See whether your microphone is picking up sound, and how it sounds, on one screen
The microphone test shows right away whether sound from your laptop's built-in mic, a headset, a USB mic or a phone mic is reaching the browser, then reads that sound four ways. Speech is the level of the louder part of the last 10 seconds, and background noise is the level of what sits underneath when you are not talking. Gap to noise is the distance between the two, so it tells you whether your voice is being buried, and peak shows whether the loudest moment reached the digital maximum and was clipped.
The upper area scrolls a waveform and a frequency map (spectrogram) of the last 10 seconds together. The waveform shows how loud the sound is, and the frequency map uses color to show at which pitch the sound sits, from low to high. A 10-second recording keeps the microphone, the voice processing setting and the readings used for it, so you can listen to different mics side by side. Opening the page does not turn the mic on, and the sound is handled only inside this browser and never uploaded.
When to check your microphone ahead of time
Microphone problems often come to light only after someone says "I can't hear you." Checking permissions, device choice, mute and the Bluetooth connection one by one at that point costs the first minutes of a meeting or class. In these cases it is safer to check a few minutes early.
- Before a video meeting or online interview: see that sound is coming in, that the wrong mic has not been picked, and that a fan or air conditioner is not drowning out your voice. For something you cannot redo, like an interview, go as far as recording a take and listening to it yourself.
- When you teach or lecture online: if you talk for a long time, a small change in the distance between mouth and mic changes the level a lot. Speak in your usual posture and check that the speech box lands on "Good level".
- When you buy a new headset or USB mic: check that the device appears in the list and that the range of incoming sound is wide enough, using the frequency map. If you record the same sentence on your old mic, you can hear for yourself what improved.
- Before game voice chat or streaming: use the peak and the clipping indicator to see whether laughing or shouting gets clipped. Clipped sound cannot be restored by lowering the volume afterward.
- When you use Bluetooth earbuds as the mic: if music sounds fine but call quality is bad, the connection type may be the cause. Looking at whether the top of the frequency map is empty helps narrow it down.
The order to follow the first time
- Press Start mic and choose "Allow" in the microphone prompt the browser shows. If you allowed it before, the mic turns on without a prompt.
- Say one sentence from the distance you normally use on calls. If the waveform grows and shrinks with your words and horizontal stripes (the harmonics of your voice) appear on the frequency map, sound is coming in.
- Look at the reading boxes. Check that speech shows "Good level", that gap to noise shows "Clear", and that the peak box shows no clipping. The verdict line tells you what to look at first in the current state.
- If you have several mics, switch in the Microphone list in the settings row below. If you switch while it is on, it reconnects right away.
- Press Record 10 s, say the same sentence, then press the play button to listen. The mic is turned off during playback.
- To compare with what you hear in a meeting app, switch Voice processing to "On" and record once more. Listening to the two takes side by side shows how much noise suppression changes your voice.
The four reading boxes and how to read them
All four boxes are calculated from 500 level measurements, one every 20 milliseconds (0.02 s), over the last 10 seconds. The unit is dBFS: the largest level that can be stored digitally is 0, and anything smaller is a negative number. The ranges below are bands this tool chose to make readings easy to follow. They are not a microphone quality score or a standard.
| Box | How it is calculated | Reading ranges |
|---|---|---|
| Speech | The level where the loudest 10% of the last 10 seconds begins (L10). If the gap to noise is under 10 dB, no speech is considered found and the box shows "—". | Below −50.0: very quiet. −50.0 up to −35.0: on the quiet side. −35.0 up to −10.0: good level. −10.0 or above: loud |
| Background noise | The level exceeded for 90% of the last 10 seconds (L90). It is close to the level of the quiet moments between words. | Below −60.0: quiet. −60.0 up to −45.0: moderate. −45.0 or above: loud. If every sample is 0: silent |
| Gap to noise | Speech, rounded to one decimal place, minus background noise | 30.0 or more: clear. 15.0 up to 30.0: noise mixed in. Under 15.0: noisy |
| Peak | The level of the single loudest sample in the last 10 seconds. The line below shows the headroom left to 0 dBFS, or the number of 20 ms windows in which a sample's absolute value was 0.999 or higher (clipping) | Even one clip triggers a warning. The more headroom, the more room for loud sounds |
The 15 dB boundary for the gap to noise follows the American Speech-Language-Hearing Association (ASHA) technical report on classroom acoustics, which recommends that speech be at least 15 dB louder than background noise. That standard concerns actual sound pressure in classrooms, so it cannot be carried over to a microphone verdict as is, but it is a reasonable way to gauge the gap at which speech stays easy to understand in noise.
