Hi @LucasCalsing,
I think that it is difficult to do keyword spotting with NanoEdge. I don't think the models are really adapted for this kind of use case, so even if you could load the data, I doubt you will have good results.
I don't see a way to avoid using big csv in your case either.
Another alternative would have been to use the ST Model Zoo services to help you train a neural network, but we do not have key word spotting models.
After a quick search, it is maybe possible to use audio event detection model for that, but this doesn't seem recommended.
I think it could be a good test if you've never used neural network, to try to use the model zoo services and see if this helps you. Maybe just do audio event detection first.
The most realistic answer here is that you have to search the internet for KWS models (key word spotting) that are very small, adapted to embedded environment, then get one and before training it, look if all its layers are supported by the st edge ai core.
If it does, you can retrain it, if it doesn't, you need another model.
Then you need to create your own application.
You can get a basic application template with STM32CubeAI Studio and then work on it to add a microphone and send output via serial for example.
In model zoo, we use the B-U585I-IOT02A and STM32n6 for audio examples, this could help you choose your hardware.
Have a good day,
Julian
In order to give better visibility on the answered topics, please click on 'Accept as Solution' on the reply which solved your issue or answered your question.
