Machine learning in new musical instruments

A new era of expressive electronic instruments

In the world of electronic music there are new synthesisers and hardware created regularly. However, the rise in machine learning is about to change this completely. In a near future, instruments will have the potential to be expressive, complex and intuitive in ways previously experienced only through traditional acoustic instruments.  

A classic example of a new instrument using machine learning is the Mogees instrument. This device has a contact microphone that picks up sound from everyday objects, and attaches to your iPhone. The idea is that any object can become an instrument.

It is clear that machine learning will be one of the most important tools with regards to new developments in arts and music, and we’re genuinely obsessed here at Amphio. We’re currently experimenting with machine learning, aimed at musicians and artists.

Here is an example of one of our early creations. We had an input; a program where you could position a big green square in a space using the computer mouse.

We had an output; an FM oscillator.

Using a tool called Wekinator, we recorded three examples of what the oscillator could be doing, and three examples of where the green cube could be. Then we trained Wekinator with this data that it had recorded.

Now when we moved the green cube, the oscillator changed based on the cube’s position. From just three simple training examples, machine learning had generated all of the spaces in-between.

This is the area where we believe machine learning has the greatest potential, creating responsive, intricate, and musical electronic instruments. Imagine a drum machine then adapts to a performer’s playing style, learning as much about the player as the player learns about the instrument… Now, that’s cool.