Install from Microsoft Store
How to read your room's noise floor
The noise floor is the level of background sound present in a room when no intentional signal is being produced, and every room has one, caused by HVAC systems, computer fans, traffic, and the thermal noise of the microphone capsule itself. In audio production, a lower noise floor means cleaner recordings with more usable headroom, and a high floor eats into that headroom and forces you to record louder, which in turn captures even more of the background in the final track.
The result is expressed in dBFS (decibels relative to full scale), where full scale, 0 dBFS, represents the maximum digital amplitude.1 All real-world signals sit below this, so noise floor readings are negative numbers, and the further from zero, the quieter. A reading of −58 dBFS is quieter than −42 dBFS, and because the decibel scale is logarithmic, that gap represents a far larger difference in actual energy than the raw numbers suggest.
A concrete reading makes the workable threshold easier to judge. If your 3-second test reports
−58 dBFS, that sits comfortably below the −50 dBFS mark most home setups
need for usable voice recording, so background noise should stay well out of the way during normal
speech. A reading of −38 dBFS, by contrast, sits above that threshold, and you would
expect HVAC hum or fan noise to be audible under quieter passages of the recording.
TIP Close doors and windows, switch off fans, and silence your phone before running the Noise Floor test. Even brief silences between keystrokes or breaths will pull the average down and give you an optimistic result. Stay genuinely still for the full 3 seconds. The reading you get is only useful if it reflects the room as it sits while you are actually recording, so test in the environment you will really use.
Is it the room or the microphone?
A Noisy grade can mean a loud room with a good microphone or a quiet room with a microphone that adds its own hiss, and the two need opposite fixes. Run the test twice at the same gain. First switch off every fan, vent and noisy program you can and close the windows, then test again with everything running as it normally does. If the second reading is much louder, the room is your limit and moving or treating it will pay off; if the two readings sit within a few dB of each other, the microphone and its preamp set the floor, and only quieter hardware or a closer working distance will lower it.
Each reading is a 3-second snapshot, so expect it to wander by roughly 2 to 4 dB between runs as intermittent sounds come and go. Run it three times and use the middle value, and treat a change smaller than about 3 dB after you add an absorption panel or move the desk as noise rather than progress. On Windows, test with the microphone's enhancements switched off first, because a driver noise suppression mode can turn a Noisy room into an Excellent reading while adding audible artifacts.2 Measure the hardware first, then decide whether you want that processing on for real calls.
Your room's noise floor sets the lower limit of audio quality for everything recorded or transmitted through your microphone. Measuring it accurately requires a methodology that isolates ambient room noise from microphone and preamp noise. The Noise Floor Grade test does exactly this by taking a 3-second RMS average during silence. Before running the test, understanding what it measures helps you interpret the result correctly and take the right corrective action.
The result is a dBFS value: negative numbers closer to zero mean louder background noise; values further from zero mean quieter. A reading of −58 dBFS is significantly quieter than −42 dBFS, despite both being negative. Common benchmarks: professional recording studios target below −60 dBFS; home recording setups in quiet rooms typically achieve −55 to −65 dBFS; shared offices near HVAC typically measure −40 to −50 dBFS. The Noise Floor Grade labels these ranges Excellent, Good, Noisy, and Very Noisy.
Check the hardware first
Before running the Noise Floor Grade, verify the microphone is active by watching the Clipping Detector for any input signal while you speak briefly. The browser reaches the hardware through getUserMedia(), which resolves to a MediaStream only once the permission prompt is accepted.1 Then sit completely still with no typing, no motion, and no heavy breathing while you run the 3-second test. Any motion during the test window contributes to the measurement. The Noise Floor Grade measures everything present in the room during those 3 seconds: HVAC, street traffic, and PC fans, as well as the microphone's own electronic noise. This comprehensive measurement approach means that the reading reflects the total noise floor that every application on your system will encounter, making it a reliable baseline for comparing different microphones, gain settings, or room positions.
What the test results mean
An Excellent grade (below −60 dBFS) indicates the room is acoustically quiet and the microphone's self-noise is low, and this is the ideal baseline for any recording or AI voice service. A Good grade (below −50 dBFS) is workable for voice recording, podcasting, and AI voice use. The Noisy range (−50 to −40 dBFS) will affect transcription accuracy, AI voice interaction reliability, and recording quality. Furthermore, the source of noise matters: microphone self-noise is fixed hardware; room noise is reducible through closing windows, turning off fans, and acoustic treatment.
