Background Noise Rejection Test: Check Mic Noise Suppression
Overview
Background Noise Rejection Test. The Background Noise Rejection Test plays background noise through your speakers while it records your voice, then plays back the result so you hear exactly how well your microphone rejects noise — instead of testing it in silence. The midi test gives you a clear answer instead of guessing.
Your background noise rejection test answers one question before a call or recording: how much of the room does your microphone setup actually keep out of your voice? Instead of a single loudness number, this noise suppression testing run measures your background noise level, your talking level and how many dB your suppression removes, so you can compare runs as you change gear and judge speech clarity and call quality with real numbers.
How the Background Noise Rejection Test Works
This is a free background noise test that runs entirely in your browser, with no registration, no account and no audio plug-in. It asks for microphone access through getUserMedia, reads the signal with the Web Audio API and keeps a rolling estimate of the level in real time. Nothing leaves your device: the analysis is local, the tool is free, there is no registration and no software to download. A real-time waveform and an FFT spectrum show what the microphone hears while each phase runs. Run the background noise suppression software test to see exactly what's working and what isn't.
- Click start and grant microphone permission when the browser asks.
- Stay silent for the baseline, then read a sentence at your normal volume.
- Switch your noise suppression on and repeat both steps.
Phase 1: Silent Baseline
For the first few seconds you say nothing. The tool records the ambient noise your mic picks up, which becomes the noise floor every later figure is compared against. Test from your usual seat: a quiet room elsewhere in the house tells you nothing about the desk you take calls from.
Phase 2: Speech Level and SNR
Next you talk. The tool measures how loud you arrive and subtracts the baseline to get SNR, the gap between your signal and the noise underneath it. A wider gap means listeners hear you, not the room.
Phase 3: Suppression On, Rejection Measured
Finally you repeat both steps with noise suppression enabled. The drop in the noise floor is your rejection in dB, and the change in SNR shows whether the processing also cost you level. The result panel shows a suppression score and a letter grade, a short test log of each phase, and a report you can save, so suppression effectiveness becomes a number rather than an impression.
Noise Suppression Formula: SNR and Rejection in Decibels
Levels in the test are measured in dBFS, decibels relative to digital full scale, where 0 is the loudest value the converter can store. Because the scale is logarithmic, you subtract levels instead of dividing them, and a 6 dB change doubles or halves the signal's amplitude. Try the free usb mic test for a quick, no-install way to check this yourself.
Signal-to-Noise Ratio as a Level Difference
The signal-to-noise ratio is the speech level minus the noise level, both in dBFS:
$$\text{SNR} = L_S - L_N = 20\log_{10}\!\left(\frac{V_S}{V_N}\right)$$Rejection is how far the noise floor falls when noise suppression is on, and the net SNR gain subtracts any loss on your side of the signal:
$$R = L_{N,\text{off}} - L_{N,\text{on}} \qquad \Delta\text{SNR} = R - \left(L_{S,\text{off}} - L_{S,\text{on}}\right)$$The share of noise energy left after suppression is \(10^{-R/10}\), so every 10 dB of rejection removes 90% of what remained.
Worked Example: Laptop Mic Beside a Desk Fan
Picture a built-in mic on a laptop with a desk fan running about a metre away. With suppression off, the baseline reads −47.3 dBFS and your talking peaks at −18.6, giving an SNR of 28.7 dB. With the meeting app's noise suppression on, the floor drops to −68.9 and your level dips slightly to −19.4.
| Measurement | Suppression off | Suppression on | Change |
|---|---|---|---|
| Noise floor | −47.3 | −68.9 | rejection of 21.6 |
| Speech level | −18.6 | −19.4 | 0.8 lost |
| SNR | 28.7 | 49.5 | +20.8 |
| Fan noise energy left | 100% | 0.69% | about 99.3% removed |
So 21.6 dB of rejection leaves the fans at roughly one twelfth of their original amplitude, and the processing costs you less than 1 dB. That is the kind of result a good noise suppression tester should show: a deep drop in the floor and almost nothing taken from the person talking.
Background Noise Level Targets for Calls, Podcasts and Streaming
A good result depends on the job and the environment. A room level in dB(A) is an A-weighted reading of sound pressure, the way a decibel meter reports it; the WHO guideline for bedrooms at night is 30 dB(A). Your microphone reading in dBFS is a different scale, so treat the targets below as working ranges for the noise level your setup should reach, not certified limits.
| Use | Room level | Noise floor (dBFS) | SNR to aim for |
|---|---|---|---|
| Video calls and meetings | ≤ 42 dB(A) | ≤ −52 | 25 dB+ |
| Podcast and voice recording | ≤ 33 dB(A) | ≤ −62 | 38 dB+ |
| Live streaming and gaming | ≤ 40 dB(A) | ≤ −56 | 30 dB+ |
| Speech recognition and voice AI | any | ≤ −50 | 22 dB+ |
Sound-meter apps often show L90, the level exceeded 90% of the time, and Leq, the energy average. For audio quality on calls and in a recording, the steady floor matters most, because short noises come and go.
