2013-08-012024-05-18https://scholars.lib.ntu.edu.tw/handle/123456789/706540摘要：在龐大的音樂資料庫中，如何搜尋自己想要的歌曲，是一個重要的研究議題。以情緒為依據的音樂分類與檢索是ㄧ個受人矚目的新方式，因為音樂情緒是直接與音樂的內容相關的，而且人們聆聽音樂的動機及反應與情感有極大的關係。我們在前期的研究中提出一個音樂情緒辨識系統，將音樂情緒定義在由激昂度與正向度兩個維度所構成之二維情緒平面上，廣受注意，非常適用於觸控式智慧型手機。未來我們想嘗試各種方法增進音樂情緒辨識功能，像是使用更貼近人類感知的中階特徵如旋律、文字介紹來辨識音樂情緒，或是使用更完善的方式來標定歌曲之情緒。這些重要研究項目將幫助我們提昇音樂情緒辨識的準確率，我們也準備應用這些音樂情緒辨識技術，發展一個創新的自動影片剪輯系統。本計畫分三年執行，主要內容包括(1)音樂旋律抽取，包含樂曲中主旋律與伴奏音樂的分離、旋律線辨識、旋律模型的建立；(2)文字標籤特徵抽取，包含文字標籤語意分析、文字標籤主題分類、文字標籤模型的建立；(3)歌曲標定方式改善，包含設計排序演算法、邀請受測者標定；(4)整合以上技術，增進音樂情緒辨識率；(5)利用音樂情緒辨識完成一個自動影片精采片段選輯系統; (6)利用音樂情緒辨識及雲端計算實現一個適用於智慧型觸控手機之音樂檢索系統。<br> Abstract: As the amount of content continues to explode in the digital age, we believe that the way music information is organized has to evolve in order to meet the ever increasing demand for easy and effective information access. We also believe music classification and retrieval by emotion is a plausible approach, because it is content-centric and functionally powerful. In the previous research period, we proposed a music emotion recognition (MER) approach that defines music in a two dimensional (arousal and valence) plane in our former research. As a promising approach for touch-based smart phones, it has received attention worldwide. Going forward, we plan to expand the scope and functionality of our MER system by using further mid-level features such as melody and text labels and by better ground truth collection. Building on top of our previous success, we also plan to develop an automatic video highlight extraction system using the resulting MER techniques. This three-year project includes five important parts: (1) music melody extraction, including separation of melody and accompaniment music, melody line identification, and melody modeling; (2) text label analysis, including label semantic analysis, topic classification, and label modeling; (3) ground truth collection refinement by a new sorting algorithm; (4) integration of the resulting techniques to enhance the accuracy of our MER system; (5) design of an automatic video highlight extraction system using the resulting MER techniques; (6) development of an emotion-based music retrieval service for smart phones by leveraging cloud computing.音樂搜尋音樂情緒辨識動影片精采片段選輯Music searchmusic emotion recognitionautomatic highlights extraction