AI Writes Its Own Music After Listening to Classic Bach

By on

Machines have always worked systematically and methodically with creativity and art considered something that is not associated with it. However, the notion may have changed as artificial intelligence was successfully created to listen, mimic and compose to the same style of the classical era’s composer Bach.

Johann Sebastian Bach’s chorale harmonization is known to be structured with various patterns that defined the baroque music. This makes the German composer a perfect fit for Gaetan Hadjeres and Francois Pachet from Sony Computer Science Laboratories in Paris to create the deep-learning machine DeepBach.

AI Can Compose To Bach’s Style

Deep learning machines heavily rely on large amounts of data that are sifted for patterns. Meanwhile, Bach’s composition has always consisted of four parts: one melody and three harmonies consisting of the alto, tenor, and bass. The algorithmic pattern makes it easier for AI to understand it.

Its consistency grants deep neural networks to be broken down into patterns found in Bach’s 352 chorales. This can then help the deep-learning machine to generate new melodies since its goals are not to create new music but to supply the harmonies in the style of Bach to the given melody.

Creativity By Machines?

To test the effectiveness of the compositions, 1,600 people who have varying expertise in classical music – from professional musicians to music students – were made to listen to Bach and DeepBach’s arrangements. Half of them thought the DeepBach harmonies were arranged by Bach himself.

Hadjeres and Pachet consider the results a “good score” especially when considering the complexity of the compositions by Bach. And the more complex the music sounds, the more people thought it was made by Bach.

This study poses the many potentials, and limitations, that the artificial intelligence can be developed into when it comes to creativity. While it may face more challenges when confronted with music with lesser or no order, it still serves a footnote to deep learning machines’ artistry, in one way or another.


About the author

To Top