Acquisition of Human Feelings in Music Arrangement

摘要

We often make decisions based on our feelings, which are implicit and very difficult to express as knowledge. This paper details an attempt to acquire feelings automatically. We assume that some relations or cnstraints exist between impressions felt and situations, which consist of an object and its environmetn. For example, in music arrangmeent, the object is a music score and its environemtn contains listeners, etc. Our proejct vaidates this assumption through three levels of experiments. At the first level, a program simply mimics human arrangements in order to transfer their impressions to another arrangement. This implies that the program is capable of distinguishing patterns that result in some impressions. At the second level, in order to produce a music recognition model, the program locates relations and constraitns between a music score and its impressions, by which we show that machine learning technqiues may provide a pwoerfu ltool for composing music and analyzing human feelings. Finally, we examine the generality of the model by modifying some arrangements to provide the subjects with a specified impression.