Sound input - does trained bit depth need to match deployed bit depth?

Hello all,

I think I know the answer to this but would appreciate confirmation please…

We’re currently training an Edge Impulse model optimized for the ESP-EYE, which features a ‘24bit depth’ microphone, the MSM261S4030H0R.

My questions is that if the model is deployed to another ESP32 board does the microphone on this board also have to feature a ‘24bit depth’ microphone for the model to perform similarly?

I suspect it does but not certain.

Thanks for looking!



  • Your training data can come from anywhere.
  • None of the training data needs to come from the ESP.

I still like to include training data from my device though so that a robust training dataset is used yielding a powerfully built Model.


Thanks for the reply.

Perhaps I can expand on my question.

If I deploy a model/ project created for the ESP-EYE to a ESP32 board which features a microphone featuring, say a 18bit depth, how do I accommodate this change in the deployed code?

Many thanks,


(Same user as leo738)