Hi there,
I'm curious if is it possible to use a piezo buzzer as a microphone? I don't want to record human voice but muscle noise.
Recently I saw an article about a new technology which allows you to wirelessly control a device using your hand. The device records the noise of the muscles then using some algorithm it figures out what did you do with your hand. I'm curious if this concept would work "at home".
You can read more about this wireless device here: https://www.thalmic.com/en/myo/
What do you guys think?
I'm curious if is it possible to use a piezo buzzer as a microphone?
Yes. I'm not sure about a "buzzer", but a piezo transducer is is often used as a microphone in [u]Acoustic guitar pickups[/u].
A "buzzer" usually has a circuit built-in to generate sound when you apply DC voltage. That circuit would-screw-up using it as a microphone. A Piezo transducer/speaker will work "in reverse" generate a signal when it's flexed/vibrated. (A regular dynamic speaker with a magnet and voice coil can also be used as a low-quality microphone, and a dynamic microphone will work as a tiny speaker.)
Picking up useful muscle noise, ignoring other noise, and "decoding" those signals.... I have no idea...
I wasn't aware of another piezo component "pieso transcuder". Now it makes more sense 
I've already ordered some of them.
I've also found a document about this topic which was written back in 2002.
You can find it here: http://files.grouplens.org/papers/p724-amento.pdf
Do you think an Arduino would be sufficient to classify the input signals and determine the gestures? Or would a stronger MCU be better?
1 Like
Normally muscle noise is not acoustic but electrical picked up by electrodes. I don't think there is any acoustic component to muscle noise
That is an interesting paper, but the authors don't seem to have followed it up.
You can certainly collect the acoustic data using the Arduino and a piezo element. However the authors used Matlab to characterize the various gestures and even though the various gestures produce quite different responses, uniquely identifying them in some repeatable way is probably the hard part. Note also in the figure below, the authors collected 6 seconds of audio data at 8,000 samples/sec, which requires at least 48K bytes of storage for each sample. One second's worth of data might suffice for a single gesture, but for analysis and comparison with other data sets using an Arduino, you would have to compress each data set on the fly to something much, much smaller. You might try contacting the authors for suggestions, Terveen is now at http://www-users.cs.umn.edu/~terveen/
