A method for finding the relationship between music mood and the information of time and frequency by music mood classification using spectrogram images

宮谷 大輝, 中山 英樹

Recently, with the development of the Internet and various music devices, it has become harder to choose appropriate music suited to particular scenes from enormous amount of songs because of diverse listening cases. Therefore, demand of music mood classification has been increasing. However, being different from other music classification tasks, mood classification is known to have less-obvious relation to musical features. Thus, in this research we estimate music mood with raw audio information, and propose method to extract influential elements in music impression. Firstly, we use visual features of spectrogram image generated audio information. Additionally, with analysis using the Fisher weight map of features extracted in each frequency, it becomes apparent how much the features effect the mood discrimination. Furthermore, dividing a music into some parts and expression one feature integrated each feature for a length of time, we identify the time period which effects powerfully on whole music mood from each probability. Finally, we verify the validity of analysis with our method through a subjective evaluation experiment.

A method for finding the relationship between music mood and the information of time and frequency by music mood classification using spectrogram images

Bibliographic Information

Search this article

Description

Journal

Keywords

Details 詳細情報について

Export

Report a problem