Introduction
In this article, we discuss the usage of Artificial Intelligence to facilitate music creation.
Researchers worldwide are studying ways to use AI to do many great things with music. Though our algorithms are far from being Beethoven, AI is getting better with music as time goes by. It is already being used to create useful music-related tools to help musicians improve their work. Some even create complete tracks!
We chose not to enter the details of how these techniques are working, as we focus on demystifying how AI is currently used to help creators in various ways. However, you can find links to interesting articles for each use case!
🎨 Note: This article is part of our series on content creation. We hope to provide you with a better idea about the usages of AI for content creators, on a wide range of domains. Don't hesitate to check our other articles!
Fully automated music generation
Creating random music is relatively easy, but making good music is more challenging. Music is hugely subjective, but we know many talented people that can create beautiful songs that most people will like. If such people could learn to do that, we can probably teach algorithms to do the same thing!
Indeed, Machine Learning algorithms can understand what we like in music before generating pieces that match our preferences. This technology is still in its early stages, but it has already produced impressive results!
The first person to have produced music thanks to Artificial Intelligence is David Cope. His program, "Experiment in Musical Intelligence" (also called Emmy), has created thousands of compositions. Here is an example of such a composition, named Taurus, made by the algorithm in the style of Vivaldi:
Link to the original video: https://www.youtube.com/watch?v=2kuY3BrmTfQ
Of course, the results were curated by David Cope. Still, the quality of this music is mind-blowing, considering an algorithm has created it!
Not every algorithm is as advanced, and plenty of researchers in the musical field are working to help and inspire musicians.
Online, AI-powered tools can create basic tracks on users' specifications, which can be customized and tweaked as needed. Such engines can make tracks in any style, from hip-hop to EDM and country.
Here is an example of a track I have just created on AIVA's website, in the style of "Cyberpunk":
The user can then adjust the result as he wants to create a track that he will be satisfied with!
Similarly, AI can not only generate music following a specific style but also transform the style of music to fit another atmosphere in a movie or video game.
Musical Style Transfer
Indeed, it is possible to perform "Musical Style Transfer," which is about applying another style to a melody. When talking about "style," we are talking about the instrument used and how a particular author would most likely write it.
Here are some results obtained with a method established by Facebook AI Research in 2017 that takes a soundtrack produced with an instrument and transforms it into another musical instrument :
Cambini wind quintet original:
Cambini wind quintet translated to a piano, in Beethoven's style:
Find more results on their website.
Such an operation requires the algorithm to understand the style of an author and how to play different instruments.
Such knowledge is obtained by reviewing all the compositions of an author. This process is often used while working on the piece of a particular author, like creating whole songs from separate parts.
Assembling pieces of music in a coherent way
Recently, researchers have been able to complete Beethoven's 10th symphony thanks to Artificial Intelligence-powered algorithms. For this challenge, researchers only had access to sketches of melodies, drafts that Beethoven didn't have the time to complete. Previously, musicians were only able to assemble the symphony's first movement. Being able to complete it entirely is an outstanding achievement.
This task was possible because of a machine learning algorithm that analyzed the sketches and drafts and then assembled them into a coherent piece. This achievement is an excellent example of how AI is used to help us complete tasks that would be otherwise impossible.
Of course, the result is probably not what Beethoven would have written, but it is a good try.
As you can see, music transcription is essential as it could help us reconstruct this beautiful song. All of this while only having access to parts of the music sheet.
Yet, many people do not know how to write them, and if they create something great, they won’t be able to share it as easily. Fortunately, once again, AI is here to help us!
Music transcription
Another specialization where Machine Learning is practical is music transcription. Music transcription is about converting a piece of music to music notation. Performing such a task is typically complex and requires experienced musicians to listen to the music carefully to understand the notes played. Even with such careful listening, experts can make mistakes.
💡 Did you know?
At the age of 14, Mozart gained fame by transcribing from memory 9 different lines of the melody of a 15 minutes song, after hearing it only one time. The music sheets of the song, Miserere mei, had been kept secret for 150 years before that.
Fortunately, today machine learning makes the transcription process easier. For now, algorithms are not perfect and are limited to simple music. Still, these algorithms are improving at a high-speed rate.
As an example, here is a piano interpretation of River Flows in You by Yiruma (sound extracted from this YouTube video):
After transcribing this audio, we can use appropriate software to obtain the music score. We can then retrieve the same melody by asking a computer to play the notes:
Note: maybe we can get a higher quality result, I think this is due to the MIDI interpreter
For those interested, here is the complete music score of this music generated by the software AnthemScore:
📎 yiruma-river-flows-in-you.pdf
Conclusion
These were some of the most exciting applications of AI-related to music we wanted to share. But there are many more out there, and many more are to come!
We invite you to read our other articles about content creation. They are highly related to this one using similar algorithms with different data types, like videos or images instead of sounds.
.jpg)
