ESP32 Forum

We just try to realize a project with an esp32-s3 platform to classify sound events. We are using the https://github.com/espressif/tflite-micro-esp-examples repo which working fine and compiles on idf master also using ai acceleration of the esp32-s3

However we need a framework for setting up and learning the cnn model and converting to tflite format. edge impulse generally is able to do this job but they do not support latest idf master (stuck on 4.4).

So the question is if anybody knows an alternative framework for this job. We know the way Ali has described here (https://blog.espressif.com/hand-gesture ... 6d7e13fd37) but it's a bit too sophisticated for guys like us not being data scientists.

Any suggestion is appreciated.

TeeVee wrote: ↑

Tue Mar 07, 2023 12:30 pm

We just try to realize a project with an esp32-s3 platform to classify sound events. We are using the https://github.com/espressif/tflite-micro-esp-examples repo which working fine and compiles on idf master also using ai acceleration of the esp32-s3

However we need a framework for setting up and learning the cnn model and converting to tflite format. edge impulse generally is able to do this job but they do not support latest idf master (stuck on 4.4).

So the question is if anybody knows an alternative framework for this job. We know the way Ali has described here (https://blog.espressif.com/hand-gesture ... 6d7e13fd37) but it's a bit too sophisticated for guys like us not being data scientists.

Any suggestion is appreciated.

It sounds like you are looking for an alternative framework that can help you set up and learn a CNN model and convert it to tflite format for your project using esp32-s3 platform to classify sound events as Edge Impulse does not support the latest idf master.

An alternative framework that you could consider is TensorFlow. TensorFlow is an open-source machine learning framework that provides tools to build and deploy machine learning models, including CNNs. TensorFlow lite is a version of TensorFlow, which is designed to run on small devices such as microcontrollers.

You can use TensorFlow to create and train a CNN model, and then convert it to a tflite format that can run on the esp32-s3 platform. TensorFlow provides many resources, including tutorials and documentations, to help you get started. You can also find many examples of CNN models that use TensorFlow on GitHub.

I hope this helps answer your question. If you need further information on this or have any other questions, please feel free to ask.


"AskRudi"