Examples of a 10-second reading
Even when the same person speaks into the same mic, the readings change with the surrounding sound and the distance. The table below was calculated by filling 500 windows of 20 milliseconds with simple levels. Because speech is defined as the louder 10%, also note that speaking for less than 1 second out of 10 means no speech is found.
| Situation | Speech | Background noise | Gap to noise | Verdict |
|---|---|---|---|---|
| Quiet room, speaking at −20 dBFS for 3 seconds | −20.0 | −62.0 | 42.0 | Clear |
| Fan noise at −38 dBFS, speaking at −20 dBFS for 3 seconds | −20.0 | −38.0 | 18.0 | Noise mixed in |
| Mic too far away, speech at −46 dBFS | −46.0 | −62.0 | 16.0 | Speech on the quiet side |
| Not speaking (only −62 dBFS) | — | −62.0 | — | No speech found |
| Speaking for only 0.5 seconds | — | −62.0 | — | No speech found |
How to read the frequency map (spectrogram)
On the frequency map, the horizontal axis is time (10 seconds ago on the left, now on the right) and the vertical axis is pitch. Low sounds are at the bottom and higher sounds are further up, drawn on a logarithmic scale that is close to how the human ear perceives pitch. Dark areas have almost no sound, and the color runs from orange to bright cream as the sound at that pitch gets louder.
A voice appears as several horizontal stripes. The vocal cords vibrate at a fundamental frequency and at whole-number multiples of it (harmonics) at the same time, and the lowest stripe is the fundamental. For adults the fundamental is roughly 90 to 155 Hz for men and 165 to 255 Hz for women, and when you raise or lower your pitch at the end of a phrase the stripes bend with it. Consonants such as "s", "sh" and "ch" show up without stripes, as a hazy patch of noise above 4 kHz.
| What you see | Usual cause | What to do |
|---|---|---|
| Curved horizontal stripes that appear only while you speak | A normal voice (fundamental and harmonics) | Just check the level and the gap to noise in the reading boxes |
| Thin horizontal lines that stay near 50 Hz and 100 Hz, or 60 Hz and 120 Hz | Hum from the mains supply. The mains frequency depends on the region (for example 60 Hz in the United States and 50 Hz in the United Kingdom), and lines appear at its multiples. | Unplug the laptop charger and look again, plug into the computer itself instead of a USB hub, move the mic cable away from power cables |
| A wide bright band that continues even when you are silent | Broadband noise from a fan, air conditioner or computer cooling fan | Turn the source off or move away from it, and bring the mic closer to your mouth |
| Short vertical lines that pop up | Keyboard or mouse clicks, tapping on the desk, the mic cable rubbing | Move the mic away from the keyboard or use a stand |
| Empty above about 8 kHz | Wideband voice on a Bluetooth call connection (16 kHz sample rate), or processing in the mic or operating system | Record the same sentence on a wired mic or another connection and compare |
| Empty above about 4 kHz | Narrowband voice on a Bluetooth call connection (8 kHz sample rate) | Same as above. This is why "s" sounds come out muffled. |
| The whole screen is dark and the waveform is flat | Muted, a different mic selected, or digital silence where every sample is 0 | See "Messages, their causes and what to do" below |
The telephone voice band is 300 to 3,400 Hz, and wideband speech coding (ITU-T G.722) covers 50 to 7,000 Hz. The sample rate must be twice the highest sound it can capture, so a signal read at 8 kHz holds sound only up to 4 kHz, and one read at 16 kHz only up to 8 kHz. The "captures up to … kHz" next to the sample rate in the settings row means this limit. Even so, that sample rate is what the browser reports, so for sound whose bandwidth was already reduced inside a Bluetooth device, the frequency map is the more accurate place to check.
