i have set up a project for the MAIX GO that displays the STFT transform of the signal acquired by the onboard microphone.

It works well with a few minor issues. I have trouble to understand the data organization at the output of the FFT and as a consequence when i perform a 512 points FFT i expect to have a range of frequency of 0 to FS/2 in the first 257 FFT samples. In my case i get 0 to FS/2 in the fist 129 samples then the mirrored version of the spectrum. I’a not sure if the problem is display related or if the FFT samples organization is incorrect.

The long-term goal is to use the STFT to classify audio samples using a neural net.


Jonathan Piat