Device test · Audio in
Microphone Test
Press Start, allow mic access, then speak and watch the meter.
Processing state is reported once the microphone is open.
Sample rate is reported once the microphone is open.
RMS and Peak are dBFS levels, relative to digital full scale. Peak and Clipping hold since you started or last reset; Reset peak & clip clears both.
Keep this tab visible. Stay quiet for the first four-second window, then speak normally without pausing for the second. The page compares the median level of each window and reads levels only, so no audio is recorded. Stop measuring ends the probe; Stop also closes the microphone.
Verdict: is the mic too quiet?
NOT RUN
—
Run Measure noise floor, then speech, to get a verdict on the speech level.
Thresholds, on the speech median: −30 to −12 dBFS with S/N of 10 dB or more is PASS. Quieter than −30 dBFS is CHECK SETTINGS (too quiet: raise the gain, then re-measure). Still quieter than −30 dBFS after the gain fix is LIKELY FAULT. Louder than −12 dBFS, or S/N under 10 dB, is CHECK SETTINGS. See the methodology.
Raise the input gain on your platform
- Settings → System → Sound → Input: pick this microphone and set Input volume to 80–100.
- Still low? Its Properties → Levels → Microphone Boost, +10 dB at a time.
- System Settings → Sound → Input: pick this microphone and drag Input volume about three quarters right.
- USB or interface mic: raise the gain knob on the device too.
- Quick Settings → the arrow beside volume → Input: pick this microphone and drag its slider up.
- Settings → Sound → Input: pick this microphone and raise Input volume to about 100% (or pavucontrol → Input Devices).
- There is no gain slider: remove the case, clear the mic holes and disconnect Bluetooth headsets.
- There is no gain slider: remove the case, clear the mic openings and disconnect Bluetooth headsets.
- Open your system's sound settings, pick this microphone as the input and raise its input volume most of the way.
Leave this tab open while you change it. The re-measure reads levels only: nothing is recorded or uploaded.
Still low after the fix? Speak 10–20 cm away, check any headset mute switch, then try another port, cable or device. A mic that stays low everywhere is a warranty case.
Record 5s and play back records for five seconds, then plays the clip automatically, and Play again replays it. The clip stays in memory on this device and is never uploaded. Stop, or leaving the page, discards it.
Playback uses your speakers or headphones. Turn the volume down before recording: the clip plays automatically when recording ends and may be unexpectedly loud, especially through headphones. Press Stop if it is uncomfortable. The microphone stays open during playback and may pick up sound from your speakers.
Copy a local diagnostic summary
The summary lists the readouts from your latest test, with no audio, microphone names or device IDs. The page copies it only when you press Copy diagnostic summary and sends the text nowhere. Your browser and operating system manage the clipboard.
Nothing has been copied.
Next steps: For a quiet or missing signal, try the microphone troubleshooting guide. To check the camera too, run the combined camera and microphone test.
How this test works
Once you allow access, the page reads the microphone stream your browser delivers. Input shows how that request went, and Signal separately shows whether audio is arriving or the browser has marked the track as muted. On each animation frame a Web Audio AnalyserNode supplies the latest 2048 samples, and the meter, waveform and numbers all come from that one window. From that window the page calculates the root-mean-square level shown as RMS. Peak holds the largest sample seen since the test started or you last pressed Reset peak & clip, and Clipping counts inspected windows that contain a sample at or above −0.1 dBFS. Consecutive reads can overlap or skip samples, which is why clipping is counted in windows. Sample rate is the rate of the Web Audio graph reading the stream, which can differ from the hardware rate because of resampling; when the track reports its own rate, the note below the waveform compares the two. Channels and the processing note come from the settings the track reports, and anything the browser leaves out reads not reported.
For the noise-floor probe, stay quiet for four seconds, then speak without pausing for four seconds. The page takes the median RMS level of each window and subtracts the quiet median from the speech median to give S/N in dB. A median keeps a single cough or door slam from setting the result, though any steady sound in the room still counts toward both windows. A gain that stays the same across both windows cancels out of the difference. Each window needs at least 80 readings at 20 per second; a slower or shorter set is withheld, and so is the ratio when the quiet window has no measurable signal. The probe reads levels and records no audio. The recording button works separately: it uses MediaRecorder to hold a five-second clip in page memory and play it on this device, and browsers without MediaRecorder cannot record, though the meter still works.
The page has no calibrated reference, so it cannot measure loudness in the room in dB SPL. Ask for unprocessed audio requests echo cancellation, noise suppression and auto gain control off, but the browser can keep some of them on, so read the processing note for what it reports. Automatic gain control lifts quiet levels and compresses the S/N difference, and changing gain or processing between the two windows changes the result. Local playback leaves out a call app's processing, codec and network path, and the other person's speakers.