How to Read Noise Checker Results
Each noise checker reading tells you something different about audio cleanliness:
- Noise floor
- The steady level while you are silent. Lower is quieter.
- Talking level
- How loud you reach the microphone.
- SNR
- The gap between the two. Bigger is cleaner.
- Rejection
- How far suppression pushes the floor down.
Noise Floor in dBFS vs dB SPL
A browser reading is a relative estimate, not a dB SPL measurement: it changes with gain settings and hardware. To calibrate against the room, hold a real sound level meter next to the microphone once and note the offset. Without that calibration, compare only runs made with the same mic and the same settings, and rerun two or three times, since one reading can catch a passing car.
Microphone Self-Noise and Equivalent Input Noise
Every microphone adds self-noise from its own capsule and preamp, quoted on spec sheets as equivalent input noise (EIN). A sensitive condenser in a silent room can show a floor that is mostly its own electronics. Automatic gain control (AGC, or auto-gain) makes this worse by raising gain during silence, so switch it off before you test, or microphone self-noise will look like room noise.
Accuracy, Limits and Privacy of Browser Readings
Phones and some headsets apply processing the page cannot disable, so background noise testing on mobile can show rejection you never switched on. A laptop's array microphone may already be filtering too. Privacy is simple: the online tool analyses audio in memory, needs no registration and never uploads or stores anything, and you can revoke access from the lock icon in the address bar.
Background Noise Testing a Spare-Room Audiobook Booth Against the ACX Floor
I'm recording a retail audiobook sample in a spare bedroom with a window air conditioner, a USB condenser about 30 cm away, and one decision to make: can software suppression get me under ACX's submission limit of −60 dB RMS for the noise floor, or does the AC have to go off for every session? The short answer, after four runs: turn the AC off, move in close and leave the processing off.
Each run is the same: four seconds of silence, then the same paragraph read at narration volume. The rejection test reports the floor, my level and the SNR, and I subtract the floors by hand to check the rejection figure.
| Run | Floor | Voice | SNR | vs −60 limit |
|---|---|---|---|---|
| AC on, suppression off | −51.8 | −20.7 | 31.1 | fails by 8.2 |
| AC on, suppression on | −63.4 | −21.9 | 41.5 | passes by 3.4 |
| AC off, suppression off | −61.2 | −20.7 | 40.5 | passes by 1.2 |
| AC off, suppression on | −69.0 | −21.6 | 47.4 | passes by 9.0 |
With the AC running, suppression rejects 11.6 dB and costs me 1.2 dB, a net SNR gain of 10.4 dB. Switching the AC off instead buys 9.4 dB with no processing at all. What stands out is the last row: in the quiet room, suppression only rejects 7.8 dB. Once the AC is gone, most of what's left is the condenser's own self-noise, which the algorithm handles less confidently than a steady fan tone.
The caveat that matters for ACX: a passing number doesn't mean passing audio. Suppression that pumps between sentences or dulls sibilants gets flagged by human reviewers even when the meter is happy, so I don't trust the second row on its own. And a 1.2 dB margin in row three is thin, since repeat runs drift by nearly a dB.
So the next step is specific: move the mic from 30 cm to about 15 cm and drop the input gain by 6 dB. My voice stays at −20.7 while the floor falls to −67.2, an SNR of 46.5 dB, clearing the limit by 7.2 dB without any processing. I'll confirm that in Audacity's ACX Check before the first chapter.
Rumble, Hum or Hiss: Noise Reduction by Source
Effective noise reduction starts with knowing which noise sources you are fighting. The spectrum view splits sound into frequency bands, and the octave bands where energy piles up during silence point to the cause; each source sits at a different frequency.
Background Noise from Fans, HVAC and Traffic
Fans, HVAC units, air conditioning and traffic produce low rumble below about 250 Hz, plus broadband hiss that sounds like white noise. In the silent baseline they show up as raised energy in the lowest bands, and because they are steady, they are the kind of noise suppression usually rejects best. A high-pass filter removes much of the low end without touching your words.
Mains Hum at 50 Hz and 60 Hz
Mains hum shows up as narrow spikes at 50 Hz or 60 Hz and their harmonics. Hum detection in the test flags this electrical hum, which usually comes from a ground loop, unbalanced leads or poor shielding. Balanced connections and shielded cables cut the interference at the source; a notch filter treats what is left.
Keyboard Clicks, Plosives and Wind Noise
Keyboard clicks are short transients that suppression often misses because they look like the start of a word. Plosive bursts and wind noise overload the capsule, causing clipping and distortion. A noise gate threshold set just above your floor mutes the gaps between words.
