Noise Suppression Test: Check Mic Noise Floor and SNR Free
Overview
Noise Suppression Test. Two six-second capture stages make up the Noise Suppression Test: hit Start Baseline and stay silent so it can read your room's raw noise floor, then hit Start Speech Capture and talk normally so it can measure your Voice Level and calculate the resulting SNR. That's different from the background noise rejection test, which plays noise through your speakers instead of measuring your room as it actually is. Try the midi test for a quick, no-install way to check this yourself.
Your colleagues hear every fan whirr and keyboard tap long before you do, so this online noise suppression test listens for you. It measures how much background noise your microphone picks up, how far your voice rises above it, and how much your noise cancellation actually removes, so you get a clear read on speech clarity and audio cleanliness in under a minute.
What a Noise Suppression Test Measures
The tool reads your microphone through the Web Audio API, and everything runs locally in your browser with no registration, so no audio leaves your device. Each reading shows how well noise suppression is working, not just how loud the room is. It reports a few real-time values, all in dBFS (decibels relative to digital full scale). That makes the reading relative: it depends on your microphone gain and hardware, and it is not a calibrated dB SPL sound meter. An FFT view plots the frequency spectrum live, so you can see which part of the noise spectrum is loudest and whether your noise level changes when suppression switches on.
Noise Floor and Mic Self-Noise
The noise floor is the level your microphone records while you stay silent. It mixes ambient noise from the room with mic self-noise, the hiss the capsule and preamp generate on their own. Spec sheets list self-noise as equivalent input noise (EIN) in dBA; a large condenser picks up far more of the room than a dynamic capsule. A lower (more negative) dBFS floor means a quieter channel.
Signal-to-Noise Ratio (SNR)
SNR is the gap in decibels between your signal level while you talk into the microphone and the noise floor while you don't. It is the single best predictor of how clean your speech will sound to a listener, because it captures both how loud your voice is and how quiet the room is.
Hum Detection at 50 Hz and 60 Hz
Mains hum is power-line energy leaking into audio as a steady tone at 50 Hz (Europe, Asia) or 60 Hz (North America) plus its harmonics. The hum check compares those frequency bins, and the octave bands around them, against the rest of the floor. A spike there points to wiring, not to your room.
How to Run a Background Noise Test in Your Browser
The free noise suppression test needs nothing installed and asks for no registration. Run the noise test twice, once with noise suppression off and once with it on, so the second pass has a baseline to beat. Use the background noise rejection test tool whenever you want a fast, repeatable way to confirm this.
- Click Start and grant microphone access when the browser asks for microphone permission.
- Stay silent for about five seconds while the tool captures the background noise baseline.
- Speak at your normal meeting volume so it can measure your voice level and SNR.
- Switch on noise suppression or your noise cancellation app, repeat both steps, and compare the two runs.
If the reading jumps around, turn off auto-gain (AGC) in your OS sound settings; it raises the gain during silence and inflates the floor. Nothing is recorded or stored, so privacy is never traded for a result. You can rerun it with no registration between app settings, and there is no registration step or account to delete afterwards.
How Labs Handle Noise Suppression Testing
Professional noise suppression tests go further than one desk check. Labs play recorded traffic or office noise conditions through different devices while a talker speaks, then score the output. Algorithms built on spectral subtraction or machine learning are compared with objective DNSMOS scores and listening panels that follow ITU-T P.835, which rates speech quality, noise intrusiveness and overall quality separately. Tools like Krisp offer a simpler version: a short test recording with comparison playback, so you can hear the processed and unprocessed audio back to back.
Reading Noise Reduction Results: Floor, SNR and Score
Every number the noise suppression tester shows comes from two measurements, the silent floor and the speaking level, so the maths is short.
Suppression Effectiveness Formulas
SNR is a straight subtraction of two dBFS readings:
$$\text{SNR} = L_{\text{speech}} - L_{\text{floor}}$$Suppression effectiveness compares the floor with processing off and on. If the floor drops by \(\Delta\) dB, the share of noise amplitude removed is:
$$\text{Reduction} = 1 - 10^{-\Delta/20}$$Use \(10^{-\Delta/10}\) instead if you want the share of noise power. Both describe the same noise profile change; amplitude is closer to what you hear.
