CofC Logo

A Musical Search Perfected

Bill Manaris, computer science professor

feature profile

Let’s say you love music. Suppose you love Beethoven, Alicia Keys and Miles Davis. Now imagine being able to upload your favorite music to Google and have the search engine find similar pieces, many of which might be completely new to you.

That concept is behind the research project for which Bill Manaris, computer science professor and amateur musician, recently received a National Science Foundation grant. His co-principal investigator is Dwight Krehbiel, an expert in cognitive neuroscience and the psychology of music at Bethel College in Kansas.

Their research is based on the principles of Zipf’s Law, which is named for the Harvard linguist who first proposed it in the 1930s. Zipf noticed that in any book, the most frequently used word (the, for example) occurs twice as frequently as the second most frequently used word (of), which occurs twice as often as the fourth, and so on. He also found that the total number of words that are used in any one piece is fairly small.

Something similar occurs in musical compositions.

The Classical Music Archives donated 15,000 MIDI-encoded pieces to the experiment. Manaris and his research team, which included graduate student Patrick Roos and senior Thomas Zalonis, used the data to analyze musical styles and “train” the computer to perform tasks such as classifying the pieces according to author, style and pleasantness. The computer can search through the 15,000 compositions in the database and recommend several musical pieces – not necessarily in the same genre – that “match.”

Simply put, the computer program can do the job of analyzing what makes particular music pleasing to your ear in terms of the proportions of notes, harmonic intervals, pitch, etc., search the database, and provide you with a list of similar musical works.

Manaris and his team are excited with the preliminary results of their research. The people they have tested indicate that the pieces the computer recommends are very similar to the input piece in terms of the emotion they evoke and mood they create. And more importantly, they report that they like the recommended pieces even more.

To see where Manaris’ research might be taking us, go to