Key Finding in a Query by Humming System
Saturday, March 26, 2005, 06:00 PM
Link to PDF
Abstract
I describe an extension to the melody matching system used in
Gersic. The system takes user input in the form of a sung or hummed query recorded through a microphone. Once recorded, the query is transcribed into a string of quantized MIDI notes which are matched to a series of similarly transcribed audio files. The matching process is executed by means of a local-alignment edit distance algorithm. The extension described here leaves the system intact, but adds a diatonic key-finding system between the transcription and matching processes. Once the diatonic key of a query has been discerned, this information can be used to reduce transcription or singing errors.
