基于旋律结构的自动作曲研究

发布时间：2019-02-22 20:38

【摘要】：自20世纪50年代以来,人工智能在自动作曲方面取得了一些进展,主要方法包括:基于马尔科夫链和人工神经网络的相关机器学习方法,基于音乐规则和音乐文法的知识库系统,遗传进化算法,以及分层结构与细胞自动机理论等方法。这些方法都有不完善的地方,尤其是有关音乐结构性的研究方面非常薄弱,生成的乐曲不具备全局结构性,导致生成的音乐美感度较低,自动化程度不高。为了解决计算机自动作曲存在的这些问题,生成满足曲式要求的旋律,本文以旋律结构化为目标,结合19世纪西方古典音乐的一种分层次的和声结构递归简化的分析方法——申克分析的分析思路,并辅助和声学、旋律发展结构的相关乐理知识,生成了兼顾局部结构性与全局结构性的旋律。首先,结合申克分析的思想以及和声进行、音程、调式等音乐理论,利用产生式规则生成主干旋律的基本框架。接着利用和声规则等相关音乐理论将其转化为主干和弦,并利用和弦映射等方法将主干和弦转变为具体的音符序列,再添加上修饰润色的和弦外音,生成主题乐句,为旋律的局部结构化奠定基础。然后,通过模糊推理方法对旋律发展进行总体控制,利用设计的重复、音高偏移、乐句反转等规则化方法将主题乐句进行扩展,从而生成全局结构化的完整旋律。最后,搭建基于旋律结构的自动作曲实验系统平台,通过该平台生成若干首乐曲。通过对这些乐曲进行抽样人工评估和机器评估,发现该方法能输出兼顾局部结构性和全局结构性的旋律,提高了音乐美感度,从而提高了作曲的自动化程度。该研究在一定程度上解决了计算机智能作曲的单调结构性薄弱的问题,促进智能作曲在结构方面的进一步研究。
[Abstract]:Since the 1950s, artificial intelligence has made some progress in automatic composition, the main methods include: Markov chain and artificial neural network based on the related machine learning methods, Knowledge base system based on music rules and music grammar, genetic evolutionary algorithm, hierarchical structure and cellular automata theory. These methods are not perfect, especially the research on music structure is very weak, the generated music does not have a global structure, resulting in a low aesthetic degree of the generated music, not high degree of automation. In order to solve these problems existing in the automatic composition of music by computer and to generate melodies that meet the requirements of musical form, this paper aims at structuring melody. Combining with the analytic thought of Schenck analysis, a recursive and simplified analytical method of classical western music in the 19th century, and assisting with the relevant musical theory knowledge of the development structure of harmony and melody, The melody of local structure and global structure is generated. First of all, combining with Schenck's theory of analysis and harmony, interval, modulation and other musical theories, the basic framework of the main melody is generated by using the production rules. Then it is transformed into the main chord by using the harmony rules and other relevant music theories, and the main chord is transformed into the specific note sequence by using the method of chord mapping, and then the modified chord external sound is added to generate the theme phrase. Lay the foundation for the partial structure of melody. Then, the development of melody is controlled by fuzzy reasoning method, and the method of design repetition, pitch deviation and sentence inversion is used to expand the topic phrase to generate a global structured complete melody. Finally, an experimental system platform of automatic composition based on melody structure is built, through which several pieces of music are generated. Through the sampling manual and machine evaluation of these compositions, it is found that the method can output the melody which takes into account the local structure and the global structure, improves the aesthetic sensitivity of the music, and thus improves the automatization of composition. To a certain extent, this study solves the monotonous structural weakness of intelligent composition, and promotes the further research on the structure of intelligent composition.
【学位授予单位】：河南师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：J614.6;TP3;TP18

【参考文献】