The Story of the Songshark Project

I originally became interested in algorithmic music when I was writing songs for hire. As seen in the accompanying video, I was writing songs about people, from commissions, based on information in their Facebook profiles, and filming myself singing the songs. I had found a market for the type of music that I was writing -- funny, uplifting tunes that were somewhat simple in structure. From writing the songs in great volume and rapid succession, I was able to gradually define a songwriting procedure.

Over time, my procedure became increasingly specific. With the music, I'd decide on a common chord sequence and a time feel. With the lyrics, I would look for rhymes, be deliberate about which pronouns I used, use specific tenses with verbs, and make an effort to include words about food, sex, birth, death...themes which I thought would capture the attention of a listener.

I eventually realized that there was a direct connection between my emotions and the music I would write; if I was happy, feeling good about the future, depressed, confused or distressed, both the music and the lyrics I would write would reflect my emotional state. In one case, I ran into a state of writers block when just after I had ended a relationship, I was paid to write a jingle for an iPhone app. Every melody I could come up with was somber, and each set of lyrics had a hint of melancholy; I was unable to produce a jingle that was appropriate in mood.

I believed that regardless of what I did, the music that I made would be some type of emotional expression, a bearing of my feelings in musical form. I had the idea that if I could create tools that would enable me to write 5, 10, or maybe even 50 songs in a single day, I could use the tools to write songs only when the feeling was right. I knew that it would be difficult for me to come up with some sort of procedure to generate an entire composition from scratch, but I became fascinated with the idea of rapid musical prototyping, and specifically the automation of certain parts of my creative process.

I first experimented with poetry generators. I think that my most successful one was this beat poem generator that I wrote. You can view the rest of the generators here. I also began making simple lyrics generators using templates. You can see an early experiment here.

For a few years, I focused on developing my general skills as a programmer and did not work on music at all. Then in January of 2014, I was encouraged by my friend Brandon Winston to try to further my research in developing a songwriting application. While Brandon took a leave from Harvard Law School, we worked together in California to research all aspects of algorithmic composition and music prototyping. We also developed our own software, and much of the software that I am archiving on this website is from that time.

Brandon Winston and I made use of our custom songwriting software in producing all the tracks on this page. The parts of the creative process that we were able to successfully automate include rhyme matching, rhyme generation, MIDI phrase generation, chord sequence generation, MIDI drum accompaniment and batch MIDI file rendering.

- Andrew Fenlon