Improving Accuracy in ESC-10 Model with MFE and Classifier

Project ID:
619008

Context/Use case:
I am currently working with Edge Impulse using the ESC-10 dataset. My model consists of an MFE block followed by a Classifier, with parameters mostly set by default in Edge Impulse. So far, I have achieved an accuracy of 80%, but I am looking for ways to improve it.

One of the main challenges I have identified is that many audio clips contain long intervals of silence, which often lead to misclassifications. I am unsure of the best approach to mitigate this issue and enhance the model’s performance.

I would appreciate any advice or suggestions on how to address this problem.

Thank you in advance for your help.

Environment:

  • Platform: Arduino Nano 33 BLE Sense