Measuring a microphone’s self-noise without an anechoic chamber
A studio condenser microphone hisses a little on its own. That hiss, its self-noise, is the number in the datasheet: 5 dB(A) for a RØDE NT1-A. Measuring it at home is harder than it looks, because the room is almost always louder than the microphone. This page is what we learned trying to measure it for the Talas One, an open-hardware dual-diaphragm microphone: what failed, what works, and the protocol for the next attempt.
In short. In a bedroom at night you only measure the room. A dummy capsule gives the noise of the electronics alone: about 5 dB(A) on our channel A, three takes within 0.1 dB. The figure for the complete microphone needs a place quieter than the microphone, and a second microphone as a witness. That measurement is not done yet; this page will carry it when it is.
1Why the room wins
Noise from the room adds to the microphone’s own. The rule given by practitioners is a background 8 to 10 dB below the microphone’s noise (recording.org). An ordinary home sits around 25–40 dB(A). To read 5 dB(A) honestly you would need a room near −5 dB(A): no home gets there.
Two consequences follow, and they organise everything below:
- Room noise only adds. A reading is always an upper bound. If it is already at or below your target, the target is met, however loud the room was.
- A known microphone is the witness. Put a microphone whose self-noise is published next to yours. If it reads close to its own specification, the place was quiet enough and your figure stands. If it reads well above, you measured the place.
“Anechoic” means without echo, not silent. A semi-anechoic room can still be too loud for this: one university room states a background below 17.5 dB(A). What to ask any lab is its background noise in dB(A).
2What failed: the bedroom at night
On 8 November 2026 we put the Talas One P3 and the NT1-A side by side in a bedroom at night, gains matched within 0.3 dB. Our measurement bench refused the run. Its guard looks at the coherence between the two microphones: if they hear the same thing, they are hearing the room, not themselves. Coherence was 0.20 against a limit of 0.05, in every band from 20 Hz to 20 kHz.
A-weighted, the two microphones read within 0.6 dB of each other, 11 dB above the interface’s own floor. That is still worth something: in the same room, the P3 is no noisier than an NT1-A. But it is not a self-noise figure.
3What works: a dummy capsule
The electronics can be measured without any sound at all. Unsolder one diaphragm’s wire, insulate it, and solder in its place a capacitor of the same value as the diaphragm. The preamp then sees exactly the impedance it sees in use, but nothing acoustic reaches it. Whatever comes out is the noise of the electronics.
- The dummy: five 220 pF C0G capacitors in series, 44 pF, close to the 41.1 pF per diaphragm of our capsule. C0G because it is stable and does not add noise of its own.
- The body closed, exactly as in use: the metal body is the shield of the input (section 4).
- Conversion to dB(A): the output noise, A-weighted, divided by the microphone’s sensitivity. Our sensitivity, about 55 mV/Pa, is taken relative to the NT1-A’s published figure, so every number below carries about ±2 dB.
| Channel A, 44 pF dummy | 11 November | 13 November (three takes) |
|---|---|---|
| Everything, A-weighted | 23.2 dB(A) | 5.1–5.2 dB(A) |
| Share of 100 Hz–5 kHz power on mains lines | 92 % | 12 % |
| Mains lines removed | ≈ 13.4 dB(A) | 5.6–5.7 dB(A) |
The first figure was mostly mains hum: 92 % of the power sat on the lines of 50 Hz and its harmonics. Two faults explained it, a body whose parts touched badly and three wrong capacitors on the XLR plugs; they are the subject of our page on hum. Once fixed, the same measurement gave 5.1–5.2 dB(A), repeatable within 0.1 dB, and 3 dB below what our simulation predicted.
What this figure is not: the self-noise of the microphone. A real capsule adds its own noise, and our second channel has not been measured with a dummy yet.
4The complete microphone: the options
| Method | What it gives |
|---|---|
| Dummy capsule | The electronics alone, ±2 dB |
| Side by side with a reference, at night | An upper bound |
| Wardrobe, blankets | A lower upper bound: textiles stop the highs, not the city’s rumble |
| Underground cellar, box of clothes | Possibly a valid figure: earth and rock stop the low end that textiles cannot |
| Steel pot with a tight lid, in a box of blankets | A good upper bound, possibly a figure |
| Add a factory-calibrated measurement microphone | Removes the ±2 dB of a relative sensitivity |
| Massive steel enclosure, as manufacturers use | The figure |
| University anechoic chamber with a quiet background, or a testing lab | The figure, with a report |
A 94 dB acoustic calibrator is often suggested. It does not fit here: it is made to seal on a ½″ or ¼″ measurement microphone, not on a 34 mm capsule behind a grille.
5Our next attempt: a rock wine cellar
A cellar dug in rock, about a metre underground, is the best free place we have: the earth is the mass that stops traffic rumble. Wine cellars, though, run at 70–85 % relative humidity, and a condenser capsule with a 1 GΩ input node does not like moisture: leakage shows as crackles and a rising low end.
- Two nested boxes. Inside, an airtight plastic box with 300–500 g of silica gel and both microphones side by side, cables out through one slot sealed with putty. Outside, a cardboard box full of clothes, against noise. Cardboard and clothes alone would soak up the cellar’s moisture and hold it against the microphones.
- Start only when the humidity inside the box has stayed below about 60 % for 30 minutes, with the hygrometer inside the box.
- The interface and a laptop on battery stay outside, at the end of the cables: a laptop fan is louder than the microphones.
- The plateau test. Add a layer of clothes, close the door, stop the cooling unit if possible, and measure again. While both levels keep falling, you are measuring the place. When they stop falling, each microphone has reached its own noise.
- Coming back up, leave the box closed for an hour so the microphones warm up dry.
| The NT1-A (5 dB(A)) reads | The microphone under test reads | Conclusion |
|---|---|---|
| 2–8 dB(A) | anything | A valid figure |
| above 8 | at or below the target | Target met (an upper bound below it) |
| above 8 | above the target, same as the NT1-A | Inconclusive: the place is too loud |
| above 8 | clearly above the NT1-A | The microphone is noisier than the place: check humidity, then the capsule |
6The tools
Everything above runs on mic-test-bench, our measurement bench, free software under AGPL-3.0: bruit_fausse_capsule.py for the dummy capsule, campagne_bruit.sh for a side-by-side self-noise run, with the coherence guard that refused our bedroom. The full protocol, with the sheet to fill in on the spot, is in the Talas One repository, and every measurement is in measurements-p3.md.