AI Revolution and One Application in Music Composition

Abstract

In recent years with the development of ChatGPT, music generation has become a very popular option for musicians or creators that want to create their own customized music for a specific purpose. In this paper we study how technological innovations over time have led us to the level of machine learning we have today, as well as an application of a recent model, the transformer, in music generation. The complexity of machine learning today has the capability for profound impacts on human lives and how the world operates. Many of the technological innovations we have today would not have been possible without the developments of old inventions. Each invention has played a significant role in preparing and allowing the next era of inventions to begin. We will discuss how it eventually led up to developments in AI and why it was necessary. Unlike simply generating text, music has very specific qualities to it that some humans can not even perfect. Along with rhythm, melody, and pitch, emotion also plays a significant role in the sound of music. There are also many different voices and sounds from different instruments. There might also be some underlying message or tone that is communicated throughout the music. While generating music is always impossible to master, the application of transformers using the attention mechanism that this paper will explore is a viable way to create music stylistically similar to any given training set. Using this method, musicians or anyone could generate a piece of music of any choice of mood, genre, etc. using a huge amount of training data.