AI That Can Generate Music

Original article was published on Artificial Intelligence on Medium

AI That Can Generate Music

A compilation of interesting tools and frameworks

Photo by Maxime Bhm on Unsplash

30 seconds to mars’s Do or Die really shows one how important music is to man. Music is everywhere, games, movies, celebrations. It would be quite hard for someone to mention an activity where music doesn’t play a role. Today, we will be exploring how Artificial Intelligence can make a difference in Music.

The history of music is not just about new creations, it is also about technological innovation. The making of music is a complicated process but with the help of AI, one can ease into the process. Recent development certainly has paved a way to create music like never before. Below we will explore 7 projects which have used AI and Machine learning to give something new to this world.

AI Music Tools And Frameworks

Flow Machines — 2012

  • What’s music without Sony, we mean they literally have the best headphones for the consumer market.
  • Sony’s Flow Machines is a research and deployment project aimed at achieving the augmented creativity of artists in music.
  • The main part of the Flow Machines project is Flow Machines Professional. It is a music composing system supported by AI. Using this system, creators can compose melodies in many different styles which they want to create, based on its own music rules created by various music analyses.
  • Generating new melody and chords is a challenge to all musicians and this application may just be the way around it. This below video contains a sample generated by it.

Computoser — 2014

  • Computoser uses an algorithm to generate music. Each generated track is compiled by a unique combination of tones, rhythm, and instruments.
  • The interface is quite clean and easy to comprehend. You just need to choose the one you need from its available options and click play to enjoy your music.
  • A major drawback is that the songs produced are pre-loaded and can’t be edited. You can explore this tool here.

AIVA — 2016

  • AIVA, founded by Pierre Barreau is an app that helps one create personalized music with the help of AI.
  • The app contains various genres ranging from soft jazz to hard rock.
  • One has the option to upload their own music or choose a set genre. Then set the genre-specific options.
  • We tried repeating the same options and realized the real work of AI setting various tunes for the same option.
  • The app is mainly based on a stochastic AI process which in simpler terms means random processes. One thing easily identifiable is the large range of hyper-parameter and parameter tuning because of the large choice of options given to users.
  • The only drawback present is that the free options limit only to 3 downloads and no download of rock and pop. Below we have attached a sample of the music we created using the app which is a blend of tango and symphonic orchestra!

Magenta — 2016

  • Google’s Magenta is practically the “Next big thing”. It defines creation.
  • The Magenta is an open-source research project exploring the role of machine learning as a tool in the creative process. It consists of both art and music.
  • In the music department, it has published many papers which can be found here. Few of the key applications such as NSynth, MusicVAE are what Google is going for.
  • Generative Networks form the base of all of them but to understand what really is going inside you need to be part of the Google brain team!
  • Magenta is the frontier that shows what AI and machine learning can do and it is just downright amazing. Do look through their other projects, some code and libraries can be found here.

MuseNet — 2019

  • MuseNet is another wonder generated by OpenAI.
  • The focus here is on classical music and finding out patterns inside them. They use a deep neural network to generate 4-minute musical compositions with 10 different instruments and can combine styles from country to Mozart to the Beatles.
  • The program doesn’t try to understand the human intellect in music but to discover patterns of harmony, rhythm, and style by learning to predict the next token in the music file. The prediction is done by a large-scale transformer model.
  • A major application could be to complete long lost symphonies created by old artists.
  • A minor limitation is the combination of odd composers and instruments such as Chopin and Red Hot chili peppers?
  • Below is a concert generated by MuseNet.

Jukebox — 2019

  • Jukebox really is a piece of magic created by OpenAI and is built to open the foundation of MuseNet.
  • With the input as lyrics, genre, and the artist, it can make its own music. This is all done by a neural network (autoencoder) which compresses audio into a discrete space and then regenerates it.
  • The possibilities are really endless with this one. Just imagine if it reaches full potential and Kanye drops an album which has a billion views and he barely had any effort at all.
  • One possible drawback could be its unwarranted usage. Random people could use anyone’s voice/style and generate music, something to ponder about?
  • This link will take you to their curated playlist which also has the paper and the code!

Amper Music — 2020

  • Amper music composes music with you. It is extremely easy to use and the dataset the algorithm is trained on has a lot of variabilities.
  • One really has a lot of options at hand to create their own music and it is fast. It creates actual audio and not MIDI files.
  • They are paving a way for everyone to create music in an easy process. This application creates an environment where anyone creative can shine, not just professional musicians.
  • The only drawback is that it is completely paid. A sample of their work can be viewed here.


Projects like Google’s NSynth are mind-blowing. The complex algorithm is trying to find new music, new instruments by combining various pieces and notes. A lot of research is going into AI and its creativity and good results are being posted such as this song about Eurovision.

The music industry can really go wild once scalable real-life applications are made. Instead of one album, there can be 50 of them from a single artist within a year. A new grammy section maybe? The possibilities are endless and it is just the beginning.

Sit back for a minute and imagine a concert generated by an artificial Michael Jackson created by robotics with the music created from scratch!