葉梅珍Yeh, Mei-Chen張書堯Chang, Shu-Yao2019-09-052017-08-102019-09-052017http://etds.lib.ntnu.edu.tw/cgi-bin/gs32/gsweb.cgi?o=dstdcdr&s=id=%22G060347041S%22.&%22.id.&http://rportal.lib.ntnu.edu.tw:80/handle/20.500.12235/106421電腦編曲是被研究已久的題目，累積了相當多的研究方法，其中電腦變奏因其多變性，至今沒有一套系統能夠解釋所有的變奏，因此找出音樂變奏的通則成為一項艱鉅的挑戰。在變奏方法不唯一的情況下，此論文根據樂理提出一個方法，我們考慮作曲家創作的手法，採用最小重複單位。而節奏的最小重複單位為動機或主題，但因為存在計算誤差，加上考慮現代音樂節奏多採用相似並帶有變化性重複， 因此採用小節作為重複的基礎單位。我們觀察節拍之間轉移的機率和小節之間的轉移機率，建立了馬爾可夫模型，計算在輸入音樂節奏的結構框架限制下，找出最合適的節奏片段作為節奏變奏。Computer composition has been studied for a long time, in which many methods have been proposed. In particular, in the task of computational variation, finding a general procedure to explain all types of variation is challenging due to the lack of the definition of variation. There are many methods to generate variation, but none of each can be applied to create all types of variation. In this thesis, we propose a systematic method to generate rhythmic variation based on theoretic composition. More specifically, we use the concept of the minimum repeated pattern in rhythm (theme or motif repeats in various way). Although using theme and motif can generate variation, we use the meter to be a minimum repeated pattern, considering the error caused by motif extraction in modern technique of composition. We further compute the probability of beat and rhythm transitions and build the Markov model, upon which we find frequent rhythm patterns (i.e., paths in the model) under the music structure constraints. Finally, we propose a method to evaluate the performance of the proposed style transfer system.音樂變奏電腦編曲機率模型馬爾可夫模型Music variationcomputer compositionprobabilistic modelMarkov model