Musical Impression Extraction Method by Discovering Relationships between Acoustic Features and Impression Terms

In this paper, we represent an impression extraction method for music by relationship between acoustic features and impression terms. Our method extracts impression terms with weights from acoustic features extracted from music as wav file. We define the acoustic features as tempo, inter-onset interval, melody register, accompaniment register, and tonality. In this paper, we use 37 kinds of impression terms to describe musical impressions. We use a data set consisting of a music file and impression terms to create a model that relates acoustic features extracted from music with impression terms using clustering and TF-ICF (Term Frequency-Inversed Cluster Frequency). By using the model, our method can extract impression terms from music as wav file. We will realize a recommendation system according to impression by our method.