🎤Mic test

✓ Tested & verified Updated: Aug 11, 2026 How the test works

Mic test

The scale is negative: 0 dBFS is the ceiling where the signal clips, not silence. Nothing is recorded and nothing is uploaded.

Loudest peak

—dBFS

Press start and talk at your normal distance for about five seconds — the meter reads the live stream and forgets it.

Average level—
Background noise—
Clipped frames—
Sample rate—

Measure the raw signal — switch off the browser's noise suppression, echo cancellation and auto gain

Try also

Is my microphone working, and how loud is it?

The mic test reads your microphone’s peak level in dBFS, so you get a number instead of a green bar. dBFS is a negative scale: 0 dBFS is the digital ceiling, and every real signal sits below it. Speech should peak between −12 and −6 dBFS. Below −20 dBFS you are quiet enough that other people will strain to hear you, and peaks touching 0 dBFS are clipping — the loudest syllables get cut flat, and turning the volume up at the other end does not repair them.

The test asks for microphone permission, opens the stream inside the page and updates the peak live, so you can talk, watch the number and adjust the input gain until it lands in the right band. Nothing is recorded, nothing is uploaded, and the stream closes when you stop the test or leave the page.

A reading is only useful once you know which band it belongs in. The scale runs downward from 0 dBFS at the digital ceiling, so every number below it is negative and −6 dBFS is louder than −20 dBFS, not quieter.

Peak while speaking	What it means	What to change
0 dBFS	Clipping — the loudest syllables are cut flat	Lower the input gain until peaks stay under −6 dBFS
−6 to −1 dBFS	Hot, but intact	Leave more headroom for laughter and raised voices
−12 to −6 dBFS	The band you want for speech	Nothing
−20 to −12 dBFS	Usable but quiet	Move to 15–20 cm, or raise the gain one step
−40 to −20 dBFS	Too quiet; noise rises with every step of gain	Confirm the right device is selected, then raise the level
Below −50 dBFS while speaking	Effectively silence	Wrong input, a hardware mute switch, or exclusive mode

Permission is granted per device, so a browser can hold the microphone and still be blocked from the camera — the camera test settles which of the two it is. For the other half of a call, the output side is covered by the speaker test and the headphone test.

Frequently asked questions

My microphone works in this test but not in Discord or Teams

That is almost always device selection or a per-app permission, not the microphone. The browser uses whatever the system calls the default input, while Discord, Teams and OBS each keep their own input setting — often still pointing at a device that was unplugged months ago. Check the app’s audio settings first, then Windows Settings > Privacy & security > Microphone, where desktop applications can be blocked individually while browsers stay allowed.

Why is my microphone so quiet?

Put a number on it before changing anything: if your speaking peaks sit below −20 dBFS, the level is genuinely low. Raise the input level in the sound settings for that device, and reach for the microphone boost only after that runs out, because boost lifts room noise at exactly the same rate as your voice. Distance does more than either — 15 to 20 cm from a headset or condenser mic is worth several decibels over arm’s length.

What is exclusive mode and why does it take my microphone?

Windows lets one application take exclusive control of an audio device, which locks every other program out until it lets go. It is a per-device checkbox: Sound control panel > Recording > your microphone > Properties > Advanced > ‘Allow applications to take exclusive control of this device’. When a call app or an audio workstation grabs the input and leaves it held, everything else gets silence rather than an error message, which is why the symptom looks like dead hardware.

I blocked the permission prompt by mistake. How do I get it back?

The block is stored per site, so the prompt will not come back on its own. In Chrome and Edge, click the icon at the left of the address bar, open Site settings and set Microphone to Allow, then reload. Firefox shows a crossed-out microphone in the address bar with a Remove option, and Safari keeps it under Settings > Websites > Microphone. A blocked site looks identical to a broken microphone, so it is worth ruling out first.

Which microphone does the test use if I have several?

The one the operating system currently calls the default input, not the one you used last. Laptops with a built-in array plus a plugged-in headset regularly default to the built-in one, which is exactly why a headset can sound distant and hollow in calls. Change the default in the system sound settings and reload the page; browsers also keep a per-site device choice that can override it.

Does the test record or upload my voice?

No. The audio stays in the page: samples travel from the microphone into the browser’s audio engine, get turned into a number, and are discarded frame by frame. Nothing is written to disk, nothing is sent to a server, and no request leaves while the meter runs. Stopping the test or closing the tab releases the device, and the browser’s recording indicator goes out with it.

How the test works

Peak level is read from the raw time-domain samples of the microphone stream rather than from a smoothed visual bar. The Web Audio API delivers those samples in the range −1 to 1; the meter takes the largest absolute value in a short rolling window and converts it with 20 × log10(peak), which is what makes the unit decibels relative to full scale and every reading zero or negative. A sample at 1.0 is 0 dBFS, half of full scale is about −6 dBFS, and a tenth of it is −20 dBFS. The wording next to the number maps it onto the bands in the table above. The browser’s own processing — echo cancellation, noise suppression and automatic gain control — is applied before the samples reach the page, so the reading describes what a call or a recording application would receive from you, which is the number that matters. Everything runs locally: no audio is recorded, stored or uploaded.

Worked example

With loudest sample of a spoken sentence 0.35 of full scale, quietest 200 ms of the room 0.008, this page works out a peak of −9.1 dBFS. This is the level broadcast desks aim for: peaks between −12 and −6 dBFS are loud without ever touching the ceiling. Leave the gain where it is.

Peak level−9.1 dBFS
Room floor−41.9 dBFS
Voice above the room33 dB

Sources:

Why a level meter beats a green tick

A green light answers the wrong question. Almost every microphone that is plugged in registers sound. The failures people actually have are levels too low to hear, peaks that clip, or a stream held by another program — and none of those show up as a yes or a no.

Headroom is the part that gets forgotten. A level that looks ideal while you talk normally will clip the moment you laugh or raise your voice, which is why the target band sits below the ceiling on purpose rather than at it.

Processing changes what the number means. Browsers apply echo cancellation, noise suppression and automatic gain control to a microphone stream by default, so a quiet input can be dragged upward and a loud room pushed down before the meter ever sees it.