Ten seconds, three answers
This online mic test tells you three things in ten seconds: which microphone your browser is using and its real sample rate, whether your level is too quiet, right or clipping, and how your voice sounds played back.
Your browser will ask for the microphone. The test then runs in two parts: three seconds while you stay quiet, which is where the background reading comes from, and seven seconds while you talk, which is what gets played back. It runs entirely inside this tab — there is no sign-in, and this page never sends your voice anywhere.
Ready.
quiet
This is the recording, held in this tab’s memory. It is never written to a file and never sent anywhere; closing or reloading the page destroys it.
When it does not work
Three switches stop a working microphone. Change the one that matches, reload, and test again.
Then the microphone is fine and the app is listening to another input. Choose the one this test named:
From each vendor’s help: Zoom, Discord, Microsoft Teams, Google Meet.
A site you blocked stays blocked until you change it. Set this site back to Allow, then reload:
From each vendor’s help: Chrome on a computer, Microsoft Edge, Safari on a Mac, iPhone, iPad, Chrome on Android.
With the system switch off, every browser and app gets an error or silence, this test included:
What it checks
Seven checks from one ten-second run, each measured from your own microphone in this tab.
| Check | How it is measured | What the page tells you |
|---|---|---|
| Which microphone | The input the browser actually opened | Its name, its real sample rate and its channel count, before anything is measured, and a picker when there is more than one |
| Whether sound arrives | The average level of the seven spoken seconds | Below −90 dBFS: no sound reached the page, and no verdict is given |
| Your level | The same average | Below −30 dBFS: very quiet, so move closer or add gain at the interface |
| Your loudest peak | The highest sample | Above −6 dBFS: running hot, with little room left for a raised voice |
| Clipping | Samples that reached full scale | Any at all: clipping, with the count |
| Background noise | The quietest four-tenths of a second in the three quiet seconds | −60 dBFS or lower: quiet enough to record in. Up to −45: fine for calls, marginal for recording. Above that: a loud background |
| How you sound | The seven spoken seconds, kept in this tab’s memory | Played back with their waveform, so you hear what the microphone sent |
What the readings mean
These two decide whether a recording is usable. Speech averaging between about −30 and −12 dBFS sits comfortably above the converter’s own noise; peaks around −6 dBFS leave room for a laugh or a raised line. Too quiet and every later stage that lifts your voice lifts the room with it. Too loud and the peaks are simply gone.
Measured during the three seconds you stayed quiet, as the quietest four-tenths of a second in them. This is the real noise floor of your room and your signal chain together. Under −60 dBFS is quiet enough to record spoken word in; down to about −45 dBFS nobody on a call will notice it; above that something is running that a recording will keep.
A sample at full scale is one the converter had no room left for, and nothing downstream puts it back — not a limiter, not a repair tool, not this site’s own fixer. One clipped sample is a warning; a few thousand is a take to record again with the gain lower.
What the browser is really running at, which is not always what the interface says: a device set to 48 kHz often arrives resampled. And one microphone should arrive as one channel — a stereo capture of a single microphone is two copies of the same thing taking twice the space.
Why the playback is part of the test
Every number above can be correct while the recording is still wrong: a rustling desk, a plosive on every p, a fan you stopped hearing an hour ago, a room with a ring in it. Those are things you hear and no meter reports. So the seven spoken seconds are kept and handed back to you with the waveform beside them.
Playing your own voice through speakers while the microphone is still open is how an echo test becomes an echo. This test deliberately switches off the browser’s echo cancellation, noise suppression and automatic gain, because all three change the thing being measured — so what you hear is what your microphone actually sent.
The room test takes longer silence and does the arithmetic against the level audiobook publishers require. The chapter checker measures a recording you have already made against ACX, Voices by INaudio, Spotify for Authors or Author’s Republic and corrects what is correctable, and the batch checker does a whole book at once. None of them uploads anything either.
Where the audio goes
Your voice never leaves this tab. The seven recorded seconds exist as numbers in the page’s own memory, are played back through Web Audio without ever becoming a file, and are discarded when you reload or close the page — and nothing on this page sends them anywhere. The test needs no sign-in, sets no cookie and stores no recording anywhere. The microphone light goes out the moment the ten seconds end, because the track is stopped in the same breath.
Straight answers
Press start and allow the microphone when the browser asks. Stay quiet for three seconds while the page measures your background, then talk for seven at the volume you would normally use. You get the device name and its real sample rate, your average and peak level in dBFS, the noise floor, a count of clipped samples, and the seven seconds played back so you can hear them.
If this test prints a level in dBFS, the microphone is working and audio is reaching your browser. If it says no sound reached the page, every sample was digital silence, which means a muted input, a device that is not running, or a different microphone selected than the one you are speaking into — not a quiet room.
Speech should average between about -30 and -12 dBFS. Below that the microphone is too far away, the input gain is too low, or the operating system is applying its own attenuation. Raise it at the interface or the system sound settings rather than in software afterwards, because software gain raises the room noise by exactly as much as it raises your voice.
This page names the four causes it can see: permission refused, no device reaching the browser at all, the device held open by another application such as a meeting or a recorder, and a device that opens but sends only silence. The troubleshooting steps on this page cover three switches behind most of the rest: the microphone picked in Zoom, Discord, Teams or Google Meet, a site the browser remembers as blocked, and the system microphone switch in Windows or macOS. A browser cannot reach past those into an operating system driver or a console headset, and this page does not pretend it can.
An echo is almost always your own speakers being picked up by the microphone. Listening back on headphones instead of speakers removes it. This test deliberately switches the browser's echo cancellation off so that what you hear on playback is what your microphone really sent, which is why an echo that meeting software hides is audible here.
It is a sample that reached full scale, so the converter had no room left for the real value and the peak was cut flat. Nothing downstream restores it — not a limiter, not a repair tool. A handful of clipped samples is a warning; a few thousand means recording that take again with the input gain lower.
Yes. Any input your browser exposes can be tested, and when more than one is available this page offers a picker so you can measure the headset rather than whichever device the browser chose. A Bluetooth headset often reports a much lower sample rate than a wired one; the sample rate reading is where you see that.