Before you start
Pick the input you actually intend to use. Laptops routinely expose an internal array, a headset, a webcam microphone and a virtual device from conferencing software, and the browser's default is often not the one on your head. Selecting explicitly removes the single most common cause of "no sound".
Close other recorders. Conferencing clients, streaming software and voice assistants can hold exclusive access or apply their own processing chain, which changes the signal before any page sees it.
Describe the fault in one line before testing: no level at all, level that overloads, gaps in otherwise normal speech, or a recording that sounds wrong on playback. The checks below map onto those four.
Run the check
Open the microphone test, allow access, and speak a steady phrase at the distance you actually use. Reading a full sentence works better than saying "test" once, because level analysis needs a few seconds of continuous signal to produce a stable peak and RMS pair.
Watch peak and RMS together. Peak is the loudest individual sample in the window and is what warns you about clipping; RMS is averaged energy and is what corresponds to perceived loudness. Both are shown in dBFS, where 0 is the digital ceiling and everything usable sits below it.
Look for clipping and silence separately. Three or more consecutive samples at 0.99 absolute value are counted as clipping, and windows of at least 120 ms at exact digital silence are counted as gaps. Those two counters answer different questions: one says the input is too hot, the other says something is cutting it.
Finally, record five seconds locally and play it back. Meters describe amplitude, not intelligibility, and a recording immediately exposes hum, room echo, keyboard noise and processing artefacts that no level display will show you.
Read your result
For a healthy speaking voice, peak should sit comfortably below 0 dBFS with a visible gap between peak and RMS. When peak and RMS converge and the clipping counter rises, the input is overloaded: reduce gain in the system input settings or increase the distance to the microphone.
If the level never moves at all, the problem is selection, permission or connection rather than gain. Confirm the selected device, confirm the browser has permission, and check that a hardware mute switch on a headset or interface is not engaged — physical mute switches account for a surprising share of dead inputs.
Gaps of exact digital silence inside continuous speech point at processing rather than hardware. Aggressive noise suppression, an automatic gate or an unstable connection all produce them. Rerun with the calling application's noise suppression disabled and see whether the gaps follow the processing.
Measurement limits
Browsers may apply automatic gain control, echo cancellation and noise suppression before samples reach the analyser, which means the numbers describe the processed stream rather than the raw microphone. That is a feature for calls and a limitation for measurement, and it is why two browsers can report different levels from the same microphone.
Nothing here can inspect an analogue preamp, a cable, a phantom-power supply or a connector. The page starts where digital samples arrive and cannot look further upstream.
Absolute dBFS values are not comparable between devices. A figure that looks low on one interface can be perfectly healthy on another, because gain staging differs. Compare a device against itself across runs, not against someone else's numbers.
What to do next
A local recording that sounds correct narrows the problem to whichever application sounded wrong: check its input selection, its gain and its noise suppression settings before touching hardware.
No level across every application, after confirming selection and permission, is when hardware becomes the leading explanation. Try the same microphone on another machine, or another microphone on this one, to decide which side is at fault.
If speech is audible but unpleasant rather than absent, the fix is usually environmental: move closer, speak across the microphone rather than into it, and reduce reflective surfaces. These change the recording far more than any setting will.
Symptom table
| Symptom | Likely layer | Next action |
|---|---|---|
| Level never moves | Wrong input, blocked permission or inactive track | Select the input and request access again |
| Strong input clips | Gain or distance is too high | Lower input gain or increase distance |
| Playback has gaps | Digital silence or processing interruption | Compare a local recording with processing changed |
Common questions
What level should my microphone show when I speak normally?
Aim for peaks clearly below 0 dBFS with headroom left over, and an RMS noticeably lower than the peak. If the two values sit on top of each other you are overloading the input, regardless of how loud it sounds.
Why does my microphone sound fine here but bad in calls?
Conferencing applications apply their own compression, noise suppression and bandwidth limits after the point this test measures. A clean local recording combined with poor call audio points squarely at the application or the network.
Can a browser mic test damage or change my microphone?
No. The page only reads samples the browser provides and, optionally, keeps a short recording in memory. It changes no system setting and writes nothing to your device.