| Summary: | 碩士 === 國立成功大學 === 電機工程學系碩博士班 === 92 === The unit of data stored in music database directly affects the data size and searching complexity of music database. Most of research in music database use note as basic unit of data stored in music database. But in musicology, the basic unit of music is not note but figure. Figure is a group of successive notes, and people usually can feel the boundary of a figure. Because figure is consistent with human perception, therefore, figure is suitable for being basic unit of music database. In order to extract figures of a music work, a music surface segmentation algorithm LBDM (Local Boundary Detection Model) is needed. Besides, there is a common observation that human usually keeps only some successive important figures of a music work in mind. And such important parts of music work can be extracted by applying a repeating pattern finding algorithm because important parts of a music work usually repeat frequently. Therefore, if figure is basic unit of music database, then it can easily find human perceptive repeating patterns in a song because figure is consistent with human perception.But, when applying repeating pattern finding algorithm on huge music database, the speed is critical because the time complexity of repeating pattern finding algorithm is very high and searching space of current music database is huge. In this thesis, a speed enhancement repeating pattern finding algorithm and an improved version of LBDM algorithm are developed. By combining these two algorithms, human perceptive repeating patterns can be found quickly and could help to increase the performance of music database for further research.
|
|---|
