Microphone readiness

PrivacyBrowser onlyDefaultsGeneral check

For calls, interviews and proctored exams: a general check, not certification. Run your provider’s own system check. Levels are relative dBFS, never room loudness or dB SPL.

Peak{{ dbText(levelPeak) }}

RMS{{ dbText(levelRms) }}

Microphone:

Access starts only from Start. Inputs appear after permission. Audio, recordings, results and device IDs stay in this browser; no uploads or storage.

Signal path:

Calls requests browser echo cancellation, noise suppression and auto gain. Raw requests all three off. Neither guarantees what a call app hears; actual applied settings appear in results. Changes apply on next Start.

Measurement window:

Blank uses the standard 10-second window. Speak normally with natural pauses. Keep this tab visible; capture automatically stops after the window. Applies next Start.

Clipping threshold:

Blank uses the standard -0.5 dBFS threshold. Samples at or above it count toward clipping share and runs. Applies next Start.

Local recording:

Normally the measurement window is recorded locally for playback. Opt out for the next Start; signal measurements still work. Recordings are never uploaded or persisted.

Measurement	Value	Interpretation	Copy
{{ row.label }}	{{ row.value }}	{{ row.note }}

Criterion	Status	This tool’s default	Next action / limit	Copy
{{ row.label }}	{{ row.status }}	{{ row.threshold }}	{{ row.action }}

{{ 'Reference lines: clipping at ' + run.clip_db + ' dBFS; noise estimate ' + dbText(values.statistics.noise_db) + '.' }} Blocks span their measured intervals. Shaded gaps contain no delivered audio; lines break across them. Zero and levels below -120 dBFS appear at the plot floor only. Measurements retain finite values; zero-derived dB values are unavailable.

{{ 'The browser reported ' + muteEvents + ' input interruption(s). Check connection and mute, then retry.' }}

No failure recorded. Click Start microphone to check access.

Similar tools

Clear speech depends on more than a microphone appearing in a device list. A muted headset, the wrong input, or a weak signal can leave a connected microphone unusable. Before a call, interview, or proctored exam, a browser check can establish whether access succeeds and audio arrives. Listening to a short recording helps confirm that the captured sound is your voice and that the words are clear, which a moving meter alone cannot establish.

Digital audio levels describe the signal after capture and any processing applied along the way. They are commonly expressed in dBFS, meaning decibels relative to digital full scale. Zero dBFS marks the nominal full-scale reference; more negative readings represent smaller amplitudes. Microphone sensitivity, input gain, speaking distance, and processing all affect these readings. They do not measure room loudness in dB SPL, or sound pressure level, without acoustic calibration.

Sample peak: The largest captured sample magnitude. Peaks reveal brief bursts that a longer average can hide, such as a loud syllable or a knock on the microphone.
RMS level: Root mean square, an average based on the energy of the samples. It describes the sustained digital signal level over an interval, including pauses. It is not a direct measure of perceived loudness.
Headroom: The space between the sample peak and digital full scale. Leaving headroom allows louder words without pushing the captured signal into overload.

Clipping occurs when a recording stage cannot represent a signal's peaks and cuts them off, which can produce harsh distortion. Gain staging means setting the level through the recording chain so speech is strong enough while its peaks still have room. Input gain affects what is captured; headphone or speaker volume affects how loudly the recording plays. Turning down playback cannot restore peaks already lost during recording.

The noise floor describes the level of unwanted sound underlying the wanted signal, including room noise and electronic hiss. If speech is weak compared with that background, simply increasing gain can raise both. Better microphone placement and less surrounding noise can improve the separation. Moving closer can help, but speaking directly into the microphone from too close can introduce breath bursts or change the tone. Natural pauses make it easier to hear what remains between words.

Calls often use processing to make conversation clearer. Echo cancellation tries to remove played sound that returns through the microphone. Noise suppression reduces unwanted background sound, while automatic gain control adjusts the signal level. These processes also change the levels and sound being assessed. A request for less processing does not guarantee an untouched recording because the microphone or operating system may still apply its own adjustments.

Access, a live signal, and understandable speech are separate checks. Site or operating-system permissions can block access, while a hardware mute switch or an unintended default input can leave an apparently connected microphone silent. A call application may choose a different input or apply different processing from a browser check, so its own audio check remains part of preparing for the conversation.

How to Use This Tool:

Use the microphone, speaking position, and room you expect to use for your call. The standard measurement window is 10 seconds, with local recording enabled. Make any changes in Advanced before starting; they apply to the next capture.

