Test Audio Input Devices: How to Check Every Microphone Your Browser Can See

Most computers no longer have one microphone. A laptop ships with a built-in array, a headset adds a second input over USB or a 3.5 mm jack, a webcam hides a third one, an audio interface exposes two channels, and a virtual cable quietly appends a loopback device that behaves like a microphone but records no air at all. When a call goes silent, the question is rarely whether the microphone works; the practical question is which of these devices the browser is listening to. This walkthrough shows how to test audio input devices one by one in a browser tab, read the level meter for each of them, and tell a dead input from a correctly selected one. The tester on this site does exactly that: it lists every input the browser can reach, opens a stream from the one you pick, and draws the live signal as a dB value and a level bar. Nothing is recorded and nothing is uploaded.

What the browser counts as an audio input

The browser does not talk to a microphone; it talks to the operating system, and the operating system presents a flat list of devices with a type attached. The MediaDevices API returns that list through enumerateDevices(), and every entry carries a kind of audioinput, audiooutput or videoinput, a deviceId, a groupId and a human label once permission exists. Everything the system recognises as a capture device appears under audioinput, and that includes hardware that has no capsule at all:

  • the built-in laptop microphone array;
  • a headset microphone on USB, Bluetooth or an analog jack;
  • a webcam microphone;
  • a USB microphone or one channel of an audio interface;
  • a virtual cable or loopback driver such as VB-Cable, Voicemeeter or BlackHole;
  • a capture device created by a screen-sharing or streaming tool.

This classification matters because a device being listed says nothing about it carrying sound. A virtual cable is a pipe between programs; until some application plays audio into it, the cable is silent by design, and testing it will correctly show a flat meter. A Bluetooth headset may appear twice once its hands-free profile activates. Enumerating the list first and testing device by device is the only way to know what each entry actually does.

Why the device list is blank until you grant access

A page can call enumerateDevices() before any permission, but the labels come back as empty strings and some devices are omitted entirely. The specification treats the hardware inventory as private information: without an active stream or a stored permission, a site could otherwise fingerprint a machine by its exact combination of microphones. So the test tool shows placeholders until you grant access, and the list fills in with real names only after that.

Two more conditions sit on top of the permission. The page must run in a secure context, which means HTTPS or localhost, because the microphone is available only to pages the browser considers trustworthy. And the permission itself is issued per site and per origin: a site blocked months ago stays blocked until someone clears that decision in the address bar settings. If you press the start button and the browser refuses without a prompt, the block is saved rather than pending, and no reload will raise the dialog again. Chrome and Edge can remember an allow decision per site; Safari deliberately asks again, which makes the prompt more visible on every visit.

Running the test on each device

The workflow is short and the order carries most of its value. Open the tester, press Start preview, and grant the microphone permission when the browser asks. From that moment the device dropdown is populated with real labels, and each selection opens its own stream. The steps:

  1. Pick a device from the dropdown.
  2. Press Start preview and speak at your normal volume and distance.
  3. Watch the bar and the dB number while you talk.
  4. Press Stop, pick the next device, and repeat.

The dropdown refreshes itself on devicechange, so plugging a headset in mid-session adds it without a reload. One quirk of the event deserves a warning: browsers fire it for changes beyond plugging hardware, for instance when the system default switches from the speakers to a headset, or when a Bluetooth device connects and immediately becomes the preferred input. A list that refreshes for no obvious reason is normal behaviour, not a bug in the page.

Test one input at a time and leave the others idle. Two open streams run in parallel without conflict, but a device that another program holds in exclusive mode can refuse the browser entirely, and the failure then looks like a silent page rather than a busy device. Closing the other application first removes that ambiguity.

What the meter actually measures

The bar is not a volume knob reading and not a loudness rating of the room. The tester routes the stream into an AudioContext, attaches an AnalyserNode, and every animation frame reads 1024 time-domain samples from it. Those samples are floating point numbers between -1 and +1, each one the height of the waveform at a moment in time. The code computes the root mean square of the whole block, which is the average energy rather than the single loudest point, converts it to decibels with 20 × log10(rms), and prints one decimal place.

The bar fill then maps that dB value onto a percentage. Anything below -60 dBFS counts as zero percent, 0 dBFS counts as one hundred percent, and the window between them scales linearly: percent = ((dB + 60) / 60) × 100. A steady -30 dBFS therefore fills half the bar, -45 dBFS fills a quarter, and -12 dBFS sits at eighty percent. The value is smoothed with an exponential filter so consonants do not make the bar flicker, while the dB number itself stays close to the raw frame value.

The window length follows the analyser's 1024 samples and the device's sample rate: at 48 kHz a block covers about 21 milliseconds, at 44.1 kHz about 23 milliseconds. That is short enough to react to a snapped finger and long enough to ignore individual samples, which is exactly what a quick input check needs.

Reference levels for a quick pass or fail

Speech at a normal distance from a built-in laptop microphone usually lands somewhere between -30 and -12 dBFS depending on input gain, and a headset positioned correctly tends to sit in a similar band. Numbers are worth reading relative to each other rather than against an absolute scale.