dBFS, RMS and peak calculations
For each 20-millisecond window, this tool squares the samples, averages them and converts the result with 10 × log10(mean square). The square root of the mean square is the RMS, so the same value can also be written as 20 × log10(RMS). The peak is the single sample with the largest absolute value in the window, converted with 20 × log10(|sample|). If every sample is 0, the result would be negative infinity, so it is shown as "Silent".
| RMS | dBFS | Meaning |
|---|---|---|
| 1.0 | 0.0 | A sound that stays at the digital maximum |
| 0.707 | −3.0 | A sine wave whose amplitude is the maximum |
| 0.5 | −6.0 | Half the RMS is 6 dB lower |
| 0.1 | −20.0 | One tenth of the RMS is 20 dB lower |
| 0.01 | −40.0 | Quiet speech |
| 0.001 | −60.0 | The bottom end of the meter |
A sine wave at full amplitude comes out as −3.0 dBFS in this calculation. Programs that follow the convention of setting a sine wave to 0 dBFS (AES17) will differ by 3 dB, so if another recording program's meter shows slightly different numbers, that does not mean either one is wrong. dBFS is the level of a digital signal, so it cannot be converted to the decibels of room noise (dB SPL), and the same voice gives different numbers depending on the microphone's sensitivity and the system input volume. To see how loud a room is, use the decibel meter.
Compare with voice processing off and on
Browsers and meeting apps can apply three kinds of processing to make a voice more pleasant to hear. By default this tool asks for all three to be off, so the reading stays close to what actually reached the microphone, and choosing "On" asks for all three to be on, which gives conditions similar to a meeting app. What is requested and what is applied can differ, so the settings row shows the actual values the browser reports and adds "Applied differently from the request" when they do not match.
| Processing | What it does | Change in the readings when on |
|---|---|---|
| Noise suppression | Reduces steady sounds that are not a voice (fans, air conditioners) | Background noise drops a lot and the gap to noise grows. Check by recording whether the ends of words get smeared |
| Echo cancellation | Removes the other person's voice coming out of your speakers and re-entering the mic | Little change when you speak alone. A difference shows up when you talk with sound playing through speakers |
| Auto volume | Raises and lowers the volume automatically to match the input level | Quiet speech gets louder, and background noise may be raised along with it when it is quiet |
If the reading with processing off is poor but the one with processing on is good, the sound is acceptable in a meeting, but for recording or streaming you should fix the mic position or the surrounding noise first. Even with processing off, the operating system or the device itself may still apply some, so "Off" is not guaranteed to match studio-grade raw sound.
What to check with a Bluetooth headset
Bluetooth Classic audio has a music connection (A2DP) and a call connection (HFP). According to Microsoft's documentation on Bluetooth audio in Windows, A2DP only plays high-quality stereo and cannot use the microphone, so when an app opens the headset's mic, the device switches to HFP. HFP carries listening and speaking together on a single (mono) channel and runs at a wideband (16 kHz) or narrowband (8 kHz) sample rate. Philips support also explains that the reduced bandwidth and lower sound quality while the mic is in use is a limitation of Bluetooth technology, not a defect.
- If the music you were listening to on the headset suddenly turned muddy the moment you started the microphone test, it has switched to HFP. This is also why the tool turns the mic off while it plays a recording: with the mic on, you would hear call quality and the recording would sound worse than it really is.
- If the top of the frequency map is empty around 8 kHz or 4 kHz, sound is arriving at the bandwidth of the call connection. Record the same sentence with your laptop's built-in mic as well and compare.
- Hearing music through the headset does not mean the headset's mic has been selected. Check the device name in the settings row and the microphone name shown on the chip.
10-second recording and comparison
If the mic is not on, the 10-second recording turns it on first and then starts, and it ends by itself at 10 seconds. To end it sooner, press the same button again. Each recording keeps the microphone name, the voice processing setting, the length, the waveform and the speech and noise readings from that moment, and up to 6 recordings stack up.
For the file format, the tool picks the first one your browser can record from WebM (Opus), MP4 audio (M4A) and Ogg, in that order, so the extension may differ between browsers. The extension always matches the real recording format. It does not convert to MP3, and simply renaming the file to .mp3 does not change the format. In a browser without recording support, the waveform and readings still work.