Why knowing your noise source matters more than the number
A Noise Floor Grade of Noisy could mean a genuinely noisy room with a great microphone, or a very quiet room with a cheap microphone that adds significant self-noise. These two situations require completely opposite fixes: the first needs acoustic treatment, the second needs hardware replacement. Running a second test after closing all noise sources in the room lets you determine which factor dominates in your specific setup, and that distinction dictates where you invest time and money for the most impactful improvement.
Platform-specific fix path
On Windows, the Noise Floor Grade changes depending on whether Windows audio enhancements (particularly Noise Suppression) are enabled. Windows exposes this as a selectable audio signal processing mode, with noise suppression offered as one of the modes the driver can apply to the capture path rather than as a separate application.2 Test with enhancements disabled to get the hardware baseline. On macOS, ensure no other application is accessing the microphone simultaneously; shared access causes additional audio processing that affects readings. On all platforms, reduce input gain until the Noise Floor Grade stops improving; once the reading stabilizes, further gain reduction only reduces signal level without improving the room's noise floor contribution.
Windows Noise Suppression applies a real-time noise suppression algorithm3 that can improve a Noisy reading to Good or even Excellent, but the improvement comes at the cost of adding processing artifacts that are audible in the Echo Loopback playback. If you measure with suppression enabled and then record with it disabled, the recording will sound worse than the grade suggested. Always measure with all enhancements disabled first to understand your true hardware baseline, then re-enable suppression only if you plan to keep it active for the actual recording or call session.
Separating room noise from microphone self-noise
Separating the room's contribution from the microphone's own electronic noise requires two separate test conditions: a maximum-silence test and a normal-conditions test. The measurement is an RMS average over the window, which is the standard way to report a broadband level because it folds every frequency into one number; 3 seconds is long enough to average over intermittent sources like distant traffic without being long enough for HVAC cycles or a passing vehicle to dominate the result.4 For the maximum-silence test, switch off all fans and ventilation, close windows, stop any processes on the computer that generate sound, and run the Noise Floor Grade three times. The result represents the closest approach to the microphone's self-noise limit achievable in your environment. This reading is hardware-bound: nothing you do to the room can improve it further. The maximum-silence test essentially answers the question "what is the best this microphone can possibly perform in any environment" and that number becomes the ceiling for what acoustic treatment in your room can ever achieve.
What the maximum-silence reading actually measures
The maximum-silence reading is the sum of two contributions: the microphone capsule's own thermal noise and the ADC's quantization noise at your current gain setting.5 At low gain settings, ADC quantization noise dominates and the reading is artificially high; at medium gain, the balance shifts toward the capsule. Run the maximum-silence test at the same gain setting you plan to use for recording so the reading represents your actual hardware floor at your working gain level.6
Running the normal-conditions test for comparison
For the normal-conditions test, run the Noise Floor Grade again with everything running as it typically would during a recording session. The difference between the two readings quantifies how much your room is contributing to the total noise floor. If the difference is 15 dBFS or more, the room is the dominant noise source and acoustic treatment or relocation produces meaningful improvements. If the difference is under 5 dBFS, the microphone's own electronics are the limiting factor in your space, and upgrading the microphone or choosing a quieter self-noise model is the next effective step.
Tracking acoustic treatment effectiveness with the Noise Floor Grade
Tracking whether acoustic treatment improvements make a measurable difference requires a consistent testing methodology. Before installing any treatment, run the Noise Floor Grade three times and record the average. Use the same microphone position, the same gain setting, and the same time of day for both the before and after tests: building noise, traffic, and HVAC loads vary with time of day, and inconsistent test conditions produce unreliable comparisons.
What constitutes a meaningful improvement
A 3 dBFS or greater improvement confirms the treatment is making a measurable acoustical difference. An improvement under 2 dBFS falls within the natural run-to-run variation of the test and cannot be distinguished from measurement noise. If panels on primary reflection points do not produce at least 3 dBFS improvement, the remaining noise floor is dominated by direct-transmission-path noise (HVAC through open vents, traffic through windows) rather than room reflections. Absorption panels address reflections; they do not attenuate direct-path noise. Understanding which source dominates before purchasing treatment prevents spending on panels that address the wrong acoustic problem.
