From psychoacoustics to deep learning: learning low-level processing of sound with neural networks