Intelligent Music System Based On Bayesian Network For Assisted Composing And Arranging

Music composing and arranging is a complex process. From the beginning of the melody composing to arranging, it takes a lot of time. However, There is a lot of repetitive, tedious and mechanical work in music composition. In order to shorten the time of composing, algorithmic composition partially or totally automates the process of music composing by using computers. Moreover, 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. Therefore, considering the key characteristics of music rhythm, pitch and dynamic as well as the uncertainty of music, the paper will investigate the method of automatic composition based on Bayesian network.The main contributions and novelties of this thesis include the following three aspects.Firstly, based on Bayesian network this paper focuses on the methods of modeling and reasoning music rhythm, pitch, and dynamic. Based on MIDI format, we will build a model about rhythm, pitch and dynamics with BNs. Then, we build a knowledge base based on the models. Rhythm and pitch of every note at each pat could be created. Then we will show the inference methods of rhythm and pitch, and propose a method of semi-supervised music composition, and rhythm and pitch of every note at each pat could be created.Secondly, We propose an algorithm of words screening in two levels and based on Naive Bayesian network, corpus of knowledge base and procedure in database designed a human-computer interaction model to provide interation between composers and music creation.Thirdly, by constructing a multilayer naive Bayesian network and extending the MVC model we program a music composition system to provide the composer with the chance to interact with the computer in composing music. We use the system to do music composition experiment. A preliminary experiment demonstrates empirically that this method for pitch inference is feasible.