Treat the 3 dBFS threshold as the gate for any purchase decision, because a smaller change can be caused entirely by the test's own run-to-run variation rather than the material you installed. Keep the before-and-after measurements at the same hour and same gain so the only variable is the treatment itself, and watch noise floor before and after treatment before committing to more panels once a real session shows the reflection-dominated noise has actually dropped by that margin.
Measure your room's noise floor when setting up a new recording space, when choosing between desk positions, when troubleshooting audio quality complaints, or to evaluate whether acoustic treatment is making a measurable difference.
Comparing two desk positions in a home office
Position 1 near HVAC return: −38 dBFS (Noisy)
Position 2 against interior wall: −61 dBFS (Excellent) — 23 dBFS improvement without hardware change
Testing acoustic panel effectiveness
Before panels: −50 dBFS from early reflections and room modes
After panels on primary reflection points: −57 dBFS — measurable improvement from Good toward Excellent
- 1.
Mozilla Developer Network, "MediaDevices: getUserMedia() method," developer.mozilla.org, accessed September 2026. https://developer.mozilla.org/en-US/docs/Web/API/MediaDevices/getUserMedia
- 2.
Microsoft, "Audio Signal Processing Modes," learn.microsoft.com, accessed September 2026. https://learn.microsoft.com/en-us/windows-hardware/drivers/audio/audio-signal-processing-modes#noise-suppression
- 3.
Microsoft, "Audio Signal Processing Modes," learn.microsoft.com, March 2025. https://learn.microsoft.com/en-us/windows-hardware/drivers/audio/audio-signal-processing-modes
- 4.
"Root mean square," Wikipedia, accessed September 2026. https://en.wikipedia.org/wiki/Root_mean_square
- 5.
Sound on Sound, "Noise: What It Is & How Can It Be Avoided?," soundonsound.com, accessed June 2026. https://www.soundonsound.com/techniques/noise-what-it-how-can-it-be-avoided
- 6.
Sound on Sound, "Q. How Should I Optimise My Gain Structure?," soundonsound.com, accessed June 2026. https://www.soundonsound.com/sound-advice/q-how-should-i-optimise-my-gain-structure
Both. The reading combines the room's ambient noise level, acoustic coupling into the microphone, and the microphone's own electronic noise. In most environments, room noise dominates. In very quiet rooms, microphone self-noise becomes the limiting factor.
Yes. The test captures a 3-second snapshot. CapyToolkit's Noise Floor Grade is most reliable when you run 3 tests and note the average; variation of 2 to 4 dBFS between runs is normal due to intermittent noise sources.
Below −70 dBFS, quantization limits of 16-bit audio and browser audio stack noise floor become factors. Most consumer microphones have self-noise above this level anyway, so the limit is rarely encountered in practice.
Indirectly. Larger rooms have more air volume but also more distance between noise sources and the microphone. The key variable is the level at the microphone capsule, not room volume.
Computer monitors generate electrical noise that can enter USB microphone power rails via the host USB controller. Running fewer powered devices on the same USB bus sometimes reduces bus noise measurably.
What is microphone clipping and why it matters
Clipping occurs when an audio signal exceeds the maximum level that a recording system can represent, and in digital audio, that ceiling is 0 dBFS. When the signal hits this limit, the tops and bottoms of the waveform are literally cut flat rather than following their natural curve.3 The result is a harsh, buzzing distortion that cannot be removed in post-production.
Common causes of clipping
Common causes of microphone clipping: gain set too high on an audio interface, speaking too close to a condenser capsule, or sudden loud sounds (plosives, table knocks) hitting an unprotected input. Transient peaks are especially easy to miss because a brief overshoot can slam past 0 dBFS even when the average level on your meters looks perfectly safe. The fix is always to reduce gain before the signal reaches the converter, not to apply limiting after the fact.
How the detector flags clipping
The Clipping Detector watches for samples at the absolute maximum or minimum of the digital scale (values of 0 or 255 in the 8-bit display range4). When more than 1% of samples in a single frame hit these limits, the CLIPPING badge appears. The threshold is intentionally conservative because even a handful of clipped samples represent permanent distortion that no post-processing can undo. If you see the badge frequently during normal speech, reduce your input gain.
