What if you train a deep neural network on the song so far, so it can generate plausible-sounding music whenever the buffer drops?

You can even hang intentionally to generate original music!

(/s, please don't)