Microphone Calibration

Most people set their mic level the same way: talk, glance at a bouncing bar, nudge a slider until it "looks right." Broadcast engineers stopped doing that roughly ninety years ago. Calibration means setting input level by measurement against a defined target — and once you've done it, an entire category of audio problems simply stops happening.

Vibes versus measurement

An uncalibrated microphone level fails in one of two directions. Too low, and your voice sits close to the noise floor — the constant bed of preamp hiss and room tone — so any later boost drags the hiss up with it. Too high, and emphatic moments slam into the digital ceiling and clip, a distortion nothing downstream can repair. Between those failure modes lies a target zone, and the point of calibration is to find it deliberately: measure where your speech peaks actually land, measure where your room's noise actually sits, and set gain so both numbers end up where the rest of the chain expects them.

The key word is measure. A level meter you aren't reading against a target is decoration. A calibration procedure reads it for you, against a number — and the number matters, because the same setup that looks fine on one app's animated bars can be 15 dB too quiet by any objective standard. Meters without references are how people end up recording a year of podcast episodes at whisper level, then boosting them in post along with every fan and refrigerator in the room.

Line-up tones: how broadcast solved this

The discipline comes from radio. A broadcast chain in the 1930s passed audio through consoles, telephone lines, and transmitter equipment owned by different people in different buildings, and the only way to make levels survive the trip was to align every stage to a common reference. Engineers sent a steady line-up tone down the chain and adjusted each device until its meter read the reference mark. The meters themselves had to be standardized for this to work: the VU meter, developed jointly by Bell Labs, CBS and NBC and standardized in 1939, gave American broadcasting a shared definition of "zero level," while the BBC's Peak Programme Meter, developed in the 1930s, did the same job in Britain with a faster, peak-reading ballistic.

The tradition carried straight into digital. European broadcasters align a reference tone to −18 dBFS (EBU Recommendation R68); American film and television practice puts it at −20 dBFS. The numbers differ, but the philosophy is identical: pick a reference, measure against it, and every stage of the chain — and every voice passing through it — lands in a predictable place. A guided mic calibration is this same ritual, with your voice standing in for the tone.

The death of "as hot as possible"

If you learned recording in the 1990s, you were probably taught to track as hot as you could without clipping. That advice was rational once. A 16-bit system offers about 96 dB between full scale and quantization noise, and early converters delivered meaningfully less; every decibel you left unused was a decibel of your recording spent closer to the hiss. Engineers rode levels near the top because the floor was genuinely close behind them.

24-bit conversion killed the argument. With a theoretical range of about 144 dB — and well over 100 dB in real hardware — the converter's noise floor sits far below the noise of any room and any microphone you'd actually use. Recording speech peaks at −18 or −12 dBFS instead of −1 costs nothing audible and buys back the headroom that protects you from your own laughter. "As hot as possible" now has all of the risk and none of the reward; the modern target is as measured as possible.

What a calibration actually measures

A useful speech calibration needs three numbers, and each requires a different measurement technique. First, speech peaks: the short-term maxima of your normal talking voice, best captured as a rolling statistic over several seconds of real speech rather than a single instantaneous reading — one plosive shouldn't define your level. (This is peak measurement, not average; see RMS & peak metering for why the two can differ by 12 dB or more on voice.) Second, the noise floor: measured by minimum tracking — watching the level during the gaps between words and phrases, where whatever remains is by definition not you. Third, the available gain: how much boost the device and the software can still apply, which decides whether a quiet raw level is fixable downstream or genuinely too low.

With those three numbers, level-setting becomes arithmetic. Peaks minus floor is your signal-to-noise ratio. Target minus peaks is the gain still needed. And gain staging tells you where that remaining gain should come from: as little as possible from the analog dial, the rest applied digitally after cleanup, where it's mathematically free.

Calibrating a voice for calls, podcasts and streams

For spoken delivery, the widely used target is speech peaks around −12 dBFS: comfortably above any plausible floor, with 12 dB of headroom in reserve. But the raw level at the mic should sit lower than that, because downstream digital gain is clean and analog headroom is precious. What matters at the physical dial is only that your voice lands inside a window the software can work with — high enough that the floor stays distant, low enough that no burst of emphasis reaches the ceiling. A good calibration flow, then, is short and honest: speak normally for a few seconds, watch a measured verdict — up, down, or good — adjust the dial, and let the software compute and apply the remainder.

Two practical notes. First, speak normally during the measurement — your actual on-call voice, not a self-conscious announcer read. Calibrating to a performance you won't sustain guarantees the level will be wrong five minutes into a real conversation. Second, calibration isn't permanent: it captures a physical arrangement of dial, microphone, distance and room. Recalibrate when any of those change — a new mic, a new desk, a gain knob someone bumped — and you'll never have to wonder whether "you're quiet" means you or your setup.

In DeskBroadcast

DeskBroadcast's guided calibration is exactly this measurement, automated. You speak normally while the app shows your raw peak level against a green target zone — by default −19…−7 dBFS raw peaks, adapting to how much digital boost your particular device has available. A rolling speech-peak marker turns the reading into a plain verdict: "Turn the mic's gain dial UP ↑" or "Too hot — turn DOWN ↓." Between your words, it minimum-tracks the room's noise floor. When you hit Done, it splits the remaining gain needed to reach −12 dBFS speech peaks across two clean digital stages — and it caps that gain so the delivered noise floor stays below about −48 dBFS, refusing to buy loudness at the price of audible hiss. From there, auto-level keeps you on target as you drift, and the spectrum analyzer shows what's actually living in your floor.