This week at ESF South Island School, the students carried on exploring how AI is being used in the music industry. Students enjoyed experimenting with AI Duet, a software that uses machine learning to allow you to play a duet with the computer. We taught the students how unlike other softwares where the computer receives a set of rules, AI Duet involves playing the computer lots of example melodies, so that over time the computer understands the relationship between notes and timings. This means that when you play a sequence of notes, the computer goes to the neural net and plays a possible response.

Students were then able to use Neural Synthesiser Instrument to interpolate between pairs of instruments to make new sounds. They learnt how the software works by finding a compressed representation of sound, the encoder network transforms a sound into this representation of it and the decoder network then converts it back into sound, so that the system is trained to reproduce sound as similar as possible to the real sound.

It was great to show the students Latent Loops and teach them about how this software uses latent space models to represent compressed data where similar data points are closer together in space because this makes it capable of learning the basic characteristics of a training dataset and capable of excluding unrealistic examples. We also brainstormed the desirable properties of a latent space with the students, including expression, realism and smoothness.

We ended the session by discussing some thought-provoking controversial questions about the future goal of the music industry and why this goal might not be to compose a masterpiece made from scratch by solely AI, but rather to use AI as a musical assistant. We got into a very interesting discussion regarding if AI could write a beautiful song, what would it value and why would it choose to write about one thing over another? These are crucial questions if we are to use AI to create music in the future!

❮ ❯