Sound Sensor

[epierre]

@Yveaux thanks, here it is:

avr_splmeter_02b.zip

http://davidegironi.blogspot.fr/2014/02/a-simple-sound-pressure-level-meter-spl.html

[Yveaux]

@epierre Had a quick look at the code. Seems the TIMSK stuff is just for timing the ADC samples at regular interval; the TIMER1_OVF_vect sets a flag (audioget_takesample) which is scanned for in the audioget_getsamples() function.
Exactly the same behavior you can get by waiting using delay() or delayMicroseconds(), and then get the sample.
That's probably easier then getting the raw timer stuff going.

[epierre]

Since @hek is interrested, I'm revivind this post with several microphone available out there, in fact impossible to know if these use real electet ones...

@BulldogLowell I've seen you've participated here http://forum.arduino.cc/index.php?topic=208520.15 but do you think this was the good solution ? if you go through http://davidegironi.blogspot.fr/2014/02/a-simple-sound-pressure-level-meter-spl.html the questions is open...

[hek]

Hmm.. not super easy to calculate an actual sound "pollution" value or do dB calculations using this I guess.

Hmm.. Adafruit using a sample window in their example.

https://learn.adafruit.com/adafruit-microphone-amplifier-breakout/measuring-sound-levels

[hek]

How does this differ from yours @epierre ?

http://www.ebay.com/itm/Sound-Detection-Sensor-Module-Voice-Sensor-Intelligent-Vehicle-LM386-Audio-Power-/261598751181?pt=LH_DefaultDomain_0&hash=item3ce88029cd

Seems to be two chips on it..

[Moshe Livne]

@epierre I have ordered almost any kind of sound sensor on aliexpress for my washing machine sensor (at first I thought to get the beeps). They were all really bad.... I went the vibration way, but if I had to start all over again I think I would get https://www.sparkfun.com/products/9964

[Moshe Livne]

@epierre Osrry, forgot to mention that the killer was the lack of amplification on the cheap modules. The diff between silence and hand clap few cm from the mic was only a few numbers on the reading. When you have amplification it stretch it across the 1023 range, I think.

[epierre]

The electret schema is here :
http://www.edutek.ltd.uk/CBricks_Pages/Electret_Microphone.html

and IMHO the best we could use is this one:
https://www.adafruit.com/products/1063

This fully assembled and tested board comes with a 20-20KHz electret microphone soldered on. For the amplification, we use the Maxim MAX4466, an op-amp specifically designed for this delicate task! The amplifier has excellent power supply noise rejection, so this amplifier sounds really good and isn't nearly as noisy or scratchy as other mic amp breakouts we've tried!

This breakout is best used for projects such as voice changers, audio recording/sampling, and audio-reactive projects that use FFT. On the back, we include a small trimmer pot to adjust the gain. You can set the gain from 25x to 125x. That's down to be about 200mVpp (for normal speaking volume about 6" away) which is good for attaching to something that expects 'line level' input without clipping, or up to about 1Vpp, ideal for reading from a microcontroller ADC. The output is rail-to-rail so if the sounds gets loud, the output can go up to 5Vpp!

Why do we need an FFT afterward ?

Using the normal Arduino analogRead() function would be much too slow for sampling audio. Instead, a feature of the microcontroller’s analog-to-digital converter called free-run mode is utilized. This automatically takes repeated analog samples at precise intervals…about 9.6 KHz for this project, the maximum a 16 MHz Arduino can handle with 10-bit samples.

The raw audio samples are converted into a frequency spectrum using a fast Fourier transform or FFT. There are a number of Arduino FFT libraries out there, but we keep finding ourselves returning to the venerable ELM-ChaN ffft library for its speed and good looks.

The FFT output still needs a bit of massaging to make for a good presentation on the limited 8x8 matrix. Several tables of scales and weights de-emphasize certain frequency ranges as they’re reduced to just eight columns. The software works at keeping the graph interesting, but some columns will always be less lively than others, especially comparing live speech against music of varying genres. If everything seems to stick toward one end of the graph, try another musician, musical genre, or different speakers.

 Finally, because ADC registers are accessed directly, specific interrupts are used, and the FFT code is in AVR assembly language, this software will not run on upscale boards like the Arduino Due, ChipKIT or Teensy 3.0. It is strictly for “classic” Arduinos.

https://github.com/adafruit/piccolo

[Yveaux]

@epierre said:

Why do we need an FFT afterward ?

We don't 'need' an fft, it simply depends on your application.
An fft breaks down your audio signal into the different frequency components it is composed of.
This can be used to display a frequency-intensity chart, as shown on the pixel display in your picture.
Low frequency (bass) is usually shown on the left, high frequency (treble) on the right.

[epierre]

@yveaux very nice to see you again !

Although I just want to get a dB mesure out, Davide Gironi described a more complex process:

sound level meter or sound meter is an instrument which measures sound pressure level. Sound pressure level (SPL) or sound level is a logarithmic measure of the effective sound pressure of a sound relative to a reference value. It is measured in decibels (dB) above a standard reference level. The commonly used reference sound pressure in air is = 20 µPa (rms) which is usually considered the threshold of human hearing. Keep in mind that 1 pascal will equal an SPL of 94 dB. Because the frequency response of human hearing changes with amplitude, a weighting have been established for measuring sound pressure. Usually the A-weighting curve is used. A weighting curve is a graph of gain across the frequency range (10Hz to 20kHz).

To compute SPL measurements, the meters loop is:

1- collects N samples
2- do FFT for the N samples collected, the signal is now transformed in the frequency domain
3 - apply A-weighting (in freq domain)
4 - get magnitude of the signal
5- get RMS value of the signal
6 - apply a time-weight filter to RMS value
7 - compute the SPL using the RMS value
8 - output data

[Yveaux]

@epierre Don't overdo it.
I'd expect mapping the analog output of the sensor to dB will work quite well too.
You need to calibrate the output once and the resulting calibration table can be used to map sensor output to dB later on.

[epierre]

@Yveaux for calibration we have some software in the smartphone, at least they would be equal of course ;-)

[Yveaux]

@epierre said:

software in the smartphone

That will be equally inaccurate I suppose...

[epierre]

@Yveaux best that nothing to start with !

[gigaguy]

Did anyone get this working? I have the 1st mic in epierre's picture above. I am getting values from it. I am currently using the analog pin, based on the posts here. The values dont seem accurate. They dont change with sound level. I am just trying to get a noise % level.

[epierre]

Hello,

First my sketch is not good enough, I don't know why it keeps increasing the value reported...

Second, I do have results just from the raw adc with these, check the wiring again. Also be reminded some sensors above with only 3 pins may have only a digital output (may so you have to check before buying). In case it is digital, the resistor knob will serve to say above which level it wil send the HIGH signal.

[gigaguy]

@epierre thanks. I do have 4 pins, and im using the analog one. I noticed the increasing value you mentioned. I tried to get raw and got a value, but when taking into the mic, the value doesn't appear to change. Does yours?