Comment by lnsru
4 years ago
I feed fft data into machine learning system for signal classification. It’s a nice real time system with midrange Xilinx ZynQ SoC. Some Xilinx IP blocks with hand written classification engine.
I guess author wants to classify sounds in his environment too.
Correct, the ultimate goal is to perform automatic activity logging.
For example my bathroom has two IR sensors (https://www.amazon.de/dp/B08LBDYPYD), one inside the shower, one outside of it. This lets me log the time and duration of when I take a shower or how often I shave.
It also has an I2S MEMS microphone which collects the data for the FFT where the goal is to replace the IR in the shower with it since the water sound is enough information.
Also temperature and humidity, where the latter also gives a good indication of when I showered and when I forgot to close the window some time after showering.
My bedroom also has the same "setup", where the FFT shows me how big of a problem I have with snoring. Eventually I can add some logic on the server to slightly wake me up if I snore, but currently I'm just logging. The PIR serves as a measurement of how much I'm moving at night without requiring a camera for this. Though I have another ESP32 under my bed fixated at the slatted frame which contains an 9dof-IMU and logs any significant movement which may be related to quality of sleep. Also the magnetic field which is why I have a strong magnet attached to my bed tablet, so I know when it changed position. Even though the tablet knows that itself because it is logging its own accelerometer every minute.
I live alone, so all of this is no problem.
At this point I'm mostly only logging, and only using the shower and bed data to see when I used them, which I can then see in Grafana. I think the data will only gain value when watched over a long time frame.
I remember seeing a Hackaday post that uses FFT/spectogram to infer IoT things such as boiling kettle without cameras. Is this related?
No, it's not related. I'm trying to get notified when my noodle water has started boiling, but even with an humidity sensor in the kitchen it's pretty hard to tell when it's about to boil, so that the heat can be lowered before it splashes around. And the FFT in the kitchen is somewhat broken so I can't use it, there may be a lot of noise in the mains or something like that. In any case I've planned to replace the setup in the kitchen months ago, but never moved on to doing it.
1 reply →