Justin Svegliato
Jul 23, 2017 · 1 min read

Great questions!

To answer your first two questions, if we specifically train our network on (1) a composer or (2) a genre, it would learn the patterns tailored to either one. When we first ran our tests, we only used music from one composer. I think it was Bach. The resulting music was very Bach-like.

Your third question is more complex. It depends on if our network can capture the elements of that genre. While I’m no jazz expert, it often uses swing and syncopation. Although our network can learn swing, it can’t understand syncopation since the input lacks the features that reflect emphasis or accent. My guess is that our current architecture only lends itself to specific genres.

    Justin Svegliato

    Written by

    CS PhD Student at UMass Amherst. Former software developer on Wall Street. I build intelligent systems that reason and learn.

    Welcome to a place where words matter. On Medium, smart voices and original ideas take center stage - with no ads in sight. Watch
    Follow all the topics you care about, and we’ll deliver the best stories for you to your homepage and inbox. Explore
    Get unlimited access to the best stories on Medium — and support writers while you’re at it. Just $5/month. Upgrade