Automatic Beat Alignment of Rap Lyrics

Rap is characterized by highly rhythmic delivery of words; the subtle ways in which syllables in the lyrics fit into the beats in the music lead to expressivity in rap music. Unfortunately, given a lack of notated score for the genre of rap, the task of determining the rhythm to existing rap performances must be carried out manually. In this paper, we explore a method of automatically aligning rap lyrics to beats using logistic regression. Specifically, we select top linguistic and audio features and compare the two resulting models. The audio model yielded about 3% higher accuracy than the linguistic model but at the cost of heavy computations in the data preparation stage. Potential applications to our technique include automatic rap rhythm transcription and style characterization and imitation of rap artists.