But it has no better accuracy or precision than a 24-bit file recorded with a good gain setting.
At present, where the 32-bit float file achieves a useful purpose is for working with and storing a DAW project. The actual recording wouldn't be any better than using 24 bits, and you'll still have to set the gain correctly, but as you work on the song and process, edit and bounce, you will create a dynamic range that is greater than 24 bits, and 32-bit float will handle everything that the DAW can produce. So you could use it for your own projects, or save a project for someone else to work on. But you probably wouldn't bounce to 32-bit float as your final mix or master. The world isn't quite up to speed with 32-bit float yet and there is a school of thought that would consider a 32-bit float file to contain a decision that hasn't been made yet, which is where to place those 24 effective bits - high, low, or medium in level.
The end of the gain control
So now we know all about 24-bit fixed point versus 32-bit floating point... Don't be silly, there's massively more complication if you want to delve further into this.
But we know enough. A 32-bit floating point file can contain any audio level from the footfall of a Texas redhead centipede with a limp in its 17th leg on the left side at 100 paces, to a Saturn V rocket exiting the launch pad (NASA still has a couple - maybe they could fix them up and show Elon a thing or two).
Now, if we could only manage to design an analogue-to-digital converter that works in 32-bit float, then a gain control would be completely unnecessary.
Well let's go back to the 1970s again and record on analogue tape with a dynamic range of around 65 decibels above the noise level. Let's also imagine that we wanted to record an orchestra. The dynamic range of an orchestra is much wider than 65 dB and is also rather unpredictable. How could we capture as much dynamic range as possible while at the same time leaving enough headroom to make sure there was no unacceptable distortion on peaks?
One answer is to use two tape recorders (or a four-track recorder). Set one (or two of the tracks) to record the loudest sounds cleanly, and let the quietest sounds sink into the noise. Set the other with a higher gain so that the quietest sounds are well above the noise, and let the loud sounds run into gross distortion. Then copy and edit the tapes adjusting for the difference in level. Hey presto - a perfect recording with the best dynamic range possible from the medium. This has been done, but not commonly. (I forget the reference so if anyone knows, please let me know.)
So in the digital world why not use two microphone preamplifiers, one set to low gain, the other to high? Then digitize the signals using two analogue-to-digital converters. Then stitch the digits together to give the very best signal, and record it in 32-bit float.
And so we have...
The Zoom F6 field recorder, pictured above. Actually I can't see any mention of it having two microphone preamplifiers per input, but it certainly does have two analogue-to-digital converters for each of its six inputs. It can record to 32-bit float with no gain control.
No gain control!
The Zoom F6 can accept a signal level of up to +4 dBu at the mic input, and resolve signals going all the way down to -131 dB below that.
That is simply phenomenal. And the Zoom F6 isn't the only unit with similar feats of achievement. There are the Sound Devices MixPre II models that can also record in 32-bit float without having to worry about gain.
So why are both of these examples field recorders?
Good question. The principle function of a field recorder is to capture audio for a film or TV production on location. Location audio is the most uncontrollable kind of audio that there is, particularly in documentary. So the sound recordist traditionally had to be very watchful over levels, even with limiters engaged. So field recording is where 32-bit float was most desperately needed, and now they have it.
But it isn't going to stay there. Who else needs freedom from the gain control?
I might say YouTube videographers. You might have noticed that the quality of YouTube productions has swung massively upwards in recent years. This is largely because of improvements in camera technology. A modern camera that is sold into the YouTuber market won't set the lighting for you obviously, but it most definitely will make the best of the lighting you have.
This is fair because YouTubers, even with subscriber numbers in the hundreds of thousands, don't tend to have a full film crew milling around them. Often the YouTuber does everything themself, or has just one assistant.
So with stuff to get right such as hair, makeup, background, lighting, framing, script and everything else that has to be done to make a great video, sound is yet another thing that needs to be right.
And if sound can be recorded without having to set a gain control - Well what's not to like about that?
So I can see the next wave of 32-bit float recorders being in the so-called prosumer market (using the first definition in the link). Maybe it will creep into the audio function of cameras themselves without announcement - auto gain setting just becomes no gain setting.
And from there it can only be a matter of time before the DAW recording studio goes entirely 32-bit float.
Where are the gain-free audio interfaces?
Google knows everything. So if a Google search for '32-bit floating point audio interface' brings up very little of relevance to music production in the first five pages, which is as far as I went, then we can assume that 32-bit floating point audio interfaces are thin on the ground.
Yes your DAW can probably record 32-bit floating point, but it's only working with the 24-bit fixed point signal that your conventional interface sends it.
What's going to be needed is an audio interface with internal preamps capable of a very wide dynamic range, followed by a 32-bit floating point converter, or converters as in the Zoom F6.
In 2022, 2023 or so I would expect to see a useful selection on the market giving us wide access to 32-bit floating point recording.
But then what becomes of our beloved preamps? The tube-driven, transformer-saturated, warmth generating tools of our trade?
Well I'm guessing that they're not going to go away. If we like the way they sound then we will continue to use them. We will record in 32-bit float but you'll still have to set the gain, but more for variation of sound texture than getting the level right.
I would imagine through that where today there are many audio interfaces with internal preamps, and we use an external preamp as an accessory, future preamps will have internal 32-bit floating point analogue-to-digital converters and will output 32-bit float directly via USB.
I'd like to summarize, but this is all so new that any meaningful summary is not going to come any time before the year 2025, and perhaps not even by then.
But I would say that 32-bit floating point audio has now come of age and that gradually it's going to take over our studio lives.
Give up the gain control? I can't wait.