Abstract
In this thesis, an approach for content-based music database retrieval is proposed. Traditionally, music is retrieved by keywords of titles, singers or lyrics. In this study, we implement a CBMR (content-based music retrieval) system that provides a friendly interface for users to retrieve music based on a few notes sung or hummed to the microphone. At database creation, we extract the tune information from the music database and translate it into a predefined mid-level representation. When the user specifies a query by singing a piece of notes, several methods are used to transcribe the acoustic input into the mid-level representation. Next, the similarity between the query pattern and database pattern is computed. Then, the CBMR system returns the results according to the similarity. All the system is coded in Matlab 5.3 and Simulink 3.0. The average computation time of our CBMR system (with 270 songs in the database) is about six seconds, including sound recording (5 seconds) and other related processing.