Noise Cancellation Settings in Zoom, Teams and Discord
Most calling apps ship their own noise suppression software:
- Zoom, Teams, Google Meet and Discord each have a setting for it in their audio options.
- Krisp, RTX Voice and NVIDIA Broadcast add a second software layer that every app can use.
They work by spectral subtraction or machine learning, estimating what is speech and removing the rest. Run a noise suppression test with each one in turn and keep the one that gives the deepest rejection for the smallest loss on your side. A free noise test run takes under a minute, so trying all of them is quick, and a free online noise suppression test of each app shows which one your ASR or your listeners will thank you for.
If the other person hears their own voice back from your speaker, that is echo, a separate problem handled by acoustic echo cancellation, not by noise suppression.
Acoustic Stress Testing for Voice Agents
Developers building voice AI use the same idea at scale. Acoustic stress testing plays recorded noise under scripted speech and measures how communication with the system holds up, from speech recognition to intent recognition. The goal is a degradation curve: performance plotted against falling SNR, with any sudden cliff marking the point where the system stops coping. Speech babble from nearby talkers is the hardest case, because babble sounds like a competing talker.
Voice Agent Metrics: Word Error Rate and Retry Rate
For voice agents, the core measure is word error rate (WER) from the ASR engine, compared between clean and noisy audio. A rising WER feeds into retry rate and task completion, because every misheard word forces a repeat. Poor voice activity detection lets noise trigger the ASR, which inflates WER again. Clean ASR input is why noise suppression often runs before recognition, though aggressive processing can raise WER if it smears consonants. Audio quality itself is often scored with DNSMOS or listening tests under ITU-T P.835, which give a MOS from 1 to 5. The SNR this page reports is the input to all of that: if your rejection run leaves you at an SNR where a model's error rate has already started climbing, fix the room or the microphone before you blame the model.
Noise Injection at Calibrated SNR Levels
Noise injection mixes real noise profiles (car, office, café) into clean recordings at fixed SNR steps, so noise robustness can be compared across ASR models and WER results on equal terms, the way the test compares one environment against another.
Improve Noise Robustness: Microphone, Gain and Room
Better hardware and a quieter environment beat any audio setting. Rerun the background noise rejection test after each change so you know which one moved your score.
Directional Mic, Mic Gain and Distance
- A directional mic with a cardioid polar pattern rejects off-axis sound from behind and to the sides.
- Move closer: each halving of distance raises your level by about 6 dB while the room stays the same.
- Lower mic gain and input gain, and turn off mic boost, so the floor drops too.
- A headset puts the capsule at your mouth, which helps more than any filter.
Room Treatment That Lowers Background Noise
Soft furnishings, rugs and acoustic panels absorb reflections that suppression struggles with. Close doors and windows, then run the online noise suppression test again to see how much the room itself improved. It is free and needs no registration, so rerun it as often as you like.
Frequently Asked Questions
- What is the difference between noise suppression and noise cancellation?
- Noise suppression removes background sound from your microphone signal, usually in software such as Krisp, RTX Voice or the setting in Zoom and Discord. Noise cancellation normally means active noise cancellation in headphones, which plays inverted sound to quiet what you hear. Only suppression changes what callers hear from you.
- How do I test my noise suppression software with this tool?
- Run the silent baseline and a short spoken sentence with your suppression switched off, then switch it on and repeat. Enter both sets of readings: the drop in the noise floor is your rejection in dB, and the change in SNR shows whether the software also took level away from your voice.
- What is a good noise floor for a microphone?
- Lower is quieter. Readings below about -60 dBFS are clean enough for most recording, while a floor above about -50 dBFS will be easy for listeners to hear. Because browser readings are relative, compare runs made with the same microphone and gain settings.
- Why is my noise floor high even in a quiet room?
- The usual causes are too much input gain or mic boost, automatic gain control raising the level during silence, a sensitive condenser picking up the whole room, USB power noise, or the microphone's own self-noise. Lower the gain, switch off AGC and move closer to the mic, then test again.
- How do I fix 50 Hz or 60 Hz hum?
- Mains hum usually comes from a ground loop, unbalanced cables or poor shielding. Try a ground loop isolator, balanced XLR connections and shielded cables, and keep the mic away from power supplies. A notch filter at 50 or 60 Hz in your recording software treats whatever is left.
- Is this an absolute decibel (dB SPL) meter?
- No. Browsers do not give calibrated sound pressure readings, so dBFS values are relative to digital full scale. For room levels in dB(A), add a calibration offset measured once against a real sound level meter.
- Does the test record or upload my audio?
- No. The audio is analysed in your browser and is never recorded, stored or uploaded. You can revoke microphone access at any time from the lock icon in the address bar.