How to know what format of data input is required for the generated model

#1

I want to deploy a model of audio classification on stm32.
So I generated a tflite model in .lite format via Edge Impulse , then used this file to generate a project file for keil compilation and flashing through cubemx, and wanted to deploy the model to stm32 through keil.
But I don’t know what format the model needs for the input data at this point

#2

Hi @WScarlett,

How are you importing the .lite file into your project? .lite files are not intended to be used as standalone models without a runtime interpreter (which is usually TensorFlow Lite).

Are you set on using the .lite file? If not, I might recommend exporting the C++ library from Edge Impulse instead, as that should be easier to work with. You can feed it raw audio data to have it perform inference. Here is an example of a keyword spotting demo: GitHub - ShawnHymel/ei-keyword-spotting. If you look in the ebedded-demos folder, there are some STM32 examples.