Query-by-Humming (QBH) is the problem of identifying songs that approximately contain a sequence of notes hummed, sung or whistled by a user. Its central challenges are: assembling a comprehensive, well-organized music collection appropriate to the task, transcribing the voice query into a sequence of note pitches and durations, and searching for that sequence in the music collection while allowing for user input errors, transcription errors, ornaments and variations. We introduce an algorithm that helps one organize a large MIDI file collection, a pitch tracking algorithm which focuses on accurately pinpointing the harmonics in a voice signal, and our adaptation of vantage object indexing, a melody matching technique due to Rainer Typke. Finally, we present the Musiseek search engine, an implementation of these algorithms. We describe the collection of query data by human subjects and its impact on evaluation and development. We conclude with directions for future research.
Copyright is held by the author.
The author has not granted permission for the file to be printed nor for the text to be copied and pasted. If you would like a printable copy of this thesis, please contact firstname.lastname@example.org.
Member of collection