Magent, a separate Google Brain group that studies the creativity of deep machine learning, has published the first melody that a neural network wrote.
The melody less than one and a half minutes long is reported by TechCrunch, the source code of the alpha version of the project is available in the GitHub repository.
Magenta officially started working in June 2016, but the first developments in the field of musical creativity of machines have already been demonstrated at the Moogfest music festival. Based on the five notes received, the program played a more complex and longer melody.
You can listen to the audio at this link
A program for importing music from MIDI files into the TensorFlow machine learning platform is the first project that the Magenta team is working on. In the future, the developers will refine the algorithm so that the neural network can compose more complex and longer melodies, and will also develop machine creativity in the field of image creation. The research team believes that such technologies will serve as new tools in the hands of artists and musicians.
The Magenta project is not the first attempt to apply machine learning technologies to creativity. Earlier, researchers from the University of Future Hakodate wrote a novel using artificial intelligence, which reached the final of the Hoshi Shin'ichi literary award. Also, researchers from the Google Brain laboratory, in the course of training the artificial intelligence system to understand human speech, taught it to create a semblance of poetry, reminiscent of teenage poetry.
Nikolay Vorontsov