AI piano music transcription improves

Japanese researchers have developed a method that uses artificial intelligence to accurately transcribe musical scores from piano performances in just a few minutes.

Researchers from Kyoto University’s Hakubi Center for Advanced Research and other entities developed a system that capture piano sounds and quickly transcribes a score on music sheets.

It would take most musicians more than an hour to transcribe a five-minute piano performance, but this AI-powered technology can reportedly complete the task in about one or two minutes.

According to Eita Nakamura, an assistant professor at the center, research into automatic transcription of music has been conducted since around the 1980s, but producing successful results has been extremely difficult because the process requires recognizing changes in rhythm and multiple sounds being played simultaneously.

For this project, the AI-powered system learned the audio data of about 200 pieces of classical music and about 800 popular Japanese and Western songs, improving its ability to accurately estimate the rhythm, pitch and intensity of the notes.

Although the system’s accuracy still has room for improvement, about 70% of 125 students surveyed at municipal Kyoto Horikawa Senior High School of Music said it “looked useful for transcribing music.”

The research findings were published in the online edition of a U.S. academic journal in March. The researchers plan to investigate whether the system also can be used for other instruments such as the guitar and drums.

“Many songs uploaded online every day don’t have sheet music,” Nakamura said. “If an individual could easily transcribe these songs, I think more people would find pleasure in performing music.” (Sept. 29)