WRITINGS
WU Hung-Hsuan / 吳宏宣
Music Information Retrieval / 音樂資訊擷取 / Yinyue zixun jiequ
Edition: Mémoire de master, Département d’informatique et d’ingénierie (資訊工程學系) de l’Université de Tamkang (淡江大學) sous la direction de LIN Hwei-Jen (林慧珍)
Date: 1998
Region: TAIWAN
Origin: TAIWAN
Type of media: Grey
Language: Chinese
Editor: LIAO Lin-Ni
Comment:
Abstract:
With the growth of internet, the digital representations of music were broadcast rapidly. In traditional way, users search the digital music by text. There are more than one dozen digital music databases had been established in last decade. User can input a fragment of music to query the database to retrieve the information they want.
We have proposed an approach to main melody extraction from MIDI files of multi-track polyphonic music. In each track of a MIDI file, the system traces the pitch contour of the polyphonic music, which can be described in a monophonic form. In each of these contours, maximal repeating patterns are found by using a correlative matrix. All these patterns are then collected in a dictionary, with which we can find the set of all maximal repeating patterns with no redundance, called the main melody, from the MIDI file of multi-track polyphonic music.
The results of main melody extraction and how well these extracted main melodies can improve the work on content-based music retrieval are given and described.