Right now we are actually looking into a different direction:
Using I2S MEMS mics (ADC is already built in the mic) which can be directly connected via I2S to an ESP32. This one for example. That should solve all our interference problems.
The ESP32 in combination with MEMS need almost no power compare to the raspberry solution. But still strong enough to run an AI model.
We are already able to record sound. But still working on it to reduce noise.
Using I2S MEMS mics (ADC is already built in the mic) which can be directly connected via I2S to an ESP32. This one for example. That should solve all our interference problems.
The ESP32 in combination with MEMS need almost no power compare to the raspberry solution. But still strong enough to run an AI model.
We are already able to record sound. But still working on it to reduce noise.
Interesting. That microphone has a specified signal-to-noise ratio of 64 dB A-weighted at 94 dB SPL, so an equivalent acoustical noise of 30 dB(A). It should outperform the Panasonic WM-61A and WM-61B with their equivalent acoustical noise level of 32 dB(A), but be far worse than a PUI Audio AOM-5024L-HD-R with its noise level of 14 dB(A).
Then again, no-one specifies what happens in the subsonic region and the noise of electret microphones always increases a lot below 300 Hz or so. No idea if the same holds for MEMS microphones.
Then again, no-one specifies what happens in the subsonic region and the noise of electret microphones always increases a lot below 300 Hz or so. No idea if the same holds for MEMS microphones.
From what I find on the Internet, they are basically micromachined condenser microphones, and the electronics are quite conventional and will therefore also have increasing noise at low frequencies. At least that's the impression I get from this datasheet Access Denied That model rolls off below 60 Hz and is therefore not suitable for elephants anyway, but it has a nice description of what's inside on page 5 (*).
I wonder if there are any manufacturers that use a different way of reading out the time-variant capacitance that does not cause a noise increase at low frequencies. It should be possible with switched-capacitor techniques, I think.
(*) Edit: in fact you see the same at page 2 of your link, https://eu.mouser.com/datasheet/2/334/DMM_4026_B_I2S_R-1761385.pdf
I wonder if there are any manufacturers that use a different way of reading out the time-variant capacitance that does not cause a noise increase at low frequencies. It should be possible with switched-capacitor techniques, I think.
(*) Edit: in fact you see the same at page 2 of your link, https://eu.mouser.com/datasheet/2/334/DMM_4026_B_I2S_R-1761385.pdf
The model we are using (DMM-4026-B-I2S) rolls off below 30Hz. So it definitely should be good for our purpose.
Hmm. Sounds interesting.