Solving Misheard Lyric Search Queries Using a Probabilistic Model of Speech Sounds

Music listeners often mishear the lyrics to unfamiliar songs heard from public sources, such as the radio. Since standard text search engines will find few relevant results when they are entered as a query, these misheard lyrics require phonetic pattern matching techniques to identify the song. We introduce a probabilistic model of mishearing trained on examples of actual misheard lyrics, and develop a phoneme similarity scoring matrix based on this model. We compare this scoring method to simpler pattern matching algorithms on the task of finding the correct lyric from a collection given a misheard query. The probabilistic method significantly outperforms all other methods, finding 5-8% more correct lyrics within the first five hits than the previous best method.