Ian Simon, Sumit Basu, David Salesin, and Maneesh Agrawala
This paper describes a method for creating new music by concatenative synthesis. Given a MIDI score and an audio recording of an example piece of monophonic music, our method synthesizes audio to correspond with a new MIDI score. The algorithm we use is based on concatenative synthesis, commonly used for generating speech. Two versions of our algorithm are explored, one in which individual notes from the example piece are concatenated, and one in which pairs of adjacent notes from the example piece are concatenated. We examine the range of example pieces and target scores for which each version of our algorithm yields good results. Our underlying framework remains general enough to be applicable to other problems, such as rendering a stylized version of the target score, and other types of sound analogies.