Low noise microphone/sound level-meter

[hans]

I would like to build a sound level meter/microphone that can detect low sound levels. My motive: I've got a noisy neighbor, which keeps talking (or shouting) continuously at night, often till 7am. In my mind there are 2 possible scenarios: A) I'm overly sensitive for this sound, and I'm just BS'ing about it. B) the apartment I'm living in is very noisy (even though it's only 15 years old with modern concrete construction), and I need data to convince my land lord to do something about the noise isolation (be it at the neighbour, the walls/windows/floors/etc.).

My intuition says scenario B is more likely, since I've heard some neighbors stories about this person, however, they don't want to deal with it any further. I also can't find much common ground to reason on with this neighbor. Unfortunately, authorities have a hard time detecting the problem, since the noise is very intermittent. Also at the end of the day, the police can't forbid someone to talk at night (shouting is something else, of course).. That brings me back to quantifying the actual sound level, and if it's too high, I could bounce the ball back to my house lord to do something about it.

The noise regulations in my area set a maximum sound level of 40dBA peak and 25dBA long-term average, both indoors. The 25dBA is a figure I'm targeting to measure for a long period of time. This seems far below the noise floor of mobile phones or a studio mic (I've a Blue Snowball), probably the audio ADC's FS + dynamic range (16-bit / 96dB) can't even represent levels as low as 20 or 25dBA.

I've had a look for COTS sound level meters, and most affordable meters have a minimum of 30dBA which they can measure. The units I found that go down to 20dBA cost far upwards of 1k$.  I could also consult a company to do sound measurements.. but considering the prolonged night time and intermittency (they could setup their sound equipment for the night, and then it so happens the neighbor is quiet that night).. I think that would be very expensive.

In case my regular options of dealing with the noise are exhausted, I'm looking if I can measure sound levels myself. If possible I would also like to record the sound so I can 'show' what kind of noise I'm talking about. My plan would be to 'calibrate' the DIY mic with a COTS sound level meter. For example, put them close together pointing at the same sound source, and then note down how much -xx dBFS equates to yy dBA on the COTS sound level meter. I assume the scaling should be fairly linear, assuming neither sensors are operated close to their noise floor, of course.

To build something quickly I'm intuitively drawn to digital MEMS microphones and interface them with e.g. a STM32F4 discovery, since that board has USB microphone code examples readily available. I found the S2GOMEMSMICIM69DTOBO1 evaluation board from Infineon, which outputs I2S so should be an almost "drop in" replacement to the onboard MEMS mic of the F4 Discovery.

That eval board has 2x IM69D130 MEMS mics plus a PDM to I2S converter. The mics should have a SNR of 69dB @ 1kHz. If I'm correct, assuming I can get the mic to operate at this spec, that should put the noise floor right at 25dBA. I imagine that if I would get 2 of these boards on 2 I2s inputs, average all 4 mics together to a single sound stream, I could increase the SNR by 3dB, putting the noise floor at 22dBA.

I must confess I'm quite ignorant on audio gear quality, therefore I'm not very confident when the noisefloor is good enough.  22dBA noise floor to detect 25dBA noise seems a bit sketchy at best. I could average out more mics, as demonstrated in this old Appnote from Analog Devices (note: not interested in the beamforming/constant group delay circuit of that appnote). However, going up to 16 digital mics is getting a tad silly.. In that case I would be better off with (multiple) analog condenser/MEMS mic and an I2S ADC. E.g. the ICS-40730  should have a 74dB SNR. If I get 4 of those averaged (similarly done as the appnote shows), that should put the sound noisefloor at 17dBA.. assuming a noise-free amplifier stage (which doesn't exist, and a NF of 3dB seems also very low for one).

Either way, perhaps the next problem is also how to power the circuit. PCs are noisy.. USB is noisy.. MCUs are noisy. So I assume I won't achieve the promised SNR figures on those noisy power rails. I suppose operating on battery power instead of USB is needed (just like the Analog appnote), perhaps even 2 battery banks for analog and digital parts. However without USB I would need to log audio data to a WAV file on a SD card instead (24-bit mono @ 48kHz = 141kiB/s of data, which would fit 12+ hrs of data on a 8GB SD card).

Another problem I foresee is the accuracy of the calibration. I'm sure that the dBFS <--> dBA mapping would stay fairly linear with a specific calibration sound or tone, however, I'm unsure how much error is introduced for other sound profiles due to the A-weighting. I assume the mic output data is proportional to SPL, and that the A-weighting needs to be applied in order to output a meaningful dBA measurement. At least, this is the qualitative reasoning I'm currently at, however I haven't found much quantitative work on this concept (e.g. design examples/books/articles) that actually showcase the A-weighting and dBA measurement in action. Does anyone happen to know where this is showcased?

Can anyone fault my reasoning or have additional/better ideas? Or perhaps other meters I haven't stumbled upon that do the job but don't break the bank..

[jonpaul]

Radio Shack meter $20.

Levels below 40 dbA are not easy to measure nor significant for noise ordinances or complaints.

