Notes on mic tests: what the check is really measuring, and why one reading is not enough

The question behind the question

Nearly everyone who opens a mic test online is asking a yes or no question: is my microphone working? The honest answer, almost every time, is that it is. Microphones fail, but they fail rarely. What fails far more often is the setup around them: the distance, the gain, the room, the cable, the charger. This is the first idea worth holding on to, because it changes what you do with the result. A test that tells you the microphone is alive has told you almost nothing. A test that tells you where the voice sits relative to the room, and where the loudest syllables sit relative to the ceiling of the recorder, has told you what to change.

Two ways to get it wrong

Signals fail in two opposite directions, and they feel quite different in the ear.

The first is too little signal. The voice arrives faint, the room arrives with it, and the recording sounds thin and hissy. Nothing is broken; the microphone is simply too far from the mouth, or the input volume is too low, so the room's own noise ends up as loud as the words. A recording like this can be cleaned up a little afterwards, but the cleanest version of it is made at the start.

The second is too much signal. The input volume is turned up to make the voice strong, the loudest syllables run past the top of the scale, and the take crackles. This one is worse, because the damage is permanent. In a digital recording, 0 dBFS is a hard ceiling. Samples cannot be stored above it. When a peak pushes past it, the waveform is sliced flat and the listener hears it as a click or a crackle that no amount of processing will remove.

Headroom is the name for the margin you leave below that ceiling. Keep the loudest syllables a comfortable distance under 0 dBFS and transient peaks, the sharp consonants and sudden emphases, stay intact. Set the level so an average looks healthy and the peaks will find the ceiling on the one word that matters.

So the real question a mic test answers is not yes or no. It is two numbers: the gap between the voice and the room, and the gap between the peaks and the ceiling.

Reading the instruments

A live level meter shows loudness moment to moment, and the temptation is to watch the average. Watch the peaks instead. The average can look perfect while a single syllable clips.

The waveform is where clipping becomes visible rather than audible. Normal speech draws a ragged, spiky shape with clear peaks above and below the centre line. A clipped signal draws flat tops, as if the peaks had been sliced off with a straight edge. Once you know the shape, you can spot a bad take in a second.

The spectrum and the spectrogram show what the microphone actually carries. A full-range result in a good setup is audible up to roughly 16 kHz. A dull ceiling on the display, with little energy in the upper frequencies, usually means the microphone is far away or the source is off-axis rather than that the microphone is poor.

And the room deserves its own number. Measure the noise floor in dBFS with nobody speaking, then measure the voice, and the gap between them is what makes a recording sound clean or hissy. A quiet room is not a luxury; it is half the result.

The worked examples on a test page make these numbers concrete. A microphone a hand's width away in a quiet room reads around −18 dBFS peak with zero clips and a voice sitting well above the room. The same microphone across the room drops to roughly −44 dBFS peak, and the gap to the room shrinks to about 17 dB. Turn the input volume all the way up instead and the peaks climb to full scale, 0 dBFS, with 18 clips on the display. Three setups, one microphone, three completely different outcomes.

Hum is not the same problem as noise

Mains hum is a steady low tone from electrical interference. It sits at the frequency of the supply and its harmonics, so it sounds like a constant note rather than a wash of sound. It is easy to mistake for general noise and easy to fix once you know it is there, provided you look for it separately.

The reason it deserves its own check is that it enters through a different door than level problems. Hum usually arrives through cables, through grounding, or through a laptop charger, not through the microphone itself. The fixes are physical: move or unplug the charger, reseat the cable at both ends, try another input. Raising or lowering the gain does nothing for it.

A fan or a fridge raises the noise floor but does not produce a steady tone. That distinction matters, because the response is different in each case. A room problem is solved by moving closer, closing a window, or switching off a fan. A hum problem is solved by chasing the electrical path.

Why a single reading misleads

Several familiar checks each answer a narrower question than the one you are asking.

A tone generator confirms that the microphone transmits sound. It says nothing about whether the voice will sit above the room.

A playback lets you judge by ear, which is the right final authority, but by the time you hear the crackle or the hiss the take is already ruined. Judgement after the fact is expensive.

A single level readout, meanwhile, can look fine on average while the loudest syllables clip. And a clipped take and a clean take can show similar averages, because clipping affects peaks, not the mean.

Checking level, noise, clipping, range and hum together is what catches the problem any one indicator would hide. A loud microphone can also be clipping. A quiet one can be perfectly clean. Only the combination tells you which situation you are in.

Where people go wrong

Turning the input gain all the way up for a stronger signal is the most common mistake, and the most damaging. Gain raises the room along with the voice, and pushes the peaks into the ceiling.

Testing in silence and calling it good is the second. A room sounds different with a voice in it, because echo arrives with the speech; an empty-room test flatters the space.

The third is treating a successful sound pickup as a pass and skipping the noise floor. The microphone picked something up. That was never in doubt.

The fourth is measuring at a different distance from the one you will record at. A test taken across the desk says nothing about a take made close to the grille.

The fifth is fixing problems by ear after a full take rather than with a short check before it. Fifteen seconds of checking is cheaper than twenty minutes of recording that has to be thrown away.

Putting it into practice

Run a short scored check before a session rather than after. A report that names each finding and gives one plain fix per finding lets you change one thing at a time, which is the only way to know what helped.

Read the score components rather than the total. A high total can hide a weak range reading, and range is exactly the component that quietly decides whether a voice sounds close and present or thin and distant.

Fix in order of cause: distance first, then input gain, then the room, then cables and power if the finding is hum. Working in that order avoids the loop of adjusting gain to compensate for a problem that distance created.

The principle holds wherever you record, whether a voiceover, a podcast or a lesson narration. For anyone working entirely in a browser, a tool such as Audio AI Studio keeps the take and the next step in the same place: recording, then transcription with speaker labels and timestamps, or a voice workflow, without installing anything. Its mic test is free and runs in the browser, reports level, clipping, room noise, frequency range and hum with a fix for each, and shows the waveform and spectrum while you speak.

Re-run the check whenever the room or the device changes. The microphone is a fixed object. It is the setup that drifts, and the check is how you notice.