In the US there is a musical group YACHT. It is possible that you have never even heard of, although they play already since 2002 and has already recorded many songs. In General, this team of musicians is always approached his work is quite creative. For example, once they sold the album without the cover. But she still could get a… Fax to the nearest post office. However, for such YACHT would hardly have appeared on the pages of our publication. But recently released a group album called Chain Tripping makes draw attention to themselves. After all, the co-author of the musicians here made artificial intelligence.
Why AI to write songs?
Group YACHT quite a long time interested in the question of how high technology can help musicians to improve their work. Vocalist Claire Evans performed at the exhibition Google I/O a presentation on this topic. Creating and recording a new album Chain Tripping YACHT took three years.
I know it’s quite hard to explain, but the use of artificial intelligence and machine-learning techniques allowed us to find a new approach to creating songs. We could add “additional layers” of music in those places, it would put them we would have guessed. said Claire Evans.
In fairness it should be noted that artificial intelligence and neural networks have been used to generate music and even text, but it is not beyond the scope of research projects designed to demonstrate the capabilities of the AI. Group YACHT decided to do something a little different. To use AI as co-author team decided not to cause a stir and say, “Look, this song has been written by a machine!”. And in order that in recognition of the musicians, their goal was to create music that they themselves would never do, but which would sound to their style. What’s your beloved band? Tell us about it in our chat in Telegram.
How to teach a machine to write songs?
In the beginning, the musicians tried to establish contact with the authors of almost all existing solutions based on AI. But none of the projects did not satisfy the musicians for a number of reasons. So they decided to take all of the known developments and to create on the basis of this something. For neural network learning, as you know, first you need to “feed” them information, based on which they will work in the future. It was decided to use discography of the team to simplify the work I had to overwrite in MIDI format completely. Songs in the discography of the group was at the time only 82 pieces. And this, unfortunately, is too small in order to fully teach the neural network to create something new based on existing.
This is interesting: How does the artificial intelligence.
However, this number was quite enough to make the AI “learned to play musical instruments”. And he mastered several instruments and options “issued” tunes, in the style of a musical group. Thus, in recognition of the musicians, they didn’t have the “magic button” which when pressed they would receive the finished song. Artificial intelligence only increased their melodies and rhythms of the additional sound. Well, the result of you can appreciate, listening to a track from the new album (and at the same time and evaluate a music video, made in a very peculiar style, as well as the music) the video is available below.