Choose Calls under Signal path for a speech check with echo cancellation, noise suppression, and automatic gain control requested. Choose Raw request for a comparison with all three requested off. The browser may not apply every request.
Click Start microphone and allow microphone access if prompted. The microphone stays off until this action, and available inputs appear only after access succeeds.
Select the intended Microphone once inputs are listed, then click Switch microphone if you need to change it. This closes the previous input and starts a fresh capture. System-default and communications entries can lead to the same physical microphone.
Speak normally with natural pauses and keep the tab visible until capture stops automatically. The peak and RMS meters should respond to the sound being delivered.
Use Stop to cancel capture. It discards the recording, even when pressed after a completed test, but keeps the measurement evidence. Let the capture finish automatically if you want to listen back.
Read Readiness criteria to find the passing, warning, unavailable, or failing checks. If access fails, open Troubleshooting and follow the guidance for the reported problem before starting again.
Play the completed capture in Local recording and listen for clear words, unwanted noise, and distortion. Playback uses the system's default audio output. Repeat the test after changing the microphone, input level, or position.

Interpreting Results:

Wait for the capture to finish before treating the verdict as a completed check. A provisional warning or failure can appear while audio is still arriving.

Ready means the full capture met every stated criterion.
Ready with warnings means the capture completed without a failing criterion, but at least one check has a warning or an unavailable value. Read the affected criterion before deciding what to change.
Not ready indicates a capture problem or a failing criterion. The criterion and troubleshooting message identify the reason.
No signal means samples arrived, but all were zero or extremely small. Check the selected input, hardware mute, and input level; permission alone does not establish usable audio.

Voice estimate and Noise estimate describe louder and quieter portions of the capture. They do not identify speech or isolate background noise. A steady tone, music, continuous talking, or strong noise suppression can make their interpretation misleading. Listen to the recording before treating a favorable estimate as evidence of clear speech.

The Level timeline helps locate brief peaks and changes during the run. Shaded gaps with broken lines mean audio was missing. Delivered silence remains a measured interval. Zero and finite values below −120 dBFS share the plot floor, so the line's position there does not establish an exact −120 dBFS measurement. Measurement CSV retains unrounded values and their availability. Timeline CSV distinguishes actual levels from plotted coordinates, including the visual floor.

When comparing Calls and Raw request, check the reported Echo control, Noise filter, and Auto gain settings. An unreported setting is unknown, not proof that processing was off. Keep the microphone, position, room, and speech pattern as similar as possible between runs.

These verdicts are a general check, not certification for any call, interview, or exam platform. They use this test's stated defaults rather than a provider's requirements. After resolving warnings and listening back, run the actual application's system check. Local playback does not test network delivery, what a remote participant receives, or a separately selected speaker or headphone output.

Technical Details:

Digital audio contains signed amplitude samples. Peak calculations use their magnitudes, while RMS calculations square them so positive and negative excursions do not cancel. All delivered channel samples contribute to the measurements.

Formula Core

For N samples, the peak amplitude P is the largest magnitude and the RMS amplitude R is the square root of their mean squared value.

\begin{array}{l} P = max_{i} | x_{i} | \\ R = \sqrt{\frac{\sum_{i = 1}^{N} x_{i}^{2}}{N}} \end{array}

Each x is a normalized sample amplitude, with magnitude 1 representing nominal full scale. N counts samples across all analysed channels. Whole-capture RMS comes from the combined squared samples, not an average of displayed dB readings.

A positive amplitude a converts to a relative digital level D in dBFS as follows. Sample peak uses D(P), and RMS level uses D(R).

D (a) = 20 \times \log_{10} (a), a > 0

For example, an ideal full-scale sine wave has peak amplitude 1 and RMS amplitude 1 divided by the square root of 2. Under this plain amplitude convention, its peak is 0 dBFS and its RMS is approximately −3.01 dBFS. Some audio meters use a different RMS reference, so their numbers need not match even for the same signal.

Zero has no finite logarithmic level. A zero peak or RMS is therefore shown as unavailable rather than assigned a finite dBFS value. Finite values below the visual plot floor remain unchanged in calculations.