Reading a frequency response curve
The frequency response display shows how much energy your microphone captures across the audible range (20 Hz–20 kHz)5 on a logarithmic scale. The horizontal axis compresses the upper octaves so that each decade (10 Hz→100 Hz, 100 Hz→1 kHz, 1 kHz→10 kHz) takes equal visual space, matching how human hearing perceives pitch.6
What the shape tells you about voice capture
A flat-ish response across the speech range (80 Hz–8 kHz)7 indicates a well-balanced microphone for voice. Weak energy below 200 Hz produces a thin, telephony-like sound, common in small-capsule USB mics with aggressive high-pass filters. Weak energy above 4 kHz creates a muffled, blanket-over-the-speaker quality, often caused by poor capsule placement or a microphone designed for instrument close-miking rather than voice.
NOTE The Frequency Response display shows what the microphone is capturing in real time, including room acoustics and background noise, so the reading shifts as the sound in the room changes. Speak or play a consistent source to get a meaningful reading, and give the display a few seconds to settle before judging the shape. The display is a visualisation tool, not a calibrated measurement, and it cannot replace a proper microphone measurement setup.
Understanding round-trip audio latency
Round-trip latency is the total delay from a physical sound entering the microphone to the system's response being heard through speakers. In the Clap Latency Test, this path is: clap → mic capsule → OS audio driver → browser → JavaScript → tone synthesis → OS → speaker → room → mic again. Every step in that chain adds delay.
Hearing yourself: the Echo Loopback test
The Echo Loopback test answers a question the meters cannot: what does your voice actually sound like on the other end? Click Record & Play Back, speak naturally through the 3-second window, and the test plays your own audio back to you through your speakers. What you hear is exactly what a caller or a recording receives: the room echo that colours your words, the background noise that rides underneath them, and your true level, none of which the noise floor grade or the clipping badge can show you from the inside.
Reading the playback is straightforward. Echo in the recording means hard surfaces are bouncing your voice back into the microphone, and background noise in the playback is what a caller hears during your pauses. Change one thing at a time: move the microphone closer so your voice dominates the room, soften nearby hard surfaces, or lower the input level feeding the capsule. Retest after each change until the playback matches the voice you expect, because the 3-second clip gives you a direct before-and-after comparison that no single number on this page can replace.
What counts as acceptable latency
Browser audio latency varies by browser, operating system, audio driver, and hardware.8 Web Audio exposes baseLatency and outputLatency estimates, and browser-based DAWs have reported best-case round-trip latency around 30 ms with 10 ms as a good target.9 Musicians monitoring in real time generally find anything at or below 10 ms comfortable, while standard operating-system drivers without dedicated audio buffers can push the delay up to several hundred milliseconds. This matters for live monitoring (hearing yourself through headphones while recording), real-time collaboration, and any application that needs tight audio synchronisation.
How to get a reliable reading
For best results: use headphones during the Clap Latency Test to prevent the reference tone from feeding back into the microphone through room reverberation. Room reflections and speaker bleed are the main confounders that make a clean single-clap measurement hard to reproduce without that isolation. Sit in a quiet space and clap sharply once, because a crisp transient registers faster than a palm slap. If the test shows "Echo not detected," increase microphone gain or reduce speaker distance.
Test conditions for your actual use case
Microphone requirements differ enough between podcasting, video calls, and music production that a single test in a single environment gives an incomplete picture. Podcasting demands a low noise floor and clean frequency response across the speech range, with recording done in a treated room at close range. Your noise floor test should reflect actual recording conditions: the same room, the same chair position, and the same nearby equipment running during a typical session. A noise floor that looks acceptable in a quiet test environment can sit above the threshold for comfortable listening if your production HVAC system runs during the real thing, and that difference only surfaces when you test under realistic conditions.
Video calls and conferencing platforms
Video calls place different demands on the hardware. Conferencing platforms apply their own noise suppression and automatic gain control, which can mask a weak or noisy microphone at the cost of voice character and intelligibility artefacts. Running the clipping test at the gain level your conferencing software applies, rather than at maximum hardware gain, shows you whether the platform is aggressively compressing your signal. Frequent CLIPPING badges at normal speaking volume with the gain your video call software sets means the input gain is misconfigured for that use case, regardless of how clean the hardware appears at lower gain settings. When the microphone is dead rather than degraded, the first checks sit outside this page: the operating system's selected input device, the physical connection, and the platform's own device and permission settings, because this tester can only see what the browser has been handed.
