In the past, the field of audio generation has mainly been associated with research for automatic speech recognition (ASR). However, recently, technologies that were originally developed for ASR have been applied to generating simple music from raw audio. As the field of music generation is relatively new, this project will explore several of these new advances in the field, including models like generative adversarial neural networks and recurrent neural networks. The goal of this project is to extend these models so that they are capable of learning a higher-level picture of musical tendencies, and, as a result, can generate more complex and longer segments of music.

I’m participating in SuperUROP because I’ve always had an interest in music and in machine learning, and now, I have the opportunity to conduct research involving both. I have taken several machine learning classes and kept up on research in this area for over a year. Through this project, I’m hoping to expand on my current experience in the field and to experience music in a different manner.