Can a machine create chorales in the same style of Bach? By the sound of what a team of French researchers have created, the answer is, in a good proportion of the time. The team has developed a neural network that has learned to produce choral cantatas in the style of the German composer, and their testing backs up the validity of the arrangements.
Artificial intelligence researchers have used deep learning techniques to create an algorithm that can write original music in the style of 17th-century German composer Johann Sebastian Bach. The work has been published online. Gaetan Hadjeres and Francois Pachet at the Sony Computer Science Laboratories in Paris call their artificial intelligence agent “DeepBach.”
The researchers work out the same facility that was behind another project this year that used artificial intelligence to create new music in the style of The Beatles.
Hadjeres and Pachet trained the AI using data from over 400 of Bach’s chorale compositions. Using neural networks, DeepBach can not only compose its own music, but also can take existing songs and re-harmonize them in Bach’s style. DeepBach can apparently do that so quickly that the researchers say it can actually be used in interactive music compositions, changing up music on the fly as needed.
"This model is capable of producing convincing chorales, even if it is trained with no other data that the 400 chorale sheets by Bach.""We developed a model of polyphonic music generation, which learns to compose chorales in the style of Bach. This model is capable of producing convincing chorales, even if it is trained with no other data that the 400 chorale sheets by Bach."
The researchers tested the success of their program by playing DeepBach compositions and asking listeners whether or not the piece was genuine. You can try the test here. (We were fooled on 30% of the answers.)
Hadjeres and Pachet claim that, DeepBach had a roughly 50 percent success rate for this test, and they noted that the percentage for actual Bach compositions was only 75 percent. While there is clearly room for improvement, the researchers believe that DeepBach demonstrates the possibilities for AI-generated music, and they hope to apply the same technique to other composers and music genres.
"We now plan to develop a music sheet graphical editor on top of the music21 toolkit in order to make interactive composition using DeepBach easier," conclude the reasearchers. "This method is not only applicable to Bach chorales but embraces a wide range of polyphonic chorale music, from Palestrina to Take 6."