Researchers at the University of Washington have released the latest music resource to aid machine learning.

A research team from the University of Washington has released a new large-scale music dataset, entitled MusicNet, to be used as a resource for machine learning methods in music research. It is believed this research will have significant ramifications for computer generated composition, note prediction and automated music transcription.

The publically available dataset comprises 330 freely-licensed recordings of classical music by ten different composers and written for 11 instruments with over one million annotated ‘labels’. The labels indicate the precise timing of each note in every recording, the instrument that plays each note and the note’s position in the metrical structure of the composition.

“At a high level, we’re interested in what makes music appealing to the ears, how we can better understand composition, or the essence of what makes Bach sound like Bach,” said Sham Kakade in an article published on UW’s website. Kakade is an Associate Professor in UW’s Statistics and Computer Science departments and Adjunct Professor in Electrical Engineering. “It can also help enable practical applications that remain challenging, like automatic transcription of a live performance into a written score,” he says. “We hope MusicNet can spur creativity and practical advances in the fields of machine learning and music composition in many ways”.

The dataset, which has been presented in an article published in the journal arXiv, represents a step forward in music analysis. “The music research community has been working for decades on hand-crafting sophisticated audio features for music analysis,” said Zaid Harchaoui, an Assistant Professor of Statistics at UW. “We built MusicNet to give researchers a large labeled dataset to automatically learn more expressive audio features, which show potential to radically change the state-of-the-art for a wide range of music analysis tasks.”

“An enormous amount of the excitement around artificial intelligence in the last five years has been driven by supervised learning with really big datasets, but it hasn’t been obvious how to label music,” said the lead author of the study John Thickstun, a computer science and engineering doctoral student. “You need to be able to say from 3 seconds and 50 milliseconds to 78 milliseconds, this instrument is playing an A. But that’s impractical or impossible for even an expert musician to track with that degree of accuracy.”

Thickstun hopes that the practical possibilities made available by this data set will lead to greater musical possibilities. “I’m really interested in the artistic opportunities. Any composer who crafts their art with the assistance of a computer – which includes many modern musicians – could use these tools,” he said. “If the machine has a higher understanding of what they’re trying to do, that just gives the artist more power.”