Harmonic Overtone Detection Based Main Melody Extraction

With the rapid growth of music information on internet, it's very hard for users to retrieval information from large scale music database. The current online music search systems are limited to categorizing and text based searching. Context based music retrieval (CMR), especially query by humming has been more and more popular among researchers, system developers and end users, as an irreplaceable method to the traditional text based searching. Query by humming (QBH), as a brand-new music retrieval method, can help user to locate the wanted piece within a huge music repository by simply sing several tones. If users forgot the name or the singer of song, they can just hum a piece of music and the system will find the song in music database. Being a new style of human machine interface, the query by humming technology has important and practical meanings to content based music retrieval online.During recent few years, by the effort of our lab, we have constructed an effective Query-By-Humming System. Various humming inputs can be accessed to the system, with a high speed and a great accuracy. But there are still some constrains. For example, the music database can only use mono-track MIDI format files, while main melody extraction from multi-track MIDI, compressed file like MP3 format files and uncompressed file like wav format files is still under exploration. In this paper, aiming the polyphonic wav format audio files, we explore a new main melody extraction...