Encoding has always been part of digital audio. Lossless formats encode everything in the recording. But if the original recording's highs are clipped, or the gain is too low to pick up quiet sounds, they can't encode what isn't there.
But what if recording levels didn't matter? What if you could forget about the problems automatic gain control (AGC) causes? What if you could start with a sound as soft as the flutter of a butterfly's wings and end with the scream of a jet engine's afterburner at full thrust, and not have to adjust gain -- all while catching every detail in the full range of sound?
Now you can. Think of it as High Dynamic Range for your ears.
Audio is encoded into files in a variety of bit depths. The fewer the bits, the less precisely the digital audio can present the analog sound waves -- and the more digital noise is generated.
16-bit audio encodes sounds into one of 65,536 levels. The highest number is the maximum volume. That bit depth enables about 96 decibels of dynamic range.
24-bit audio -- the most common pro format today -- supports over 16 million levels of sound, which translates to 144.5 dB of dynamic range. Better, but still well below the maximum sound pressure at sea level of about 210 db.
Both 16- and 24-bit audio is encoded as fixed point numbers. A new format, called 32-bit float in audio circles, encodes audio in an IEEE-754 standard single precision format: 1 bit for positive or negative; 8 bit exponent; and 23 bit fraction.
Translated into decibels, that gives a range of more than 1500 dB. That's way more range than you'll ever need.
Furthermore, unlike fixed point encoding, there is no specified maximum sound level -- thus no clipping of loud sounds.
Turn the gain all the way up on your 32-bit float capable recorder! You can adjust the sound levels later in your 32-bit digital audio workstation (DAW). Most pro DAWs handle 32-bit float today.
As always, there's no free lunch: 32-bit files are about a third larger than 24-bit files. But given today's multi-terabyte hard drives and SSDs, that's hardly a problem.
The bigger problem is that few audio recorders support the 32-bit float format. But they are out there -- check out Sound Devices and Zoom -- and they cost a few hundred dollars. I expect to see more choices soon.
Once you have 32-bit float sound coming from your audio interface, there is plenty of software that can use it. The free and open source Audacity -- available for macOS, Windows, and Linux -- edits 32-bit float. So do Protools and Logic Pro.
Now for a couple of caveats. If distortion is introduced before recording -- overloaded mic capsule, power line hum -- 32-bit float won't help.
Also, 32-bit floating point digital signal processors (DSPs) cost more than fixed point DSPs. When packaged into a quality audio interface, the difference isn't huge -- but they aren't the cheapest recorders today.
32-bit float is not a panacea for all audio issues, but it relieves the recording engineer of monitoring for clipping during a recording. She can focus on mic choice and placement, room tone, and ensuring AGC is turned off.
Audio is usually the most important and difficult part of any video project. Removing audio level issues from recording makes a tough job easier.
Comments? If you've been using 32-bit float I'd like to hear what you think. I'm planning an audio project and wondering if I should upgrade my Apogee One interface.