SynthMaker Production Manual Open SynthMaker (opens in a new tab)

What are decibels, and what is headroom?

A decibel (dB) compares two levels: every 6 dB up doubles the size of the signal, about 10 dB up sounds twice as loud, and headroom is the gap, in dB, between a mix's loudest peak and the ceiling it can't go past.

Decibels compare two levels

A decibel never says how loud something is on its own. It says how much bigger or smaller one level is than another.

Change in sizeIn dB
× 2 (double)+6.02 dB
× 0.5 (half)-6.02 dB
× 0.1 (a tenth)-20.0 dB

Decibels add where sizes multiply. Halve twice and you have a quarter: -6.02 dB + -6.02 dB = -12.04 dB.

Loudness is different again. Your ears don't hear twice the size as twice as loud. A common rule of thumb, from listening tests, is that about +10 dB sounds twice as loud. So a part at -20 dB sounds roughly a quarter as loud. The rule is approximate: it changes with how loud you listen and with the sound.

The digital ceiling

Digital audio has a hard ceiling: the biggest number it can store. It's called 0 dBFS (decibels relative to full scale), and every level below it is negative.

Push a sound past it, and everything that would go over is cut off flat. That's clipping. The flat tops add harsh, buzzy overtones. Done on purpose, it's a kind of distortion (chapter 3). On a whole mix, it usually sounds broken.

Headroom

Headroom is the room left above the loudest peak. A mix whose highest peak is at -6 dBFS has 6 dB of headroom: it could get twice as big before it clipped.

While you mix, leave plenty. Peaks from different parts sometimes land at the same moment. Lift, a song played by SynthMaker, shows why. Each of its stems (each part as its own audio file) peaks at -4.3 dBFS or lower. Added together, they peak at +1.9 dBFS: over the ceiling.

At the master, the song is brought close to the ceiling, but not right up to it: Lift keeps 1.5 dB of headroom. Why not use it all? The smooth wave between two samples can rise a little above both, and poke past the ceiling when a player rebuilds it. That in-between peak is the true peak, and SynthMaker's published music keeps it under -1 dB.

SynthMaker has a safety net. The Master at the end of every sound has a limiter, a control that turns down only the loudest peaks. It keeps every peak under -0.3 dBFS, so a patch never clips at its output.

Listen

1. Steps down: Steps down. An even "shhh", like a waterfall (pink noise), at 0 dB, then -6 dB, -10 dB and -20 dB. Listen: -6 dB is a clear step down but not half as loud; -10 dB should sound about half, and -20 dB about a quarter.

Steps down0:00 / 0:13

2. Headroom, and none:

  • Clean: 16 seconds of Lift, finished.0:00 / 0:16
  • Clipped: the same made 4 times bigger (+12.04 dB), with everything past the ceiling cut flat. That cuts 30.8% of the samples. It's turned back down to the same loudness, so only the clipping differs. Listen to the kick and the bass, where the biggest peaks are.0:00 / 0:16

Try it in SynthMaker

  1. Open the Init sound: a sine wave through a VCA to the Output. Put an Analyzer between the VCA and the Output, choose its Scope, and hold a note.
  2. Set the VCA's level to 1, then 0.5: the wave is half as tall, 6.02 dB quieter. At 0.1 it's a tenth, -20 dB.
  3. Now clip it. With the level back at 1, put a Shaper between the VCA and the Analyzer, shape set to Hard. It multiplies the sound by its drive and cuts anything past the ceiling flat. Turn drive up from 1 towards 4 (+12.04 dB). The wave's round tops turn flat, and the pure sine turns buzzy. Set mix to 0 to hear the clean sound again.

Where the numbers come from

  • Worked out: every dB in the table and the steps (dB = 20 × log10 of the change in size).
  • A rule of thumb, not measured here: +10 dB sounds about twice as loud.
  • Measured from Lift's stems and the finished song: the peaks, and the share of samples clipped in example 2.
  • From SynthMaker's engine: the Master's -0.3 dBFS limiter, and how the Hard Shaper clips.