How to run Emlearn model on ESP32s3 based custom board

[ujur007]

I have ESP32s3 based module and PCB that has microphone and IMU sensors. I want to try emlearn micropython on this board.
I do not have much knowledge of emlearn. But I can learn it.
Just if someone can point out to me some real world example of doing some keyword detection or some motion based ML model using emlearn, it will be helpful.
If there is no such example, then please point out the method or ways to process further in this direction.

[jonnor]

Hi @ujur007

There is audio processing example here - only DSP, no trained models - https://github.com/emlearn/emlearn-micropython/tree/master/examples/soundlevel_iir

There is example for IMU (accelerometer) here, using machine learning, https://github.com/emlearn/emlearn-micropython/tree/master/examples/har_trees

For more information, see the documentation: https://emlearn-micropython.readthedocs.io/en/latest/user_guide.html

[ujur007]

Ok. For the IMU example, is there a specific format requirement? Or how to collect this data that the emLearn understands.? I have the sensor data that I can store in files. such as .csv, text, etc.
I do not have the exact same hardware in hand.

[jonnor]

@ujur007 there is data collection support in har_trees. It stores into .npy files. This is then converted into a dataset (in .parquet files) using the .py tools also provided there. You can also convert other data to the same .parquet format.