Estimating speech and background

Short blocks, normally about 100 milliseconds long, provide a rough distinction between quieter and louder parts of a capture. Sort their RMS amplitudes from lowest to highest. For m blocks, the background estimate uses the value at rank m divided by 10, rounded upward; the voice estimate uses rank 9m divided by 10, also rounded upward. Ranks start at 1, without interpolation.

\begin{array}{l} B = D (r_{⌈ \frac{m}{10} ⌉}) \\ V = D (r_{⌈ \frac{9 m}{10} ⌉}) \end{array}

Here r is a block RMS amplitude in that sorted list, B is Noise estimate, and V is Voice estimate, both in dBFS. Each block receives equal weight in the ranking. Blocks can be shorter at startup, at channel changes, around gaps, or at the end of a capture.

The difference between these estimates gives Voice margin. Headroom uses the distance from the sample peak to full scale, with a minimum reported value of zero. Both differences are expressed in dB.

\begin{array}{l} M = V - B \\ H = max (0, - D (P)) \end{array}

Natural pauses are an assumption of the background estimate. With continuous sound, quieter blocks may still contain speech or a test tone. Processing can also suppress pauses or change gain between them. The margin is an estimated separation within this capture, not a calibrated signal-to-noise ratio. If either selected block amplitude is zero, its dBFS estimate and the margin are unavailable; headroom is unavailable for a zero peak.

Clipping and near-zero samples

A sample contributes to Clipping share when its absolute amplitude is at or above the selected threshold. The threshold c defaults to −0.5 dBFS, with −1 and −3 dBFS alternatives. A more negative threshold flags smaller amplitudes; changing it does not change input gain or remove distortion.

\begin{array}{l} A = 10^{\frac{c}{20}} \\ C = 100 \times \frac{N_{c}}{N} \end{array}

A is the threshold amplitude, N_c counts samples whose magnitude is at least A, and C is the clipping share in percent. This is a threshold-crossing measure of clipping risk. It neither proves analogue distortion nor measures peaks between digital samples. Floating-point audio can also contain values above nominal full scale.

Clipped runs counts sequences of threshold-crossing samples separately for each channel. Runs can continue across adjacent blocks. The longest recorded run of L samples converts to milliseconds using the analysis sample rate f_s in samples per second; missing audio cannot establish what happened during a delivery gap.

t_{ms} = 1000 \times \frac{L}{f_{s}}

The near-zero boundary is an absolute amplitude of 0.00001, equivalent to −100 dBFS. Samples at or below that boundary count toward Near-zero samples. Delivered zeros count as audio data, so they are distinct from a failure to deliver audio.

Capture duration and coverage

Duration counts delivered audio frames, where one frame contains one sample for each channel. Coverage compares delivered frames with the full interval from the first block's start to the last block's end, including any gaps.

\begin{array}{l} t = \frac{F}{f_{s}} \\ Q = 100 \times \frac{F}{S} \end{array}

F is the number of delivered frames, S is the number of frames spanned by that interval, t is Measured time in seconds, and Q is Sample coverage in percent. The usual target is 10 seconds, with 5- and 20-second alternatives. Automatic capture ends after the target amount of audio arrives; gaps can still make coverage insufficient.

Input rate is the rate reported for the microphone, while Analysis rate is the rate used for these calculations. They can differ because the browser resamples audio. Analysis channels reports the largest channel count received, whereas Samples examined counts individual samples across every channel. These properties do not establish the microphone's acoustic accuracy or native bit depth.

Readiness rules

The following are this test's own general checks. They are not published microphone-quality standards or provider acceptance limits. Signal-level checks remain Not tested until samples arrive.

Exact pass, warning, unavailable, and fail conditions for the microphone readiness checks
Criterion	Pass	Other status	Fail
Complete capture	Automatic completion, full requested duration, and coverage ≥99%	Not tested before capture; Measuring while active	Capture ends without meeting all completion requirements
Signal present	At least one sample above −100 dBFS	Not tested without samples	All received samples at or below the near-zero boundary
Voice estimate	≥−35 dBFS	Warning from −45 dBFS inclusive to −35 dBFS exclusive	Below −45 dBFS, or a zero selected amplitude
Clipped samples	0%	Warning above 0% through 0.1% inclusive	>0.1%
Peak headroom	≥3 dB	Warning below 3 dB; unavailable for a zero peak	No separate fail condition
Background estimate	≤−45 dBFS	Warning above −45 dBFS; unavailable for a zero selected amplitude	No separate fail condition
Voice / background	≥15 dB	Warning below 15 dB; unavailable if either estimate has a zero amplitude	No separate fail condition
Input interruptions	Samples received with no browser-reported input mute events	Warning after a reported mute event, even if input resumes; otherwise Not tested without samples	No separate fail condition

The all-near-zero condition takes priority and produces No signal, including during an incomplete run. Otherwise, a failed criterion or capture error takes priority over the completed-run verdicts. Recording availability does not determine the readiness verdict.

