Every microphone sends something, even when you say nothing. Fans, traffic, the mic’s own hiss and a nearby charger’s hum add up to a steady bed of sound. That bed is the noise floor: the level your mic sends into a call when nobody is talking.
It matters twice: the people on the call hear it in every pause, and your voice has to stand clear of it to sound clean.
The unit: dBFS
Digital audio has a hard ceiling: the largest value a sample can hold. Levels are measured down from that ceiling in decibels relative to full scale, written dBFS. Full scale is 0 dBFS, so every real reading is negative, and a bigger negative number is quieter. dBFS, explained goes into the scale in more detail.
Decibels are logarithmic. Every 6 dB is about half or double the signal level, and every 20 dB is a factor of ten. A floor of −40 dBFS is ten times the level of a floor of −60 dBFS, not a little more.
For a sense of scale, the WebRTC audio standard that browser calls follow asks software to bring active speech to a level that works out to about −22 dBFS on the scale this site uses. RFC 7874 gives it as an RMS value of 2600 on a 16-bit scale whose ceiling is 32767. A floor at −60 dBFS sits roughly 40 dB below that.
What the test reads
The noise test asks for four seconds of quiet. It cuts them into 100-millisecond frames, takes the root-mean-square (RMS) level of each, an average of the signal’s power, and reports the median frame as your floor, so a single cough or click does not count. These are the bands it uses. They are our rule of thumb, not a published standard:
| Noise floor | What others get in your pauses | Test verdict |
|---|---|---|
| −60 dBFS or lower | Nothing much reaches the call | Good |
| −45 to −60 dBFS | Some noise; suppression will help | Borderline |
| Above −45 dBFS | Noise that competes with your voice | Noisy |
If you compare with another meter: the test reads plain RMS, so a full-scale sine wave reads about −3 dBFS here. Meters that scale a full-scale sine to 0 read up to 3 dB higher.
What mics add on their own
Every mic has some self-noise, the hiss of its own electronics. Makers rarely publish it in dBFS, and headset and USB mic specs use different units, so the reliable number is the one you measure.
One published target gives a sense of scale. For mic arrays built into devices, Microsoft recommends a sensitivity of −26 dBFS for a 94 dB SPL tone (sound pressure level, how loud a sound is in the air) and a signal-to-noise ratio of at least 64 dB. A mic built to that hisses at about −90 dBFS on its own (−26 minus 64), some 30 dB below the test’s good line.
So when your floor reads −50 dBFS, the mic’s own hiss is rarely the reason. The room is, or the gain that lifts it. Mic type still changes how much of the room you get: RØDE notes that condenser mics are very sensitive and pick up more background noise than dynamic mics.
Why the floor is higher than you expect
- Distance. Your voice gets quieter the further you are from the mic: in open space, Audio-Technica gives about 6 dB for every doubling of distance. Moving the mic closer raises your voice far more than it raises a noise source across the room, so the gap between them grows even if the floor stays put.
- Automatic gain. Browsers and apps can adjust your level automatically to keep it steady. TeamSpeak describes automatic gain as lowering a strong signal and raising a weak one, and in your pauses the weakest signal is the room. The browser reports whether it applied auto gain, and the test shows you what it said.
- Mains hum. Power wiring runs at 60 Hz in North America and at 50 Hz in the United Kingdom and most other countries. A laptop charger or a ground loop can put that hum into your audio, and recorded hum often carries harmonics above it. The test looks for that pattern separately, because a hum can sit at a low overall level and still be the thing people notice.
Suppression hides the floor; it doesn’t remove the cause
Noise suppression turns down sound it takes for noise. Classic filters work best on steady noise such as fans, and pushing any filter harder can take parts of your voice too. The test can record with your browser’s suppression off and then on, so you hear how much the software is doing.
If the floor is high with suppression off, fix the room or the mic first: turn off the fan, close the door, move the mic closer. Then let the app’s suppression clean up what is left.
Measuring your own
Run the test where you take calls, with the room as it normally is. Stay quiet for the first step; that gives you the floor. Read the sentence in the second step; that gives you your speech level, your signal-to-noise ratio and your peak. Then turn the noise source off and run it again. The drop in the floor is what that source was adding to every call.