Worked Example: Before and After Suppression
Say your laptop microphone shows a floor of −47.3 dBFS and speech at −18.6 with processing off. That is an SNR of 28.7 dB. With software noise suppression such as RTX Voice or NVIDIA Broadcast switched on, the floor falls to −63.8 dBFS while speech dips slightly to −19.4 dBFS.
- New SNR: −19.4 − (−63.8) = 44.4 dB, 15.7 dB better than before.
- Floor drop: 16.5 dB, so about 85.0% of the noise amplitude (97.8% of its power) is gone.
- Speech cost: 0.8 dB, small enough that your voice stays natural; a Krisp or app setting that costs several dB of speech is too aggressive.
Those figures become your noise suppression score: big floor drop, tiny speech loss.
Good SNR Targets for Calls, Podcasts and Streaming
| SNR you measured | Noise floor to aim for (dBFS) | How your voice sounds | Next step |
|---|---|---|---|
| Under 18 dB | Above −44 | Speech competes with the room | Turn on noise suppression or voice isolation |
| 18–27 dB | −44 to −53 | Clear enough for calls and video conferencing | Move the mic closer |
| 27–38 dB | −53 to −61 | Good for streaming and remote work | Use a cardioid or dynamic mics |
| Over 38 dB | Below −61 | Ready for a podcast or music recording | Add acoustic treatment |
Checking an Audiobook Room with the Noise Suppression Tester
A narrator preparing a first audition for ACX, Audible's production marketplace, needs a yes or no on one question: can this spare bedroom pass ACX's room-tone rule, a noise floor no higher than −60 dB RMS, without heavy processing? The answer turns out to be yes, but not by leaning on suppression.
The tester's first pass runs with processing off. The USB condenser sits 15 cm from the narrator's mouth, input level unchanged from narration sessions. Five seconds of silence read −54.7 dBFS; a paragraph read aloud averages −20.9. The signal-to-noise ratio is −20.9 − (−54.7) = 33.8 dB, and the floor misses the target by 5.3 dB. To close that gap, something has to remove about 46% of the noise amplitude, since 1 − 10−5.3/20 ≈ 0.457.
The second pass switches on NVIDIA Broadcast at its default strength. The floor drops to −68.2 and the voice to −21.6, so the ratio climbs to 46.6 dB, and the tester's before-and-after floor comparison puts the noise amplitude removed at roughly 79%. On paper that passes easily. On playback, though, the narrator hears a faint watery edge on soft "s" sounds, the kind of processing damage ACX reviewers reject.
| Run | Floor | Voice | Ratio | Meets −60? |
|---|---|---|---|---|
| Processing off | −54.7 | −20.9 | 33.8 dB | No |
| Suppression on | −68.2 | −21.6 | 46.6 dB | Yes, with audible damage |
| Off, fridge unplugged | −61.3 | −20.9 | 40.4 dB | Yes, 1.3 dB to spare |
The spectrum view shows a hump near 120 Hz, so the third run unplugs the fridge in the hallway. With processing off, the floor falls 6.6 dB to −61.3, passing with nothing applied.
My judgement: the number to watch is the margin, not the pass. A 1.3 dB cushion disappears on a warmer day when the air handler runs, so the plan is to record with the fridge off and keep suppression at its lightest setting as insurance. One caveat a quick search won't tell you: many browsers apply their own noise processing to the mic stream by default, so the on-page reading can look cleaner than the raw file your DAW captures. Before submitting, check the room tone in the actual exported file, not only in the tab.
Noise Gate vs Suppression vs Voice Isolation
These three tools solve different problems, and the test shows which one you need. A noise gate mutes the mic whenever the level falls below a threshold; its attack, hold and release time settings decide how fast it opens and closes. It does nothing while you talk. Noise suppression estimates the unwanted part of the signal and removes it continuously. Voice isolation goes further and tries to keep only the main talker, which helps with competing speech but can leave artefacts.