Reading What it means Action
Bar moves with your voice, idles near zero healthy input none
Bar stays at zero while you speak wrong device, hardware mute or a closed path reselect the device, check the mute switch
-30 to -12 dBFS on normal speech typical working range none
Bar pinned at 100% on speech input gain too hot, signal clipping lower the input level in the system settings
Bar jumps wildly between words automatic gain control working on the signal compare with AGC disabled if the app allows it

Clipping is the failure worth taking seriously. Once a sample reaches full scale the waveform is cut flat, and the distortion that follows cannot be repaired later. If the bar saturates on normal speech, lower the input level in the operating system before recording or calling.

Virtual cables and loopback devices

Virtual audio devices break the mental model of a microphone, and they cause a large share of confused test results. A virtual cable installs a pair of endpoints: a playback device that applications send audio to, and a recording device that hands that audio back to whichever program opens it. In the browser both halves appear, and the recording half is listed under audio inputs like any microphone.

Two consequences follow. First, a virtual cable tested on its own shows the meter idle at zero, because nothing in the system is currently playing into it; the silence is the correct answer, not a broken device. Second, when a program does route sound through the cable, the meter moves with that program's output, which makes the tester a quick way to confirm whether the routing actually works. The same logic covers loopback drivers that mirror the speaker mix. If a meeting app reports no input but the virtual cable moves when music plays, the fault sits in the app's device selection, not in the audio path itself.

Sample rates and channel counts on interfaces

Audio interfaces expose several capture channels under one box, and the list often shows them as separate inputs: analog input one, analog input two, and sometimes a loopback of the interface's own output. Each channel deserves its own test, because a microphone plugged into the second input of a two-channel interface is invisible on the first, and the meter will read zero until the right channel is selected. The browser does not mix them together.

Sample rate is the other variable. The AudioContext runs at the device's rate, commonly 44.1 or 48 kHz, and a mismatch between the system setting and the interface can leave the stream flat or produce a pitch shift rather than an error message. If one input works and another from the same interface does not, compare their configured rates in the interface's control panel before blaming the browser.

Bluetooth headsets change the picture

A Bluetooth headset is two devices in one, and its microphone exists only while the hands-free profile is active. When the headset switches from high-quality music playback to the call profile, the system swaps the audio device underneath the application, and an input that worked a minute ago can disappear from the list or come back with a different label. The classic symptom is a mic test that passes, followed by a call where the headset microphone sounds thin or refuses to start; the profile switch happened in between.

When a Bluetooth input misbehaves, test it twice: once with the headset connected and idle, once during a call or playback, and compare the two readings. If the browser receives sound only in the first case, the profile negotiation is the problem, and a wired or USB headset sidesteps it entirely.

Silent signal with the right device selected

The meter shows zero and the dropdown shows the correct microphone; the diagnosis moves down the chain, one stage at a time. Check the hardware mute switch first, because analog headsets still ship with one and it silences the capsule before any software sees it. Then confirm the operating system's input level for that device is not at zero, and that the device is not held by another program in exclusive mode. On systems with per-app privacy switches, the browser itself must be allowed to use the microphone at the operating system level, and the site prompt alone does not override that switch.

If every stage checks out and the meter still sits at zero, select another input and speak: sound appearing on a sibling device proves the audio stack works and narrows the fault to the one device's cable, port or driver. Swapping the device into a different port costs a minute and eliminates the adapter question.

Reading the idle noise floor

Stay silent for ten seconds with a working input and watch the bar. A quiet room typically leaves the meter at zero percent, because the -60 dBFS floor of the scale sits above the room's residual noise; a noisy office or a mic with high gain will keep the bar a few percent up, flickering with fans and distant voices. That idle level is the noise floor, and it sets how much clean signal you have to work with. A conversation that raises the bar only slightly above an active idle reading will sound distant to the other side, which is a cue to move closer or raise the input gain moderately.

The tester reports what the machine delivers, device by device, in one session. Enumerate the inputs, test each one with your voice, compare the numbers, and the question of which microphone the browser actually hears stops being a guess.

FAQ

01Why does my microphone show no name in the dropdown?

Device labels are hidden until the page receives microphone permission; press Start preview and allow access, and the real names appear. A page blocked in the past shows no prompt at all until the site permission is cleared.

02How do I know which of my inputs is the right one?

Select each entry from the dropdown and speak; the input that moves the meter is the one carrying your voice. Built-in, USB and headset microphones all appear in the same list, and virtual cables appear there too but stay silent unless audio is routed into them.

03The meter moves but the level stays very low. What does that mean?

A bar that barely leaves zero on normal speech points to a low input gain, a distant microphone or a device with its own attenuator. Raise the input level in the system settings, or move closer, and watch the reading change; typical speech sits well above the idle floor.

04Can I test a microphone that another app is using?

Only if the app releases it. A device held in exclusive mode refuses the browser, and the page then shows no signal even though the microphone works elsewhere. Close the other application and start the test again.

05Why did a device disappear from the list in the middle of testing?

The list refreshes on every device change, including Bluetooth profile switches and default-device changes at the system level. If the headset switched to its call profile, its microphone may come back under a different label; reconnect it and test again.

06Is anything recorded while I test?

No. The stream is analysed frame by frame inside the page and discarded immediately, and stopping the test closes the microphone completely. Nothing is stored locally and nothing is sent to a server.

Kay
Kay

Manchester, United Kingdom · SO reputation 19956 · Badges: 73🥇190🥈307🥉 · SO member since 2012