基于贝叶斯网研究自动作曲中音高的表示和推理

Representing and Reasoning Pitches in Algorithmic Composition Based on Bayesian Networks

自动作曲或称算法作曲是利用计算机进行自动或半自动的音乐创作过程。算法作曲的关键之一是生成音高。然而,不确定性是音乐本身固有的特征。贝叶斯网是不确定性知识的表示和推理的典型工具,已经成功应用到很多领域。在MIDI格式的基础上,利用贝叶斯网在算法作曲中生成音高,首先建立一个关于音高的贝叶斯网模型并基于此模型建立知识库。其次,基于贝叶斯网对音高进行推理,生成给定节拍处的每一个音的音高。实验表明,所提出的音高推理方法是可行的。

Algorithmic composition is the partial or total automatic process of music composition by a computer.One of the challenges in algorithmic composition is to create pitches.However,uncertainty is an intrinsic feature of music.Bayesian network（BN）is an effective and popular framework for representing and reasoning knowledge under uncertainty,and BNs have been successfully applied to a variety of problems.Based on MIDI format to create pitches in algorithmic composition,we firstly built a model about pitches with BNs and a knowledge base based on the model.Moreover,based on bayesian inference,the pitch of every note at each pat could be created.A preliminary experiment demonstrates empirically that this method for pitch inference is feasible.