When you compare, change only one thing at a time. Record the same sentence from the same place and distance while changing only the mic, or keep the mic and change only the voice processing, and you can tell where the difference came from. The speech and noise-gap numbers in a recording row are calculated over that recording's full 10 seconds.
Messages, their causes and what to do
| What you see | Cause | What to do |
|---|---|---|
| Stuck on waiting for permission | The permission prompt has not been answered. Without an answer the request never finishes and keeps waiting. | Find the permission prompt near the address bar and answer it, or press the button again to cancel and start over |
| Microphone permission was denied | Microphone access is blocked in the browser's site settings or in the operating system (NotAllowedError) | Allow it in both places. If only one is allowed, the mic will not turn on. |
| No microphone was found | No microphone is connected, or the chosen device was removed (NotFoundError) | Check the connection and the system's list of input devices, then choose again from the list |
| The microphone could not be opened | Permission was granted, but another app is using it, or the device or driver has an error (NotReadableError) | Close meeting and recording programs and start again, or unplug the device and plug it back in |
| The microphone cannot be used on this page | The connection is not secure (https), or the environment has no microphone support, such as a messaging app's built-in browser | Check the address and open the page in a regular browser |
| "Silent" in the background noise box, all samples 0 | Input muted in the operating system, the headset's mute switch, or a virtual device with no sound selected | Unmute, and check the system input volume and the selected mic |
| The waveform moves but no speech is found | Only noise is coming in, or you spoke for less than 1 second out of 10 | Speak two or three sentences in a row at your normal volume |
Works here but not in a meeting app
If sound comes in here, the microphone device and the browser permission are fine. A meeting app has its own input device choice, permission, volume, mute and voice processing, so what is left to check is on the app's side.
- In the app's audio settings, see whether the input device is the same as the device name you confirmed here.
- In your operating system's settings, check that app's microphone access permission. It is separate from the browser permission.
- Check the app's own mute, input volume and push-to-talk settings.
- Turn off the mic on this page, then check again with the app's own microphone test. Some devices fail on one side if two programs open the same mic at the same time.
Meeting apps such as Zoom, Discord, Google Meet and Microsoft Teams use their own noise suppression and auto volume, so the same mic may reach other people differently from what you heard here. A recording made here is material for narrowing down the cause, and it does not reproduce the final sound your listeners get.
Where sound stays and when the mic turns off
Sound analysis and recording are handled entirely inside this browser, and neither sound nor device information is sent to a server or kept in browser storage. Recordings exist only in the memory of the page that is open, so reloading or closing the page erases them. Download any recording you want to keep first. The delete button on a recording row removes only the recording on the page and does not delete a file you have already downloaded.
The mic turns off when you press Turn mic off, when a recording plays, when you change the mic or the voice processing and it reconnects, and when you switch to another tab or leave the page. While you are speaking, incoming sound is not sent straight to the speakers, so the feedback squeal of speaker sound re-entering the mic does not happen.
Common ways to misread the results
- Thinking a moving waveform means your voice is coming in: the waveform draws all incoming sound. If the waveform and frequency map look much the same after you stop talking, it is noise coming in.
- Reading dBFS as the loudness of the room: −20 dBFS means 20 dB below the digital maximum, not a 20-decibel sound. If you raise the input volume, the numbers rise in the same room.
- Taking a result with voice processing on as the mic's performance: with noise suppression on, background noise can drop below −90 dBFS. When comparing the mics themselves, use the readings with processing off.
- Thinking a high sample rate means a good mic: 48,000 Hz and 2 channels are only connection settings. Sensitivity, noise, distance and the room's reverberation affect the sound far more.
- Thinking no clipping means clean sound: the clipping indicator only checks whether the digital maximum was reached. It cannot catch distortion from the analog amplifier inside the mic, so listen to a recording to check.
Limits
- The readings are calculated from the digital sound the browser received. The tool does not measure a microphone's sensitivity, directionality or frequency response, and does not grade its quality.
- Speech and background noise are split by the distribution of levels over the last 10 seconds, so the tool does not actually tell a voice from noise. Loud music or other people's conversation mixed in is counted as speech too.
