Training Watson To Be Your Musical Muse

Share this post:

Watson already helps chefs and home cooks alike come up with never-before-eaten recipes. It also assists artists and helped one color coordinate a mural, and another make a cognitive dress for the Met Gala in New York. Now researchers are working to make the system capable of collaborating with musicians to produce original songs and scores.

Watson Beat is a cognitive cloud-based app being developed by machine learning and artificial intelligence expert Janani Mukundan, a Dire Straits fan with no musical training but a PhD in computer engineering, and Richard Daskas, a composer and professional musician.

Janani Mukandan (L) and Richard Daskas ready to demo Watson Beat in the lab.

Together, they have taught Watson about the specifics of rhythm, pitch, and instrumentation, as well as the differences in genres. All of it combines in algorithms running through Watson’s neural network to help artists create a original compositions.

“Watson Beat knows what sounds good to us. And based on input, it knows, for example that if a user wants a ‘dark’ or ‘gloomy’ song, that it should use a minor key,” Janani said.

Listen and learn

Watson Beat then keeps learning from its users. In as few as eight bars or 15 seconds of input, it can churn out a multi-instrument, minutes-long track that can be adjusted according to the mood and genre any budding Beethoven chooses. Here is a sample of an original piece by Richard:

Richard only wrote it for, and played it on the piano. But he decided to amp it up – give it an up-tempo sound with a bass line, drum kit, and a Middle Eastern beat. With just the sample above, here is what Watson Beat banged out:

“Watson Beat can understand MIDI (file) inputs of entire songs – even multiple songs – or individual instruments. Or, like the sample above, it can comprehend live play. I plugged a keyboard into my laptop and ran the app as I played,” Richard said.

Richard determined the upbeat style of the song, as well as the variation of the additional instruments, by introducing perturbations into Watson Beat’s neural network. These adjustments, done via an app, tell the system how similar or different the collaborative composition should be, compared to the original. This sliding scale of cognitive computing influence can come in especially handy for those whose last piano lesson stopped with Mary Had a Little Lamb!

The sound of opportunity

Watson Beat, like its smart app-in-the-cloud brethren, collaborates. It can help a musician get over writer’s block or remove hidden biases in genre or style by introducing something quite literally unheard of before. It might also be used in a music therapy setting.

For now, Janani and Richard’s creation is still a proof of concept. Attendees at SXSW, Tribeca, and Moogfest had a first-hand opportunity to listen to and even try creating their own music. Longer term, the pair wants to offer Watson Beat as an app for the public.

Listen to more samples on our soundcloud playlist.