Hard to point at one thing! I started with a deep learning for artists course, and adapted the stuff I saw there to an audio mini project. And then started reading the big papers in neural networks independently.

I found it helpful to read the source material (eg, ml text books and academic papers), paired with blog entries (which often have a good basic idea, but skip out get wrong some details), and actually building things for my own interests.

On the work side, I'm doing low bit rate speech coding, which is good fun.