Distinct representation of music in deep neural networks trained for natural sound detection without music. a Examples of log-Mel spectrograms of natural sound data in the AudioSet. b Deep neural network architecture used to detect natural sound categories in the input data. The purple box indicates the average pooling layer. c Performance (average mean accuracy, mAP) of the network trained without music for music-related categories (top, red bars) and other categories (bottom, blue). d Density plot of the t-SNE integration of the feature vectors obtained from the network in C. The lines represent iso-proportion lines at the 80%, 60%, 40%, and 20% levels. Credit: Natural communications (2024). DOI: 10.1038/s41467-023-44516-0
Music, often considered the universal language, is known to be a common component of all cultures. Could the “musical instinct” be something shared to some extent, despite the vast environmental differences between cultures?
A team of KAIST researchers led by Professor Hawoong Jung of the Department of Physics used an artificial neural network model to identify the principle that musical instincts emerge from the human brain without special learning.
The research, led by first author Dr. Gwangsu Kim of the Department of Physics at KAIST (current affiliation: MIT Department of Brain and Cognitive Sciences) and Dr. Dong-Kyum Kim (current affiliation: IBS), is published in Natural communications under the title “Spontaneous emergence of rudimentary music detectors in deep neural networks”.
Previously, researchers attempted to identify the similarities and differences between music that exist in various cultures and attempted to understand the origin of universality. An article published in Science in 2019 revealed that music is produced in all ethnographically distinct cultures and that similar forms of rhythms and melodies are used. Neuroscientists have also learned that a specific part of the human brain, the auditory cortex, is responsible for processing musical information.
Professor Jung’s team used an artificial neural network model to show that cognitive functions in music are formed spontaneously as a result of processing auditory information received from nature, without being taught music. The research team used AudioSet, a large-scale collection of sound data provided by Google, and trained the artificial neural network to learn the different sounds.
Interestingly, the research team found that some neurons in the network model would respond selectively to music. In other words, they observed the spontaneous generation of neurons that responded minimally to various other sounds like those of animals, nature or machines, but showed high levels of response to various forms of music, both instrumental than vocals.
The neurons in the artificial neural network model showed responsive behaviors similar to those in the auditory cortex of a real brain. For example, artificial neurons reacted less to the sound of music that was cut into short intervals and reorganized. This indicates that spontaneously generated music-selective neurons encode the temporal structure of music. This property was not limited to a specific genre of music, but emerged in 25 different genres, including classical, pop, rock, jazz and electronic.
Additionally, suppressing the activity of music-selective neurons significantly impairs cognitive accuracy for other natural sounds. That is, the neural function that processes musical information helps process other sounds, and this “musical ability” may be an instinct formed as a result of an evolutionary adaptation acquired to better process the sounds of nature.
Professor Jung, who advised the research, said: “The results of our study imply that evolutionary pressure helped form the universal basis of musical information processing in various cultures. As for the significance of the research, he explained, “We look forward to this artificially constructed model with human-like musicality becoming an original model for various applications, including music generation through AI, music therapy and music cognition research. »
He also commented on its limitations, adding: “This research does not, however, take into consideration the developmental process that follows music learning, and it should be noted that this is a study on the foundations of processing of musical information during early development. »
More information:
Gwangsu Kim et al, Spontaneous emergence of rudimentary music detectors in deep neural networks, Natural communications (2024). DOI: 10.1038/s41467-023-44516-0
Provided by Korea Advanced Institute of Science and Technology (KAIST)
Quote: Research team breaks down musical instincts using AI (January 23, 2024) retrieved January 23, 2024 from
This document is subject to copyright. Apart from fair use for private study or research purposes, no part may be reproduced without written permission. The content is provided for information only.
