South SFV Today

Researchers at Stanford University are exploring the potential of artificial intelligence to reduce injury risks among professional pianists, particularly those with smaller hands. At the SIGGRAPH Asia 2024 conference on December 5, Stanford Engineering researchers introduced an AI-trained model that replicates the hand movements required for playing complex musical pieces. The aim is to identify and mitigate long-term injuries common among piano players.

Elizabeth Schumann, Billie Bennet Achilles Director of Keyboard Studies at Stanford and co-author of the research paper, highlighted the issue: “We would never expect a world-class athlete to compete with equipment that does not fit their body. Yet we ask pianists, particularly women, to adapt to a one-size-fits-all design that was never built with them in mind.” She emphasized using this research to make piano performance more sustainable.

The standard modern piano keyboard was designed in the 19th century for the average European male. However, many contemporary players have smaller hands than ideal for these keyboards. Traditional methods of studying this impact involve lengthy cohort studies which may not effectively quantify injury risks or test solutions. Instead, researchers recorded hand movements of elite-level pianists and employed AI to predict how different hand sizes would perform new music.

The study involved 15 elite-level pianists who played ten hours of music while cameras captured their hand movements from all angles. Using advanced computer vision techniques, researchers reconstructed these motions in three dimensions without attaching sensors that might disrupt performance. Karen Liu, professor of computer science at Stanford and lead author on the paper, noted: “The quality of data that we were able to achieve is unprecedented.”

Graduate student Roucheng Wang and postdoctoral researcher Pei Xu used this dataset to train a model capable of generating accurate piano-playing data from unseen sheet music like Beethoven’s "Für Elise." Elizabeth Schumann remarked on its accuracy: “I was just so stunned at how accurately this model could simulate elite-level technique.”

While current models simulate physical hand movements without accounting for muscle and tendon strain, future developments aim to include biomechanics predictions related to muscle tension and potential injury risks.

This research also extends beyond piano players; similar techniques have been applied successfully to guitar player datasets. Liu expressed optimism about expanding these models across various musical performances: “With high-quality data, we could model the 3D movements needed for other types of music performance as well.”

Liu is affiliated with several Stanford initiatives including Bio-X and Wu Tsai Human Performance Alliance. Schumann also belongs to Wu Tsai Human Performance Alliance. Graduate student Haochen Shi contributed as a co-author.

Funding for this project came from Wu-Tsai Human Performance Alliance, Stanford Institute for Human-Centered Artificial Intelligence, and Roblox.

For further inquiries contact Jill Wu at jillwu@stanford.edu.