音樂旋律意象感知在自動作曲之分析研究

Abstract

音樂是由旋律、節奏與和聲等三個要素交互作用所組成的，尤其以旋律要素最為重要，我們歸納構成音樂旋律的兩個重要因子為節奏與調式，並作進一步進行音樂訊號聽覺實驗，分析不同旋律所造成的音樂意象感知。希望藉由了解調式與節奏來分析旋律的多元化。在本篇論文中，我們以自動作曲為理論基礎，並以感性工程為實作方式，結合多變量統計以及其他統計相關理論和方法，經由資料轉換到法則的建立，建構出一套感性工學資訊系統音樂評估描述方案。有關聽覺實驗的進行，實驗前我們藉由文獻資料與專家的協助，挑選39個形容詞對進行旋律意象調查，並以馬可夫鍊分析現有的調式音樂來建構中國的五聲調式的轉換表，並以 Metric Complexity 節奏複雜度建構節奏；最後使用自動作曲系統依照不同的調式與節奏交互而成產生30種音樂樣本，針對35位受測者進行音樂旋律意象的聽覺語意實驗。實驗資料藉項目分析與信度檢定篩檢無效樣本與選項，最後得到 Cronbach α 為0.887的良好信度結果與14組形容詞對；由因素分析歸納可以出三個意象因子，分別是熱情冷淡、優美醜陋、浪漫死板。利用多變量變異數分析來找尋調式與節奏複雜度對14組形容詞對的關聯，並繪製出估計平均數。再利用多元尺度法繪製出兩向度的空間知覺意象圖來建構資料庫。由實驗的各種結果指出，我們發現調式與節奏皆可以有效的協助音樂家或電腦作曲系統創作具有不同感知意象旋律。

Many researches review that music is composed by three factors which are melody, rhythm and harmony, and melody is the most important of all. We concluded that rhythm and mode are contributed to musical melody, and then makes these two factors as variable to do the musical auditory experiments. In this research, the theoretical base is automatic composition and we use kansei engineering as the experiment method. Therefore, we can construct a subject of kansei engineering system musical analysis by it. The part of hearing experiences, before it we selected 39 adjectives for 35 testers by document research and assistances of experts. We applied markov chains to analyze tonal music to construct markov chains tables of pentatonic mode, then use metric complexity to contribute rhythm. At last, we adopted computer automatic composition system to generate samples from 5 sorts of pentatonic mode and 6 kinds of rhythm complexity, then producted 30 varities music samples.The results of the experience, we can conclude three image perception factors by Factor Analysis, which are passionate - indifferent、beautiful - ugly、romantic - rigid. Moreover, we use MANOVA and MDS to get image perception scale at the same time. Above all, we realized mode and rhythm can help musicians and computer compose system to create difference melody affective efficiently.