I haven’t got the chance to play with Edge Impulse yet (waiting on my training data) but as far as I understand then one could use Edge Impulse on an esp32 through Arduino IDE and I was wondering if I could do the same with my nRF51822 board since it would be the perfect size for my project.
In theory, should be possible as long as you got the core Arduino IDE support. This has certainly been verified with the ESP32, but I’m not sure if it has with the nRF51822. Another option would be to export as a C++ library and integrate that into your project. Let us know how you get along, but it might be worthwhile to just avoid any potential headaches and go with a more recent device from Nordic like the nRF52833 for example. Here’s an example of open source hardware that supports it: https://ruuvi.com/ruuvitag-specs/
Hi @tostuk, it really depends on what you want to do. For audio classification the nRF51 is much too small, but anything with low-resolution sensor data (anything under a 100Hz that doesn’t require lots of preprocessing) will work quite well as long as you keep an eye on the RAM usage (which is shown in Edge Impulse at every DSP and ML step). Some guidance here on model sizes: https://docs.edgeimpulse.com/docs/inference-performance-metrics
Easiest way is just using the data forwarder to get data off the device, and then use the example sketch on that page to run inference.