What this is for
Microphone pages read the input in this tab. Relative level, clipping, and a local playback are the whole product.
A microphone test in a browser is a level, a recording, and a playback. It is not a studio shootout. The useful questions are small. Does the meter move when you talk? Does it move when you are quiet? Does it pin when you speak normally? Does the recording sound like the mic you think you selected? Those questions can be answered without uploading anything.
Allowing the microphone lets this tab read samples. The level meter is the root of the mean square of those samples, drawn as a bar. It is a relative number. It is not decibels SPL, because the browser does not know the sensitivity of your capsule or the gain in the interface. Do not publish the figure as a room measurement.
How to use the result
Clipping is flat tops. When samples hit the digital ceiling, the counter moves and the sound goes harsh in a way that no later volume change fully repairs. Lower the input gain in the system or on the interface. Moving away from the mic changes the tone of the room as well as the level. Gain is the cleaner fix.
Noise is what remains when you are quiet. Sample it on purpose, in the room you will actually use, with the fridge and the fan in their real states. A higher number means this input is hotter or the room is louder. It still is not an SPL reading. Compare two setups with the same page rather than comparing your number with a number from someone else’s interface.
Recording stays in the tab as a blob URL. Playback uses the audio element so you hear the file, not a second path that could hide a problem. Compare mode keeps two takes so you can move a mic, or switch from a webcam mic to a headset, and hear the difference without a project file. Close the tab and the takes are gone.
What it cannot claim
Webcam microphones sit far from the mouth and close to the fan. They will lose to a headset boom in a noisy room every time, and the meter will show room noise even when you think you are being quiet. Headset mics fail at the mute switch on the boom and at the plug more often than they fail in the browser. Tap the capsule. If the meter ignores the tap and jumps when you hit the desk, you are listening to a different mic.
A frequency-response curve is a lab product. Playing a sweep out of the same computer and recording it measures the speaker, the room, the mic, and the gain. That recording can still teach you that a capsule is dull or that it rattles, if you listen with that humility. It cannot be a published specification. The guide page says this because the temptation to screenshot a graph is strong.
When nothing moves, walk the stack. Site permission, operating-system input device, input level above zero, hardware mute, then another application holding the device in exclusive mode. If the meter on the troubleshooter moves, the browser can hear you and the other application is the remaining suspect. If it does not move, stay in the system sound panel. Reloading the website will not unmute a privacy switch.
Practical notes
A useful session is short. Open the meter, speak one sentence at the level you will record, and watch the bar. It should move and still leave a gap before the end. Record that sentence and play it back once. If the playback has a hum that the room does not, move off the laptop’s power brick or a USB port on a noisy hub. If the playback is clean and the other app is not, the other app has the wrong device. Two minutes of this beats a long thread about drivers.