以情緒感受為基礎之自動音樂選曲系統

Abstract

音樂聆賞者在使用數位音樂檔案聆賞音樂時，最常遇到兩種問題：(1)需要花費大量時間整理並管理數位音樂檔案資料庫；(2)現存的音樂播放軟體中不存在以情緒為依據之自動選曲功能，因此在過去的聆賞經驗中，尚未有符合一般認知的自動音樂選曲系統。本文應用內涵式音樂資訊檢索技術開發基於情緒感受之自動化音樂選曲系統以解決上述的兩項問題。研究中以基於音樂理論之聲學特徵萃取音樂訊號之聲學特徵，並以相似度量測演算法計算得出相似的音樂檔案，最後以主觀標記的音樂類型與情緒感受等資料判斷檢索結果是否符合一般認知，並測試所有演算法的檢索效能。在本次研究準備之測試音樂內容之下，音樂類型的測試裡最高檢索效能達94.17%；在情緒感受的測試裡檢索效能最高有98.75%。

There are two main issues for listeners when they using digital music files. First, manual music classification and management in digital music library take a large amount of time. Second there is no automatic music track selector which is based on emotional appeal in existed audio player. Therefore, in the past there is no such system which can choose music tracks automatically and fits music common sense at the same time. This paper proposes a system which can automatically select music tracks based on emotional appeal to solve the two issues that we have mentioned and this system is an application in content-based retrieval. In this research, the acoustic features are extracted from music signals and based on music theory. After that similarity functions are used to pick the similar music files. Finally evaluation the preferences based on music common sense. Using the music content that we prepare for subjective tests and evaluation methodologies that we esign get the maximum average precision in music genre test is 94.17% and in emotion test is 98.75%.