Michael Langford, a computational engineering and music double major at The University of Texas at Austin who is graduating this spring, is working on an honors thesis to study machine learning methods for music composition. The idea for the thesis was sparked in part by the “fireworks” he experiences while listening to music due to his associative chromesthesia—a form of synesthesia that invokes an involuntary experience of color—along with his interest in how computers can interpret data to predict future scenarios. We sat down with Michael to learn more about his thesis, its potential applications for the future, and how he plans to develop his concept.

michael langford photo

In a few words, how would you describe your thesis? What is your goal for this project?

I want to develop and compare new methods of composing music using computer software. I am studying two different types of neural networks—simple recurrent neural networks and long short-term memory (LSTM) neural networks—and investigating their effectiveness in modeling large-scale patterns in music. Specifically, my goal is to have these programs compose fugal music—a Baroque-era style in which a central theme is recalled throughout a piece—which is very mathematical and precise. Hopefully, the computer programs will have an easier time modeling these musical sequences due to the mathematical nature of the genre. 

Many people confuse computational engineering with computer science. How do you define computational engineering?

Whereas computer scientists may write programs or create apps and web pages, computational engineers use computer languages for engineering purposes. For example, we may be given a new design of an airplane that an engineer designed and be asked to determine whether or not this design will fly properly.

When did you first notice your associative chromesthesia? How did you react?

I think I've always had it—I just assumed everyone had it so I didn’t think much of it as a child. My specific case means that each letter and every number has a specific color associated with it. I don’t have the type of synesthesia where I project these colors onto the world; they’re just kind of in the back of my mind. It’s especially vibrant when I'm playing music because all the sounds and notes I hear get translated into color—it’s like fireworks going off in my head.

Michael Langford, a computational engineering major, developed an honors thesis to study machine learning methods for music composition. This computational video and sound byte is small sample of the output from the neural network that Langford trained. The network learned patterns from the sequences of notes from 113 performances of Bach’s preludes and fugues to create an original piece of music.

My second year I took a course offered by the natural sciences department for their Polymathic Scholars program, where we were required to develop a new field of study we were interested in. I thought it was perfect. If I wanted to do my engineering honors thesis, I might as well start my second year developing a topic that I'm really interested in. After I saw preliminary work people were doing in applying computer intelligence in music composition, I decided that’s an area I really wanted to explore because it combines my two interests, music and computational engineering.

Will you apply your chromesthesia into your work? Will you attempt to teach the machines which music evokes which color?

I don’t know if I'd have the abilities to do this, but I could also write a program that goes through a piece of music and shows the colors I would be experiencing while listening to it. The problem with chromesthesia is that the majority of people who have it experience the colors differently, which is unfortunate because one of my favorite composers, Alexander Scriabin, also had chromesthesia and was obsessed with color, but his color experiences don’t line up with mine. It would be cool to show, but others may disagree with me as strongly I disagree with Scriabin.

How might you apply what you learn from this project in your future work?

There are many more people just trying to write algorithms that can compose music. I think it’d be cool to do something like this in the future as a career. There are very few companies working on this as a career path that I know of. 

Have you thought about the implications of this work if it is successful? Writing music may become a job for machines in the future; how would you feel about this?

I think there’s a good chance that yeah, in the next 30-50 years a lot of computers will be doing music composition, but some different writers have said that we shouldn’t see it as computers are automating composers out of a job; we should see it as composers are gaining new tools to help them compose. There’s a chance, yeah, but machine learning and artificial intelligence approaches, like my thesis, can be really narrow. They may have a good idea at the beginning that never gets repeated. I’ve found it hard for machines to do that when a human composer can do that so easily.