A Music Retrieval System with a Seamless Query Interface by Humming or Song Title

We propose a music retrieval system that enables a user to retrieve a song by two different methods: by singing its melody or by saying its title. To allow the user to use those methods seamlessly without changing a voice input mode, a method of automatically discriminating between singing and speaking voices is indispensable. We therefore designed an automatic vocal style discriminator and built a music retrieval system that retrieves a song by query-by-humming for singing voice or by dictating the song title by automatic speech recognition (ASR) for speaking voice. Experimental results using a large music database that built for singing research show that our system is able to discriminate between singing and speaking voices with 96.1%. The average retrieval rates of correct songs in the top 10 of 100 songs by query-by-humming and ASR for song titles are 50.5% and 96.7% respectively.