自動化作曲系統及其圖形化人機介面之研究

Abstract

本論文主要係研究電腦人工智慧作曲，或稱準則作曲，是音樂和電腦兩類專業知識所開發出的音樂自動創作系統。本系統運用馬可夫鏈 (Markov Chain)，其技術是按照一個轉換表來依次選擇功能和聲的連接；該轉換表就像一個函數，其引數是當前的和聲級數，而函數值則是下一個要出現和聲級數的可能性。轉換表可以針對某一特定作曲家或某一時期風格的音樂作品進行收集和統計，來構造出相應風格的轉換表。而此轉換表，定義了這些特定音樂風格的作品中，曲調進行與其和聲外音出現之可能性，並加入上下行音程的出現機率，成為本系統核心的旋律產生引擎。 此外本核心引擎亦結合篩選理論 (Sieve Theory) ，可將不需要的音篩選掉，來製造出教會調式、全音階以及中國五聲、七聲音階風格的旋律。本系統大量分析過去偉大音樂家的三個具代表性風格時期之作品，其方法可產生類比音樂家思考的人工智慧，並配合簡易上手的圖形化使用者介面，來達到使音樂創作平台可大眾化之理想。

This thesis is mainly the research of artificial intelligence composition, or so called algorithmic composition. It develops the automatic composition system integrating both fields with knowledge of music and computer. This system uses “Markov Chain”, adapting the technology of choosing the function harmony sequentially according to the conversion table. This conversion table is a function, and its input argument represents the current root of the chord. Finally the possible harmonic progressions are guided and counted with the function output. It can be collected with the proper analysis for the music works of some specific styles of the composers in a certain period with the conversion tables, to construct the next state of the table for the corresponding style. Somehow the works of these specific music styles and the possibility of the melodic lines with its non-chordal tones are well defined in the conversion table, with the probability of the interval up and down, to form the melody generator engine of the system. In addition, this engine has also combined the “Sieve Theory”, which can filter out the unwanted pitches, to create the melodic scales including Church Mode, Chinese pentatonic, and Chinese septatonic styles. This system analyzes tremendous works of the past great musicians’ works in three typical periods. The analytic methodology is to generate the artificial intelligence simulating the musicians’ thinking, with the user friendly graphic interface, to reach the ideal of establishing a popularized music composition platform.