[jononor]

Warning: shameless-self-promotion. For those that wish to go a bit beyond this article, I gave a presentation on the topic at EuroPython. https://www.youtube.com/watch?v=uCGROOUO_wY It explains how to build models that can make use of temporal variations and learn the feature representations based on the (Mel) spectrogram. Especially suited if you are already familiar with image-classification using Convolutional Neural Networks.