Music Emotion Maps in Arousal-Valence Space
In this article we present the approach in which the detection of emotion is modeled by the pertinent regression problem. Conducting experiments required building a database, annotation of samples by music experts, construction of regressors, attribute se
- PDF / 1,364,203 Bytes
- 10 Pages / 439.37 x 666.142 pts Page_size
- 9 Downloads / 213 Views
·
Audio feature
Introduction
Emotions are a dominant element in music, and they are the reason people listen to music so often [12]. Systems searching musical compositions on Internet databases more and more often add an option of selecting emotions to the basic search parameters, such as title, composer, genre, etc. The emotional content of music is not always constant, and even in classical music or jazz changes often. Analysis of emotions contained in music over time is a very interesting aspect of studying the content of music. It can provide new knowledge on how the composer emotionally shaped the music or why we like some compositions more than others. Music emotion recognition, taking into account the emotion model, can be divided into categorical or dimensional. In the categorical approach, a number of emotional categories (adjectives) are used for labeling music excerpts. It was presented in the following papers [5,6,11]. In the dimensional approach, emotion is described using dimensional space - 2D or 3D. Russell [13] proposed a 2D model, where the dimensions are represented by arousal and valence; used in [15,18]. The 3D model of Pleasure-Arousal-Dominance (PAD) was used in [3,10]. Music emotion recognition concentrates on static or dynamic changes over time. Static music emotion recognition uses excerpts from 15 to 30 s and omits changes in emotions over time. It assumes the emotion in a given segment does not change. Dynamic music emotion recognition analyzes changes in emotions c IFIP International Federation for Information Processing 2016 Published by Springer International Publishing Switzerland 2016. All Rights Reserved K. Saeed and W. Homenda (Eds.): CISIM 2016, LNCS 9842, pp. 697–706, 2016. DOI: 10.1007/978-3-319-45378-1 60
698
J. Grekow
over time. Methods for detecting emotion using a sliding window are presented in [9,11,15,18]. Deng and Leung [3] proposed multiple dynamic textures to model emotion dynamics over time. To find similar sequence pasterns of musical emotions, they used subsequence dynamic time warping for matching emotion dynamics. Aljanaki et al. [1] investigated how well structural segmentation explains emotion segmentation. They evaluated different unsupervised segmentation methods on the task of emotion segmentation. Imbrasaite et al. [7] and Schmidt et al. [14] used Continuous Conditional Random Fields for dimensional emotion tracking. In our study, we used dynamic music emotion recognition with a sliding window. We experimentally selected a segment length of 6 s as the shortest period of time after which a music expert can recognize an emotion. The rest of this paper is organized as follows. Section 2 describes the music annotated data set and the emotion model used. Section 3 presents features extracted by using tools for audio analysis. Section 4 describes regressor training and their evaluation. Section 5 presents the results of emotion tracking. Finally, Sect. 6 summarizes the main findings.
2
Music Data
The data set that was annotated consisted of 324 six-second
Data Loading...
