With Edge Impulse, I was able to build a machine learning model, which can distinguish between spells like “Lumos” and “Nox” and general noise. Because I really enjoyed this side project, I wrote a little article on my website on: https://jan-fiedler.net/tinyml-audio-recognition/
This may be helpful for anybody, who is also looking for an approach to TinyML audio recognition.
I ended up building a fully embedded wand. Check the article, if you would like to see it in action
We’ve added some new things this week to make life easier for building these types of models by the way, you can now click on the three dots and select “Split sample” to extract just the keywords. We’ll also be adding ways to automatically add noise to datasets in the near future. But already fantastic work!