Musicians and live monitoring
Round-trip latency matters most for musicians recording with software instruments or monitoring through headphones during a take. Latency above about 10 ms can become noticeable for monitoring, and 25 to 50 ms of software monitoring can start to sound more like double tracking than a clean echo.10 Ensemble research also shows that 10, 20, and 40 ms delays change coordination patterns, so the clap latency test gives you a sanity check before committing to a session rather than a promise that every setup can hit a fixed target.11 Consistently high latency results point to the driver buffer size setting in your audio interface software rather than a problem with the microphone itself.
Microphone Quality Reference
- Workable noise floor Below −50 dBFS
- Comfortable monitoring latency ~10 ms or less
- Perceived as double-tracking 25–50 ms
Run the noise floor and clap latency tests above and compare your own readings against these figures.
A 90-second microphone check before a call
Start by confirming which microphone you are testing. Your browser uses the operating system's default input unless you pick another one in its permission prompt, and a Mac can switch that default by itself when you plug in a headset or connect a display with audio.12 Sit where you will sit during the call, with the microphone where it will be, because the noise floor and the frequency display both describe that exact spot. A test run at the other end of the desk tells you very little.
Then work through the tests in order, since each one makes the next easier to read. Run the Noise Floor Grade and fix the room if it is Noisy. Run the Clipping Detector while you speak at your loudest normal volume and lower the gain if the badge appears. Put on headphones, record an Echo Loopback and listen for hum, echo or harsh consonants. Finish with the Clap Latency Test, still on headphones, so the tone cannot leak back into the microphone. The whole sequence takes about a minute and a half. If every test passes but people still say you sound bad, the call app's own noise suppression or gain control is the next place to look, not your hardware.
An online microphone check before a call or session takes less than 90 seconds and prevents the audio quality problems that derail meetings and recordings. The five tests in this tool (Noise Floor Grade, Clipping Detector, Frequency Response, Echo Loopback, and Clap Latency Test) cover every aspect of microphone performance that affects call quality and AI voice service reliability. Running them in sequence gives you a complete picture in under two minutes.
The recommended sequence for a pre-call check is: run Noise Floor Grade first (confirm the room is quiet enough), run the Clipping Detector while speaking at your loudest normal volume (confirm gain staging is correct), play the Echo Loopback test and listen for echo or distortion (confirm the signal path is clean), then check the Clap Latency Test with headphones (confirm acceptable round-trip delay). Any result outside the Good grade thresholds indicates a specific problem with a specific corrective action.
Check the hardware first
The Noise Floor Grade test is the entry point for every pre-call check. A Good or Excellent grade confirms the microphone is active, the room is quiet enough, and you are positioned correctly. If the grade is Noisy or Very Noisy, fix the environmental issue before running other tests; there is no point evaluating clipping or latency until the noise floor is addressed.
The Clipping Detector running simultaneously while you speak at normal volume confirms gain staging immediately. If the noise floor is already Noisy, clipping detection becomes less useful because the elevated background noise makes it harder to distinguish between gain staging problems and environmental noise. Address the noise floor first by reducing ambient sound sources, then run the Clipping Detector in the improved conditions to get a clean gain staging assessment.
What the test results mean
Noise Floor Grade Excellent or Good: proceed to remaining tests because the room environment is quiet enough for all five tests to produce meaningful results. Clipping Detector shows badge during normal speech: reduce gain 5 to 10 dB to bring the peak levels below the digital ceiling1. Echo Loopback reveals reverb: you are too far from the microphone or need absorptive material in the room to reduce reflections.
Clap Latency Test above 80ms: your audio stack is adding latency; check USB hub usage, audio buffer settings, or consider switching from Bluetooth to wired. Each result points to a specific corrective action that can be taken before the call begins. Mapping each test result to a single fix prevents the common troubleshooting mistake of changing multiple settings simultaneously, which makes it impossible to identify which change actually resolved the problem.
The pre-call test sequence is designed so that each test depends on the result of the one before it. You test the noise floor first because a Noisy room makes every other test harder to interpret. You test clipping second because clipping distortion in the playback masks room reverb that the Echo Loopback should reveal. And you test latency last because it is the most expensive to fix and the least likely to produce an urgent quality complaint compared to audible noise, clipping, or echo. Following this dependency chain means you fix the most impactful problems first and never waste time on downstream diagnostics that are invalidated by upstream issues.
Platform-specific fix path
On all platforms, the most impactful pre-call action is confirming the correct microphone is selected as the default input device in your OS settings before running these tests. The browser's getUserMedia selects the OS default unless your application overrides it2. If tests show unexpected results, check which device is actually in use via your OS audio settings. On Windows, the default device can be wrong after a system restart if multiple audio devices are connected. On macOS, the Sound preference panel shows the active input device.
