A data set of 330 songs / 1 million tones will be made public that makes it possible for AI to understand "What is Bach's feeling?"

ByGrfx_guru

Humans listen to music and can analyze "This song is Bach's composition" or "This phrase is similar to Mozart". In that case, analysis of the height and length of the sound in the brain and checking against your own memory ... is done, but the computer also has a similar ability to make it possible Detailed data set has been released.

What makes Bach sound like Bach? New dataset teaches algorithms classical music | UW Today
http://www.washington.edu/news/2016/11/30/what-makes-bach-sound-like-bach-new-dataset-teaches-algorithms-classical-music/

In November 2016, the research team of the University of Washington, USA gathered a huge dataset that analyzed classical music "MusicNetWe announced. This data set is an analysis of 330 songs licensed free sounds, which shows the pitch of all the sounds being played, the timing of sound generation / final sound, etc. from "◯ minutes Δ △ seconds to × ◯ seconds" "The sound is ringing" has succeeded in converting the data to very fine levels. As a result, the number of sound data has reached more than one million items.

An example of that image is released in the movie. Beethoven 's "given as a sound sourceString Quartet No. 11"The height and length of the sound, as well as the musical instruments being played are converted into detailed data.

MusicNet - YouTube


That is to say, it is data in which extremely detailed music score was recorded based on actually sounded sounds. Based on this data set, let AI learn "music" and reconstruct classical music Attempts are being made. Associate Professor Sham Kakade of the University of Washington said, "We are at a high level" Why does music listen to ears "and how to better understand the composition of the songs or what is" It is interested to understand whether it is connected to the "likeness" of the music, "he says about MusicNet's possible things" From the number of directions, MusicNet creates creativity and actual production in the field of machine learning and composition We hope to demonstrate the benefits of it. "

Until now, analysis of music was often advanced by human ears and hands, but MusicNet made it possible to perform these actions almost automatically based on enormous data as a big advantage in the future It is said that. This is an attempt to realize the same thing as "ImageNet" which can analyze the contents of photographs with big data and AI and analyze reflected objects and people even in the music world.

The data set acquired by the research team is released on the following site, so that anyone can download it.

Getting Started
http://homes.cs.washington.edu/~thickstn/start.html


Although the paper by the research team is under peer review,Preprint serverIt is possible to view it with "arXiv" of.

[1611.09827] Learning Features of Music from Scratch
https://arxiv.org/abs/1611.09827