Your voice never reaches a call alone. Whatever the mic hears around you comes with it. The signal-to-noise ratio, or SNR, measures how far your voice stands above that noise, in decibels (dB).
The formula
When both levels are in dBFS (decibels relative to full scale, the unit the noise test uses), SNR is a subtraction:
SNR = speech level − noise floor
Speech at −22 dBFS over a floor at −62 dBFS gives an SNR of 40 dB. Underneath, it is 20 × log₁₀ of the ratio between the two signal levels, so 40 dB means your voice is 100 times the level of the noise, and 20 dB means 10 times.
The noise test measures both halves. The floor is the median level of four seconds of quiet, read in 100-millisecond frames. Speech is the loud end of your normal talking: of the frames that sit more than 10 dB above the floor, the test takes the level that 90% of them fall below. That leaves out the pauses between words without being thrown by a single shout.
What 20, 30 and 40 dB mean
There is no single published standard for SNR on a call, so the test uses bands of its own. The middle column is plain arithmetic; the verdicts are our rule of thumb:
| SNR | Noise level compared with your voice | Test verdict |
|---|---|---|
| 40 dB | One hundredth | Clean |
| 30 dB | About a thirtieth | Clean |
| 20 dB | One tenth | Usable; a closer mic would make it cleaner |
| 10 dB | About a third | Poor; listeners strain to pick out your voice |
For comparison, one professional body does publish a figure for rooms where people need to follow speech. The American Speech-Language-Hearing Association’s acoustic standard for classrooms says the signal-to-noise ratio should be no lower than +15 dB. The test’s 20 dB line sits a little above that.
The easiest way to raise it
SNR has two halves, so you can raise it from either side. Lowering the noise means the room: the fan, the door, the TV. Raising your voice at the mic is usually easier. Audio-Technica gives about 6 dB more level for every halving of distance in open space, so moving the mic from 20 cm to 10 cm from your mouth can add up to 6 dB to your speech, while a noise source across the room barely changes.
Turning up the gain doesn’t help. Gain multiplies everything the mic hears, so it lifts the noise by exactly as much as your voice, and the SNR stays where it was.
What suppression does to SNR
Noise suppression works on the bottom half. It turns down sound it takes for noise, so the floor in your pauses drops and the measured SNR rises, sometimes by a lot. Two cautions:
- It isn’t free. Audacity’s manual warns that a more sensitive setting removes more noise but may take some of the sound you want to keep, and Discord says Krisp can lower voice quality in a quiet room with a good mic.
- It works best with a gap to start with. Audacity says noise reduction works well when the voice is much louder than the noise, and can leave the result too distorted when it isn’t.
Zoom and Teams both offer stronger levels for louder rooms. Zoom’s High removes background speech and may use more CPU, and in the Teams iPhone app High suppresses all background sound that isn’t speech. Use them when the room calls for it, not by default.
The noise test can compare your browser’s suppression off and on and play both clips back. The difference in the floor is what the filter does for you; listening to the clips tells you what it costs your voice.
Mic specs are a different SNR
Mic makers quote an SNR too, and it means something else: the gap between a standard test tone and the mic’s own self-noise. For the mic parts in a speech device, Microsoft recommends at least 65 dB, measured with a 94 dB SPL tone at 1 kHz. That number describes the hardware in a silent room. The SNR the noise test reports describes your voice in your room, which is the one your listeners hear.