Check a microphone before a call or a recording. The meter reads the signal arriving from the device in dBFS, where 0 is full scale and every real level is negative.
dBFS, not decibels of loudness
The number here describes the digital signal, not the sound in the room. Full scale is the largest value the format can store, so the scale runs downward: -6 dBFS is half of full scale in amplitude, -20 dBFS is a tenth.
That makes it a headroom reading. A voice peaking at -6 dBFS has 6 dB left before the converter runs out, which one raised sentence will use up. A voice peaking at -30 dBFS is using a fraction of the available resolution and will need lifting later, along with whatever the room and the preamp contributed.
Peak, RMS, and the hold
Peak is the highest sample in the last frame, and it jumps with every consonant. RMS is the energy over that frame, which tracks how loud the signal seems rather than how tall it is. The gap between them is roughly the crest factor: speech runs 12 to 18 dB, a compressed broadcast voice much less.
The hold keeps the loudest peak since the test started. Read the hold to set a level, and the live peak to see whether anything is arriving at all.
Clipping and the noise floor
Clipping is counted, not estimated. Any sample at or within a whisker of full scale is tallied, so a count above zero means the signal was flattened at the top, not that it came close.
The noise floor is the quietest level measured since the test started, which means it only means something after a few seconds of nothing but the room. A laptop microphone in a quiet room lands near -60 dBFS. A fan, a hard drive, or a laptop under load sits 10 to 15 dB above that, which is the difference between a recording that cleans up and one that does not.
Stereo devices and dead channels
Opening the device in stereo meters both channels separately. A stereo interface with a cable in only one input, a headset reported as stereo but wired to one side, or a channel lost in a driver all look the same in a mono meter: a level that reads slightly low. Two meters make it obvious.
Voice processing
Echo cancellation, noise suppression, and automatic gain control are call features. They are applied by the browser to the microphone track before anything else sees it.
They are worth hearing deliberately. Automatic gain control lifts the signal during pauses, which is why a room sounds like it swells between sentences on a call. Noise suppression removes steady sound, and it takes the tail of a sustained note or a soft consonant with it. Both are useful in a meeting and unwanted in a recording, so the switch changes the constraints rather than hiding them.
Where a browser test stops
The browser reports the stream it was given, after the operating system’s own input processing. If the level is wrong at the system level, or a driver resamples, that is upstream of anything measurable here.
Latency is also invisible to a meter. Monitoring the input tells you whether the path works, not how many milliseconds it takes to get back.