6 min read
32 bit float recording: what it fixes on set, and what it does not
It ends clipping. It does not end noise, distance or bad mic placement. Here is the honest version.
Every recorder released in the last few years advertises 32 bit float. On a documentary shoot where a subject whispers then shouts, it genuinely changes the day. It is also widely misunderstood.
What the format actually does
A 32 bit float sample stores a mantissa and an exponent instead of a fixed integer scale. The usable dynamic range is so large that a signal recorded at what looks like +20 dBFS can be pulled back down in post and reappear intact. There is no digital ceiling to hit.
What it does not do
- It does not lower the noise floor of your preamp or microphone.
- It does not rescue a signal that clipped in the analogue stage before conversion.
- It does not fix a boom that was three metres away from the actor.
- It does not remove room reflections, wind or a fridge in the next room.
The safety net is digital only. If the preamp itself distorts, the float file records that distortion faithfully. Set a sane gain, then stop worrying about peaks.
Delivering to post
Many editors still want 24 bit files, and plenty of pipelines choke on 32 bit float assets. Normalise the float master to a sensible level, then export 24 bit at the project sample rate. Keep the float originals archived, because they are your only route back if a level decision turns out wrong.
A practical set routine
- Record 32 bit float on the recorder, monitor as usual.
- Aim dialogue around minus 20 dBFS so the mix still reads normally.
- At the offload, split poly takes into named mono stems.
- Convert working copies to 24 bit, archive the float masters untouched.
The sample rate and bit depth converter here handles that last step in the browser, with a preview before export and no upload.