集錦音樂自動作曲演算法之研究

Abstract

本文開發了一個音樂作曲的演算法，將來自於不同歌曲的音樂片段拼貼在一起而成為一段完整的音樂。音樂片段來源是現代流行歌曲。本文採用統計機率模型加上一些音樂理論為基礎的概念來設計演算法。在音樂理論上，和弦在音樂和諧度上扮演重要的角色，所以和弦是本論文主要去探討的特徵。和弦的轉換及音樂架構也是我們探討的對象。最後提出一套兩階層的貪婪演算法，包括和弦模型及節奏架構來達到拼貼作曲的效果。而在得出主旋律後，我們還用動態程式規劃設計一個以和弦為基礎的自動伴奏演算法來為主旋律搭配伴奏。最後實驗部份採取人為受訪的實驗方式，實驗結果顯示本系統完成的作品效果不錯。

This thesis presents a composition algorithm by sewing several unrelated music swatches into one piece of music, based on music harmony and several musical features. All segments in our system are from popular songs in modern life. Statistical concepts and music theory-based rules are applied in our algorithm. The chord pattern is the main feature for that chord plays an important role in music harmony. A training model of chord transition and music structure analysis are both proposed. The composition approach is a two-level greedy algorithm combining the chord model and rhythmic features. A chord-based automatic accompaniment algorithm using dynamic programming is also proposed in this paper. Subjective experiments were executed and the result shows that the music generated by our algorithm performs pretty well.