How the test works
Press Start microphone and the browser asks for permission; nothing happens before you agree. Once the stream opens, the dark panel becomes an instrument: the trace shows the waveform coming from the microphone, the bar along the bottom shows the current level with a thin marker that holds the recent peak, the number in the corner is the peak of the last frame in decibels, and the CLIP lamp lights when a peak reaches the top of the scale. The list above the buttons lets you switch between microphones; switching mid-run ends the current run and starts a new one on the chosen device, so each result belongs to one microphone.
Record 10 s captures a short clip from the same stream and stops on its own after ten seconds. Play back plays it through your speakers or headphones. The clip lives in the tab's memory as a temporary object and is discarded when a new clip is recorded or the page is closed. Nothing is sent anywhere; the "Local by design" note on the side describes exactly what the page touches.
The level is measured in decibels relative to full scale, so 0 dB is the loudest value a digital signal can carry and quieter sounds are negative. The scale on the panel runs from −60 dB, which is effectively silence for a consumer microphone, to 0 dB. The measurement runs with the browser's processing switched off (no echo cancellation, no noise suppression, no automatic gain), so the meter reflects the microphone and the operating system's input volume, not a filtered version.
What the result means
When you press Stop, the card shows the loudest peak of the run in dB and, underneath, a one-word reading, the average level and the length of the run. The reading is a rule of thumb for speech:
- Good level. Peaks landed between roughly −30 and −0.5 dB without repeated clipping. Video calls and recording apps will work with this input as it is.
- Too quiet. Nothing peaked above −30 dB. Either you were far from the microphone, the input volume in the system is low, or the browser picked a microphone that does not point at you. Apps will compensate by amplifying, and noise comes up with the voice.
- Clipping. More than a handful of frames hit the top of the scale. The input gain is too high or the microphone is too close; the distortion this causes cannot be removed afterwards.
The waveform is also diagnostic. A steady thick band while you are silent is a noise floor: a fan, a hum from a cheap USB hub, or interference on a long analogue cable. Regular spikes with nothing said point to a loose connector. A waveform that only shows one side of the centre line is a faulty microphone or a damaged cable.
What to do if something is wrong
Start with the picker: if the meter is flat, try every device in the list, because the browser's default is often not the microphone you expect. Then open the sound settings of the operating system, raise the input volume of the chosen microphone to about 75 percent and check that no mute is engaged; many laptops have a microphone mute key with its own LED and some headsets have a switch on the cable. On Windows, check that the browser has microphone access under Privacy settings; on macOS, under Privacy & Security; on phones, in the site permissions of the browser. Then reload this page and run the test again.
If the level is fine but the recording sounds thin or muffled, move closer, take the microphone out of a case or a sleeve, and clean the small opening on a phone or laptop; a layer of dust or a screen protector over the microphone hole can cost 10 dB. If the recording has a hum, plug the computer and any audio gear into the same power strip and use a shorter cable.
When to return or claim the warranty
A new microphone or headset that produces no signal on any device, shows one-sided waveforms, or crackles when the cable is moved is defective and should be exchanged within the seller's return window. Record a clip here with the fault audible, note the device name shown in the result, and bring both to the claim; the technician can reproduce the test in a minute. A low noise floor and a modest level from a built-in laptop microphone are normal for the size of the part and are not grounds for a claim; a dedicated USB microphone will beat any built-in one by a wide margin.
Frequently asked questions
Is my voice uploaded anywhere?
No. The browser hands the microphone stream to this page, the page draws the waveform and measures the level with the Web Audio API, and the optional recording is kept as a temporary file inside the tab. There is no server involved; you can open the network panel of the browser while the test runs and see that nothing is sent. Closing the page discards everything.
The meter does not move at all. Is the microphone dead?
First check the device picker above the buttons: laptops often have two or three inputs and the browser may have picked a headset that is not plugged in or a monitor's microphone over HDMI. Then check the input volume in the operating system and whether the microphone is muted by a key or a physical switch. If a different microphone works on the same page, the silent one or its cable is the problem.
What level should I aim for?
For speech, peaks between −12 and −6 dB with the average around −20 dB is a comfortable recording level: loud enough to be clear, with room before the top. Peaks that stay below −30 dB mean the microphone is too far away or the input gain is too low, and every app will have to amplify noise along with your voice. Peaks that hit 0 dB light the CLIP lamp and distort.
Why does the recording sound different from the level I saw?
The meter measures the raw input with the browser's echo cancellation, noise suppression and automatic gain turned off, so you see what the microphone actually delivers. Some browsers record through a compressed codec, which softens transients, and your speakers or headphones add their own colour on playback. Compare recordings made with two microphones on this page rather than comparing the recording to the meter.
Can I use this for a Bluetooth headset?
Yes, but pick the headset in the device list and be aware that many Bluetooth headsets switch to a low-quality telephone profile as soon as their microphone is in use, which makes music playback sound flat while the test runs. The level reading is still valid. A short delay between speaking and seeing the waveform move is normal on Bluetooth.