Displayed numbers use up to two decimal places, while calculations and threshold comparisons use unrounded values. A rounded display value can therefore look equal to a cutoff while its criterion lies just above or below it.

Privacy Notes:

Microphone audio, recordings, results, and device identifiers stay in your browser and are not uploaded by this test. Captured samples, recordings, and device identifiers are not saved in browser storage. Capture requires an explicit start, including when the browser already remembers permission.

Local recording is enabled by default. Disable it in Advanced before the next capture for measurements without a playback recording. If recording is unsupported or fails, signal measurements can still work.
A completed recording is held temporarily for the open page. Discard recording in Advanced removes it without changing the measurements. Starting another capture also discards the previous recording. Leaving or reloading the page removes the temporary recording; a file you explicitly download remains on your device.
A WAV download is a local 16-bit PCM conversion of the browser recording. It is a listening copy, not an exact export of the samples used for the measurements, and conversion cannot undo any earlier recording compression.
The page address can retain signal-path, window, threshold, and recording preferences. It does not contain captured audio, results, or device identifiers, and opening it does not start capture.

The microphone is released on Stop, before switching inputs, and when you leave or close the page. Hiding the tab during startup or capture also interrupts the test and stops microphone access. Returning to the tab does not resume capture; a fresh explicit start is required.

Worked Examples:

More gain without better separation

Suppose a completed capture has a voice estimate of −30 dBFS, a noise estimate of −42 dBFS, and a peak of −8 dBFS. The estimated voice margin is 12 dB and headroom is 8 dB. The voice level passes, but background and margin produce warnings. If every amplitude were doubled without other processing changes, both estimates would rise by about 6.02 dB, leaving the margin at 12 dB while headroom falls to about 1.98 dB. Improving microphone position or reducing background noise addresses the poor separation; additional gain alone does not.

A clipping boundary hidden by rounding

A 10-second mono capture analysed at 48,000 samples per second contains 480,000 samples. If exactly 480 samples meet the selected clipping threshold, the share is 0.1% and Clipped samples shows Warning. With 481 qualifying samples, the share is approximately 0.100208% and that criterion becomes Fail, although both percentages can display as 0.1%. Follow the criterion and reduce input gain or increase microphone distance before repeating the capture.

FAQ:

Why is microphone access blocked after I allowed the site?

Website permission and operating-system permission for the browser are separate. Check both, then open the page directly over HTTPS in a current browser. An embedded page or an organization policy can also block capture, so a permission error does not necessarily mean you clicked Deny. The browser and operating-system choices under Troubleshooting provide relevant help. Start again after correcting the permission.

Why can the browser find my microphone but not read it?

Another meeting or recording application may be holding the input, or the device or connection may have failed. Close competing audio applications, check the connection, and retry. If a selected input disappeared, choose System default and start again. The error identifies an access problem; it does not establish which application or hardware part caused it.

What do “No audio arrived” and “Audio stopped arriving” mean?

The test stops if no audio block arrives within 5 seconds after access succeeds, or if delivery pauses for 3 seconds after audio has started arriving. These are delivery failures, not judgments about how quietly you spoke. Actual zero-valued samples still count as delivered audio. Check the input connection and browser, then explicitly retry; any partial measurements remain evidence of an incomplete run.

Why does my Bluetooth headset sound different when its microphone is active?

Bluetooth Classic can use different profiles for listening and conversation. On Windows, A2DP, the Advanced Audio Distribution Profile, provides stereo playback; HFP, the Hands-Free Profile, supports microphone capture with playback. Opening the headset microphone can switch to the conversation profile and change the sound. Behavior depends on the headset, operating system, and Bluetooth audio technology. Recheck the input and output combination used by your call application; this test does not identify or select the Bluetooth profile.

References:

Media Capture and Streams, W3C, Candidate Recommendation Draft, 9 October 2025.
Web Audio API 1.1, W3C, Working Draft, 22 September 2026.
Level Meters, Room EQ Wizard.
Signal Generator, Room EQ Wizard.
Techniques for Gaining Stage Your Microphone During Recording, RØDE, 26 March 2026.
How to Position Your Microphone for Streaming, RØDE.
Use your camera and microphone in Chrome, Google Chrome Help.
Bluetooth Classic audio, Microsoft Learn, 14 November 2024.

Microphone Test