Development Of A Multi-channel Audio Retrieval System Based On Content

Currently, web music retrieval is confined in categorized scanning and text basedsearching. To find a song, the user has to know a series of information about its topic,composer, and performer. Otherwise, it can be explored tediously along the category. Such atext based retrieval technology is far from satisfactory for music searching. For example, itwould be extremely difficult, if one want to find a song by groaning its melody. Thus, themusic retrieval should be capable of recognizing and extracting the audio features of music.Namely, the retrieval technology is to be improved from text-based to content-based.In the recent years, content-based music retrieval attracted extensive study. Fruitfulresults in algorithm for single-channel melody features extracting were appeared in foreignliterature and relevant retrieval software was available. Researchers in Hong Kong andTaiwan districts contributed in this area by improving these theory and method. Quite anumber of domestic research works were also reported. But, there is some disparity incomparison with the level already achieved.Work of the theses is multi-channel music melody features’ recognition and extraction.Single-channel audio information can be extracted sequentially. But, it is much morecomplicated for multi-channel ones. Based on music theory, representation of the melody isanalyzed and extraction for multi-channel the algorithm is modified. A MIDI formatmulti-channel melody database is designed to establish content-based audio retrieval system.The music query can be responded by humming (QBH). A WAV format database system isalso proposed.Validity of the designed system is verified by laboratory test-run of the prototype.