Query by humming with the vocalsearch system

Systems able to find a song based on a sung, hummed, or whistled melody are called query by humming (QBH), and are used by the VocalSearch systems. The system provides performance with the best current systems and involves a simple design and user interface. Queries are in the form of a user-supplied melody, theme, hook, instrumental riff, or some other memorable part of a piece. The typical QBH melody-comparison techniques, string alignment, n-grams, Markov models, dynamic time warping, compare monophonic melodies in the database to a monophonic query. Each song in VocalSearch's database is represented by a theme for the verse and a different theme for the chorus. The QBH system considers tens of thousands of themes. VocalSearch uses a probabilistic string-alignment algorithm to measure similarity between targets and queries. The melodic encoding used for queries and targets has a strong effect on the kinds of errors a QH system can handle.