We use GenRad 1983 meters with 30-130 dB calibrated


Jon

[HB9EVI]

if your neighbour keeps you from sleeping by night, call the police, file a report and announce that to your landlord.

this isn't a technical problem. I don't know the local laws, but recording others without their consent could cause you trouble, even if the person is yelling it all around the house.

[hans]

I've called them and that hasn't helped so far, because they don't hear it outside.

W.r.t. privacy concerns, good question.. hasn't occurred to me directly. Personally I can't comprehend what is being said, as it seems only low frequency noises protrude. Nonetheless, the problem potentially still stands.

[HB9EVI]

who's talking about outside; if somebody complains about loud neighbours, they go to ring there, ask them to stop the music/to stop yelling or whatever and file a report. no idea what incapable police you have there, if they don't do their duty. at least here it works like that.

it's not turning about who can understand what on a recording (what would be the sense to record somebody without being able to identify the person after all); recording others is not without problems; what do you think why the surveillance cams in the streets just record picture but no sound.

[tszaboo]

UNI-T makes a relatively cheap sound level meter. BTW, I think one never can settle with noisy neighbors. Owning a shotgun is illegal here, so probably best to get used to it, move, or make them move.

[hans]

@HB9EVI The police are not the call-in thugs that will instantly fight your case without any investigation of their own. Honestly I would be worried if it's so easy to get someone in trouble with just an one-way report.  So far they haven't been able to hear anything outside, a few times out of coincidence, but perhaps also because there are different rules for outside and inside noise (and I sleep inside, so there's that).
Next time I will probably try to ask if they can listen inside my house as well..

I've been asked to provide an audio recording of the problem, but the sound level is slightly above whisper level, and therefore it's hard to record with a microphone that's intended to be close to your face (e.g. studio or mobile phone mics). But if your concerns about privacy are correct, and I'm to believe they are since recording laws are quite strict, then I'm also not allowed to what I'm requested by my land lord.

I've seen that Uni-T meter I think. I can try it first to see if 30dB lower bound is sufficient. As I said, I perceive it as whisper level or slightly louder, but due to the limited frequency content it's inaudible. And in any case it's kind of a gadget to have, for if I don't proceed with this project..

[CaptDon]

Not exactly sure of the legal ramifications here, however, the sound is in 'your' house so it belongs to you. You have not bugged premises not owned by you so really you should be in the clear. Anyway, there is a great program called ScannerRecorder. It will record on your laptap including timestamp. You can set up a 'trigger' threshold above which the recorder activates. It will take some correlation to match the amplitude of the recording to absolute SPL. Cheers. B.T.W., there are many acoustic ways to drive your neighbor nuts before he has the same effect on you.

[TimFox]

According to one source, 40 dB(A) is the noise level in a quiet library.  You may have trouble convincing a disinterested observer of the nuisance, even if a calibrated meter gives you above 30 dB(A).
My antique General Radio 1565-B Sound-Level Meter, ca. 1971, is a typical instrument for sound or noise measurement, but was designed for occupational safety measurement rather than nuisance abatement.  Therefore, it is only usable from 40 to 140 dB(A).

[tszaboo]

I've seen that Uni-T meter I think. I can try it first to see if 30dB lower bound is sufficient. As I said, I perceive it as whisper level or slightly louder, but due to the limited frequency content it's inaudible. And in any case it's kind of a gadget to have, for if I don't proceed with this project..

I try to find mine tomorrow, and see what the noise floor is.

[Conrad Hoffman]

Various thoughts, probably mostly useless. 1) You could easily be bothered by very low level sounds at night, probably under 25 dBA. Those levels are difficult to measure. 2) A-weighting rolls off the low end to supposedly reflect how people hear, but it's well established that it's a very flawed weighting. It's just been the standard for so long that we're stuck with it. Your apartment likely has lots of LF noise like air handling, traffic and such, that needs to be rolled off, but you may need 1/3 octave or FFT display to separate the offending noise. I'd record and  use something like Audacity to analyze it. 3) You can certainly calibrate quite well using the method you describe, but only for single tones or narrow noise bands. There's lots of info online about SPL meters and calibration. 4) MEMs mics and inexpensive electret cartridges are quite good, but noise at low levels is usually their downfall. (See "build a measurement microphone" type articles.) Larger diameter studio mics usually do better, unless you spend crazy money. The 1/2" electret for our SPL meter at work is a simple (and fragile) un-amplified B&K module for about $1100. Then you need the meter to attach it to.

[Terry Bites]

You need to convince your local Authorities that the noise excededs thier theshhold for nuiscance and get thier advice. Dont argue tech stuff with your landlord. Your tests mean zip to them.

In the UK we are entitled to the "quiet enjoymnent of our homes."

When the say quiet the mean a quiet lilfe not a silent one. Log your noise problems

You could start chcucking down secondhand carpet tiles, making false ceilings etc. Then you'll be moving on to tin foil hats, sniper chic and night vision insanity.

[Conrad Hoffman]

Remember too, it can be easier to get rid of the complainant, rather than fixing the walls or building or getting people to keep the same hours you do.