Keyboard Clicks, Fan Noise and Hiss
Steady background noise like air conditioning, HVAC rumble and white noise is the easiest for noise suppression to learn and remove. Transients are harder: a noise gate hides clicks between sentences, yet once you speak the gate opens and they pass straight through. Test while typing and talking at once.
Nearby Voices and Office Chatter
Another person's voice shares the same frequencies as yours, so even voice isolation reduces it rather than erasing it. A noise gate can't help either, since a nearby talker crosses its threshold just as you do. Push the strength too far and your own speech turns robotic or muffled. Treat reduction as an improvement, never as a privacy guarantee.
Test Noise Cancellation in Zoom, Teams and Discord
Each app adds its own noise suppression on top of any system tool, so test inside the app you actually use. Stack two aggressive layers and artefacts multiply; set processing strength to moderate in one place first.
Gaming Chat and Push-to-Talk
For gaming, combine noise suppression with voice activation so the channel stays quiet between callouts, and fall back to push-to-talk in a loud room. Rerun the test with voice activation on and check that the floor between callouts drops well below your no-processing baseline before you trust it in a match.
Work Calls in Google Meet and Other Apps
- Check the noise cancellation option in Zoom and Teams before important meetings.
- Google Meet suppresses noise by default on most accounts; compare its result with your own app.
- A headset microphone beats a laptop microphone for speech because it sits closer to your mouth.
Improve Audio Quality After a Noise Checker Reading
Fix the source before you reach for more noise suppression software; the cleaner the input, the less processing your voice has to survive. Plosive and wind noise respond to a pop filter in front of the microphone and a high-pass filter, while most other problems are hardware or setup.
Electrical Hum and Ground Loop Fixes
Electrical hum that the test flags usually comes from a ground loop or unshielded cable. Try an isolation transformer, shielded cables, and moving your audio interface away from power bricks to cut RF interference. A notch filter is the software fallback.
Room Noise, Input Gain and Mic Boost
- Lower the input gain until speech peaks stay well below clipping.
- Turn off mic boost; it lifts background noise as much as your voice.
- Cut room noise at the source: close windows and point the microphone away from the fan.
- Rerun the test after each change so you know which fix moved the audio numbers.
Frequently Asked Questions
- What does a noise suppression test measure?
- It measures your microphone's noise floor while you stay silent, your signal level while you speak, and the gap between them (the signal-to-noise ratio). It also checks for 50 Hz and 60 Hz electrical hum. All readings are relative dBFS values, not calibrated dB SPL.
- Is noise suppression the same as noise cancellation?
- Not quite. Noise suppression removes unwanted sound from the microphone signal in software. Active noise cancellation in headphones plays an inverted sound wave to cancel noise at your ears. Apps often use the two names for the same mic feature.
- How do I test software like Krisp, NVIDIA Broadcast or Discord's suppression?
- Run the test with the software switched off and note the noise floor and signal level. Switch it on, select its virtual mic if it uses one, and run the test again. Enter both sets of readings to see the floor reduction and SNR improvement.
- What SNR is good enough?
- Around 20 dB keeps speech clear on calls, roughly 30 dB suits streaming and remote work, and recording a podcast or audiobook calls for about 40 dB or more with a low noise floor.
- Why is my noise floor high in a quiet room?
- Microphone self-noise, too much input gain or mic boost, a sensitive condenser capsule, USB power noise, or automatic gain control raising the level during silence can all lift the floor. Lower the gain, turn off AGC and move closer to the mic.
- How do I fix 50 Hz or 60 Hz hum?
- Hum usually comes from a ground loop, unbalanced or unshielded cables, or power supplies near your interface. Try a ground loop isolator, balanced or shielded cables and a different USB port. A notch filter is the software fallback.
- Does the test record or upload my audio?
- No. The analysis runs locally in your browser through the Web Audio API, and no audio is stored or sent to a server.