On February 26th, an interview article about the research of Associate Professor TAKAGI Yu of Computer Science Group was published on the BBC (web version), the UK's public broadcaster. This article was published in "BBC Future," the BBC's science and technology section, and is also featured on their homepage. Please take a look.

BBC website

BBC Future homepage

The featured article, "How AI can read our scrambled inner thoughts," highlighted cutting-edge research using AI to reconstruct perceptual experiences from human brain activity, and introduced Associate Professor TAKAGI's "AI-based brain reconstruction research on music."

This research was published in 2025 as a joint research project with Google Inc., and was led by Associate Professor TAKAGI as co-first author and co-corresponding author. They demonstrated that by associating fMRI (functional magnetic resonance imaging) data collected while listening to music with "MusicLM," a music generation AI developed by Google, it is possible to regenerate songs similar to the music being listened to. Because fMRI measures every second, it is considered difficult to reproduce music that changes over time. However, it was confirmed that the generated music can be reconstructed in terms of basic categories such as genre and atmosphere.

Furthermore, this study analyzed the correspondence between the semantic representation of music in the human brain and the internal representation in generative AI models, revealing new insights into the neural basis of music perception.

Research points
- Reconstructing music from brain activity using a generative AI model
- Elucidating the correspondence between human musical semantic expression and internal representations in generative models.
- A fusion approach of neuroscience and generative AI

This paper has attracted international attention as cutting-edge research that fuses AI and neuroscience, and has even been featured in the British science media outlet "New Scientist."

Paper Information

Journal name: Nature Communications
Title: Text-to-music generation models capture musical semantic representations in the human brain
Authors: Timo I. Denk†*, Yu Takagi†*, Takuya Matsuyama, Andrea Agostinelli, Tomoya Nakai, Christian Frank & Shinji Nishimoto
(†Co-first author, *Co-responsible author)
DOI：10.1038/s41467-025-66731-7
URL：https://www.nature.com/articles/s41467-025-66731-7

Related Links