No, the data already comes in the “correct” format as a data stream from a sensor. The DNN models are trained to work on that data stream, on an edge device you usually don’t have the performance to pre process anything, except for an FFT which is a fixed function accelerator that can process the data in real time.