Why the default device changes after connecting new hardware
When you connect a new USB microphone, Bluetooth headset, or HDMI display with audio capabilities, many operating systems automatically switch the default input device to the newly connected hardware3. This means the microphone you tested five minutes ago may no longer be the active input when you open your conferencing application. Running the Noise Floor Grade in a browser tab immediately before joining the call provides a real-time confirmation that the expected device is still selected and producing a usable signal, catching silent default-device switches that would otherwise result in joining a call with the wrong microphone or no microphone at all.
The pre-call test sequence in detail
Before any important call, the order of the tests matters as much as the tests themselves. Noise Floor Grade first: it establishes whether the room environment is usable. If the grade is Very Noisy, stop and address the noise source before running any other test. Continuing to the Clipping Detector with an unusable noise floor wastes time because the acoustic environment is the blocking problem. Once the grade reaches Good or Excellent, run the Clipping Detector while speaking at the volume you expect to use in the call.
The Echo Loopback test requires headphones for accurate results: without headphones, the playback re-enters the microphone and creates an echo-of-an-echo that obscures the diagnostic. Connect headphones, run the 3-second recording, and listen specifically for low-frequency rumble (desk vibration), sibilance harshness (gain too high), reverb (room reflections), or hum (electrical interference). Then, with headphones still connected, run the Clap Latency Test: the test needs a sharp clap in a quiet room, and the headphones prevent the test reference tone from re-entering the microphone and confusing the timing measurement.
The Echo Loopback and Clap Latency tests both involve audio output that can re-enter the microphone if you are using speakers instead of headphones. For the Echo Loopback, speaker playback creates a feedback loop where the recorded audio includes the playback itself, making it impossible to distinguish between a genuine room echo and a speaker-to-microphone coupling artifact. For the Clap Latency Test, the reference tone must reach the microphone only through the air from your clap, not through a speaker that plays it back into the capsule. Using headphones for both tests is the single most important procedural step in the entire pre-call sequence.
When multiple tests show borderline results
When all your test results sit at the boundary between Good and the adjacent grade, prioritizing which to address first depends on what you are testing for. A noise floor at the boundary (around −50 dBFS) is the most important variable for AI voice services: even a 3 dBFS improvement reduces false activations in speech recognition4. Latency at the boundary (around 60ms) matters more for real-time conversational AI than for calls with another person, where latency below 150ms is perceptually unnoticeable5.
Using multiple runs to confirm a borderline grade
A Noise Floor Grade result that is Good on one run and Noisy on the next indicates an intermittent noise source. Run five consecutive tests and count how many fall in each grade range. If four of five return Good, the room is functionally Good with occasional intermittent noise that will not dominate a recording. If the result splits evenly between grades, the intermittent source is significant enough to address before the call. CapyToolkit allows rapid repeated testing to assess this variability quickly, making the five-run check a practical pre-session step rather than a time-consuming diagnostic.
Run the five tests back to back in the same sitting rather than spread across the day, because the environmental noise the check is designed to catch also drifts with time and would otherwise contaminate the count. If four of five readings land in Good, treat the line as crossed and proceed; reserve the deeper investigation for the even split, where the intermittent source is real enough to justify a change before the call.
Deciding when to proceed despite a borderline result
The decision to proceed with a call despite a borderline Noise Floor Grade comes down to the consequence of a degraded result. For an informal video call, a borderline grade is acceptable. For a podcast recording, an AI transcription session, or a call where word-for-word accuracy matters, run the full pre-call mic routine before accepting a borderline grade. Closing a nearby door, moving away from an HVAC vent, or turning off a fan takes less than two minutes and frequently improves a borderline Noisy result to Good.
Run this check before any call, podcast recording, voice AI session, or streaming setup. The 90-second pre-call check eliminates the most common causes of audio quality complaints before anyone else hears them.
Pre-call check before an important job interview
Echo Loopback reveals desk vibration hum — keyboard on the same surface as the mic stand
Placed mic on separate boom arm with feet off desk: hum eliminated, Excellent noise floor
Weekly podcast preparation
Clap Latency Test: 105ms — Bluetooth headset, accepted latency as unavoidable
Switched to USB wired headset: 31ms — noticeably improved recording sync and monitoring feel
- 1.