- The frequency map is drawn with an FFT of size 2048, one row every 20 milliseconds. Very low sounds (below about 100 Hz) have wide bins and look blurred, and precise acoustic measurement needs dedicated equipment.
- The actual voice processing values and the sample rate are what the browser reports. In a browser that does not report them, the tool shows "Unknown".
- The sound a meeting app transmits, network quality and the other person's playback device cannot be checked with this tool.
As of The source documents were checked on 2026-10-22.
Sources
- 01MDN: MediaDevices.getUserMedia()Only available in secure contexts (https, localhost); the request may never finish if the user does not respond; what exceptions such as NotAllowedError, NotFoundError and NotReadableError mean
- 02MDN: MediaStreamTrack.getSettings()Actual values of an audio track: autoGainControl, echoCancellation, noiseSuppression, sampleRate, channelCount, latency
- 03MDN: MediaRecorder.isTypeSupported()Checking which file formats the browser can record
- 04Microsoft Learn: Bluetooth Classic AudioA2DP is playback only, the mic uses HFP, HFP is wideband 16 kHz or narrowband 8 kHz, switching to HFP when an app opens the mic
- 05Philips Support: why Bluetooth headphone sound quality drops when the mic is usedThe bandwidth reduction while the mic is in use is a limitation of Bluetooth technology (article dated 2026-01-19)
- 06ITU-T G.722: wideband speech codingWideband 50 to 7,000 Hz
- 07ITU-T P.310: transmission characteristics for telephone-band digital telephonesTelephone band 300 to 3,400 Hz
- 08ASHA: acoustics in educational settings (technical report)The speech-to-background-noise ratio should be no lower than +15 dB
- 09Wikipedia: Voice frequencyFundamental frequency of adult voices (men about 90 to 155 Hz, women about 165 to 255 Hz)
- 10Wikipedia: Mains electricity by countryMains frequency by country, for example 60 Hz in the United States and 50 Hz in the United Kingdom
Frequently asked questions
Does the microphone turn on as soon as I open the page?
No. The first screen is a sample. The browser asks for microphone permission only after you press Start mic or Record 10 s. Even if you allowed it before, merely opening the page does not turn the mic on, and switching to another tab turns it off.
The speech box shows "—" and tells me to say something.
If the gap between the louder and the quieter parts of the last 10 seconds is under 10 dB, no speech is considered found. That happens when you have not spoken, spoke for less than 1 second, or the noise is as loud as your voice. Speak two or three sentences in a row at your normal volume.
How loud in decibels is −20 dBFS?
It cannot be known. dBFS says how far below the digital maximum a signal is, so it cannot be converted to dB SPL, the loudness of sound in a room. The same voice gives different numbers depending on the mic's sensitivity and the system input volume.
Is it normal that I cannot hear my own voice through the speakers while I talk?
Yes. Sending incoming sound straight to the speakers could feed it back into the mic and cause a squeal, so the tool does not do it. Record 10 seconds and use the play button to listen. The mic is turned off during playback.
Can I save a recording as MP3?
No. It is saved as one of WebM, MP4 audio (M4A) or Ogg, whichever your browser can record, and the extension matches the real format. Renaming a file to .mp3 does not change its format, so use a converter if you need MP3.
My Bluetooth earbuds' mic sounds muffled.
While the mic is in use, a Bluetooth headset switches to its call connection (HFP), and the sample rate drops to 16 kHz or 8 kHz, which cannot hold sound above 8 kHz or 4 kHz. If the top of the frequency map is empty, this is the case. Record the same sentence with a wired or built-in mic and compare.
With voice processing on, the background noise drops a lot. Is the mic better?
No. That is noise suppression removing sound that is not a voice. Meeting apps do similar processing, so it helps you guess how you sound in a meeting, but to compare the microphones themselves use the readings and recordings made with processing off.
Sound comes in here but not in Zoom or Discord.
If sound comes in here, the microphone and the browser permission are fine. Check the app's input device choice, the operating system's per-app microphone permission, and the mute and input volume inside the app. It also helps to turn off the mic on this page and test again with the app's own microphone test.