Google continues to make advances in all of its artificial intelligence endeavors, and the most recent fruits of its labors comes from Project Magenta. Magenta comes to use from the Google Brain team and was tasked with actually creating music using a trained neural network.
Using the open-source machine learning platform TensorFlow as a basis, the Magenta team was able to turn the table on the tools that have previously been used to decipher human content in order for the machine to have its own “voice”. “We believe that the models that have worked so well in speech recognition, translation and image annotation will seed an exciting new crop of tools for art and music creation,” said Douglas Eck, one of the engineers behind Magenta.
“Artists and musicians draw our attention to one thing at the expense of another,” Eck explains. “They change their story over time—is any Beatles album exactly like another?—and there’s always some element of surprise at play. How do we capture effects like attention and surprise in a machine learning model?”
We can say with great certainty that Magenta has not advanced to the point where it can produce music that can rival the Beatles, but its first self-created song is a 90-second piece that shows some promise (listen here).
The melody was produced using four notes to start, then proceeds to crank it up with a verse and some drum beats. It’s a good first step for Magenta, and well beyond anything that Ross Gellar could possibly come up with on his own.
"With Magenta, we want to explore the other side -- developing algorithms that can learn how to generate art and music, potentially creating compelling and artistic content on their own,” adds Eck.
The developers behind Magenta are hoping to fine tune their models and tools, and distribute them via GitHub.