Sound on Sound, "Q. How Should I Optimise My Gain Structure?" soundonsound.com, accessed June 2026. https://www.soundonsound.com/sound-advice/q-how-should-i-optimise-my-gain-structure
- 2.
W3C, "Media Capture and Streams," w3.org, October 2025. https://www.w3.org/TR/mediacapture-streams/
- 3.
Apple, "If Your Mac Switches to a Different Audio Device," support.apple.com, accessed June 2026. https://support.apple.com/guide/mac-help/mac-switches-a-audio-device-mh35883/mac
- 4.
M. W"olfel and J. McDonough, "Distant Speech Recognition," John Wiley & Sons, 2009, pp. 173-180. https://dl.acm.org/doi/10.1002/9780470714161
- 5.
ITU-T, "Recommendation G.114: One-Way Transmission Time," itu.int, May 2003. https://itu.int/rec/T-REC-G.114-200305-I/en
Under 2 minutes for all five tests. The Noise Floor Grade is 3 seconds; the Clipping Detector runs in real time; Frequency Response is live; Echo Loopback is 3 seconds recording plus playback; Clap Latency Test is a single sharp clap. The entire sequence takes under 90 seconds.
Yes. The Noise Floor Grade and Frequency Response display both reflect room acoustics at your specific position. Run the tests from your actual call position for a representative result.
The problem is likely in the application's audio pipeline (compression, noise reduction, bandwidth management) rather than your hardware. A clean result here combined with poor call quality points to the application's processing.
No. CapyToolkit doesn't store any audio data or test results. Each test session is fresh. Note your passing grades if you want a reference for comparison in future sessions.
The Clap Latency Test measures browser audio round-trip locally, so it does not involve network traffic. VPN cannot affect the Clap Latency Test result. VPN does affect call application latency to the call server, which is separate.
- 1.
Wikipedia, "dBFS," en.wikipedia.org, accessed June 2026. https://en.wikipedia.org/wiki/DBFS
- 2.
Microsoft, "Audio Signal Processing Modes," learn.microsoft.com, accessed September 2026. https://learn.microsoft.com/en-us/windows-hardware/drivers/audio/audio-signal-processing-modes
- 3.
BAVC AVAA, "Digital Clipping," github.com, accessed June 2026. https://github.com/bavc/avaa/blob/master/_artifacts/digital_clipping.md
- 4.
Mozilla Developer Network, "AnalyserNode: getByteTimeDomainData() method," developer.mozilla.org, July 2024. https://developer.mozilla.org/en-US/docs/Web/API/AnalyserNode/getByteTimeDomainData
- 5.
R. Nave, "Sensitivity of Human Ear," hyperphysics.phy-astr.gsu.edu, accessed June 2026. http://hyperphysics.phy-astr.gsu.edu/hbase/Sound/earsens.html
- 6.
Physics LibreTexts, "Frequency Ratios and Pitch Perception," phys.libretexts.org, accessed June 2026. https://phys.libretexts.org/Bookshelves/Waves_and_Acoustics/Understanding_Sound_(Abbot)/02%3A_Perception_of_sound/2.05%3A_Frequency_ratios_and_pitch_perception
- 7.
Wikipedia, "Voice Frequency," en.wikipedia.org, accessed June 2026. https://en.wikipedia.org/wiki/Voice_frequency
- 8.
Mozilla Developer Network, "AudioContext: outputLatency property," developer.mozilla.org, April 2023. https://developer.mozilla.org/en-US/docs/Web/API/AudioContext/outputLatency
- 9.
W3C, "Audio latency in browser-based DAWs," w3.org, March 2021. https://www.w3.org/2021/03/media-production-workshop/talks/ulf-hammarqvist-audio-latency.html
- 10.
Martin Walker, "Dealing With Computer Audio Latency," soundonsound.com, April 1999. https://www.soundonsound.com/techniques/dealing-computer-audio-latency
- 11.
Auriel Washburn, Matthew J. Wright, Chris Chafe, and Takako Fujioka, "Temporal Coordination in Piano Duet Networked Music Performance (NMP): Interactions Between Acoustic Transmission Latency and Musical Role Asymmetries," Frontiers in Psychology, 2021. https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2021.707090/full
- 12.
Apple, "If your Mac switches to a different audio device," support.apple.com, accessed June 2026. https://support.apple.com/guide/mac-help/mac-switches-a-audio-device-mh35883/mac