Levels and gain
A recording that's too quiet fills up with noise when you turn it up; one that's too loud breaks and can't be fixed. Learn to set the gain so your voice goes in clean and with headroom.
Gain is not volume
They're often confused, but they're different things:
- Gain: how much the microphone signal is amplified before it's recorded. It decides the level your voice goes into the file at. It's on the interface, the recorder or the system's input setting.
- Volume: how loud you hear the signal in your headphones or speakers. It changes nothing about what's recorded.
The typical mistake is turning up the headphone volume because “it sounds quiet” and leaving the gain alone, or the other way around: lowering the gain because the headphones are too loud. Set the gain first by watching the meter; then set the listening volume to taste.
How it's measured: dBFS
In digital audio, level is measured in dBFS (decibels relative to full scale). The maximum is 0 dBFS and everything else is a negative number: −6, −12, −20… The closer to zero, the louder.
| Peak level | Diagnosis | What to do |
|---|---|---|
Below −24 dBFS | Too quiet | Turn up the gain or move closer to the mic; noise will surface when you raise it later |
Between −24 and −12 dBFS | A bit quiet, usable | Fine at 24 bits; if you can, go a little higher |
Between −12 and −6 dBFS | Ideal zone for voice | Leave it there |
Between −6 and −1 dBFS | Risky | A laugh or a shout could clip; back off a little |
0 dBFS | Clipping | Redo the take with less gain |
Try it in the simulator above: raise and lower the gain and watch the waveform go from a nearly flat line to filling the space and, in the end, being cut off at the top and bottom.
Clipping can't be fixed
When the signal tries to go above 0 dBFS, the converter can't represent it and cuts off the wave: the crests become flat. You hear it as harsh distortion or a crackle. Lowering the volume afterwards doesn't remove it, because the information in those crests no longer exists.
Some programs try to rebuild the wave (declip), and with mild clipping they help, but the result is never as good as a clean take. The rule is simple: when in doubt, record a little lower.
If the light on your interface or the meter goes red during a sentence, that sentence is probably clipped. Note it down and, if it matters, redo it right then. It's much easier than trying to rescue it in editing.
24 bits: your safety margin
Bit depth determines how much distance there is between the digital noise floor and the 0 dBFS ceiling:
- At
16 bitsthe theoretical range is about 96 dB; at24 bits, about 144 dB (in practice it's limited by the noise of the mic and interface, but it's still much more headroom). - Thanks to that margin, at 24 bits you can record with peaks around −12 dBFS without worry: no digital noise appears when you raise the level in editing.
- Some recorders and microphones offer 32-bit float, which practically can't clip inside the file. Even so, the preamps have their own limit and it's worth setting them carefully.
For podcasts, record in WAV, 48 kHz and 24 bits whenever your gear allows. The MP3 comes at the very end, when you export.
The level check
The most common mistake is setting levels by saying “one, two, testing” in a weak voice and then speaking twice as loud once the episode starts. Do a real test:
- Sit at your final distance, with the microphone already fixed in place.
- Speak as you will in the episode: tell a story with the same enthusiasm.
- Laugh loudly and say something energetic: those are your maximum peaks.
- Set the gain so normal peaks land between −12 and −6 dBFS and the loudest don't go above −3.
- Record 20 seconds, listen on headphones and check there are no noises or distortion.
If there are several of you, do the test with each person on their own mic: each voice needs its own gain.
Many programs, apps and USB microphones come with automatic gain control (AGC). It raises the level during silences and makes the background noise “breathe”. Turn it off and set levels manually whenever you can.
Meters and a safety limiter
Meters show the level in real time. The most common are peak meters (they show the instantaneous maximum, which is what matters to avoid clipping), and many add a peak-hold marker that stays up for a few seconds. Watch that marker, not the constant movement.
Some podcast interfaces and recorders include a limiter on the input: if a peak approaches 0 dBFS, it catches it before it clips. It's good insurance against laughs and unexpected bumps, but:
- It's no substitute for well-set gain; if it's working all the time, the voice sounds squashed.
- It should only kick in occasionally.
- The episode's final loudness (the LUFS) is set later, in the master, not while recording.
- Gain decides what's recorded; volume only what you hear.
- Voice peaks between −12 and −6 dBFS; never at 0.
- Clipping can't be fixed: when in doubt, record lower.
- Record in WAV, 48 kHz and 24 bits.
- Do the level check speaking as you will in the episode, and turn off automatic gain.
Your turn: find your gain
Record three takes of the same story with the gain too low, right and too high, and compare the waveform and the sound.
- Tell a 30-second story with a laugh in the middle, with peaks around −30, −10 and 0 dBFS.
- Raise the level of the quiet take to match and listen to the background noise; look at the cut-off crests on the clipped one.
- Note where the gain knob sits for the correct take with your voice.
Frequently asked questions
What level should I record my voice at?
Set the gain so peaks land between −12 and −6 dBFS. That gives you enough level and headroom for laughs or raised voices.
Can clipped audio be fixed?
Only partly. Declip tools improve mild clipping, but the lost information can't be fully recovered. The safe option is to redo the take with less gain.
Is it better to record at 16 or 24 bits?
At 24 bits, because it gives you much more headroom to record with moderate peaks without noise appearing when you raise the level later. The final file can be exported at 16 bits or as MP3.