Academia.eduAcademia.edu

Figure 4 – uploaded by Konstantinos Drossos

See full PDFdownloadDownload figure
Figure 4: The images used for arousal estimation, from lower to higher, as proposed in [38] Figure 5: The images used for valence estimation, from lower to higher, as proposed in [38]

Figure 4 The images used for arousal estimation, from lower to higher, as proposed in [38] Figure 5: The images used for valence estimation, from lower to higher, as proposed in [38]

Related Figures (7)

Figure 1: The list of adjectives with 8 classes ac- cording to [26, 27] The list of adjectives proposed by [26] represents an alter- native discrete emotions modeling approach. Emotions are not referred as single words, like in the case of basic emo- tions. Instead, a number of word groups is used, with most words in the same group to be synonyms [27]. These groups serve as a way to refer to and model emotions or emotional states. Each group is denoted as a class and consists of words capable for describing a similar emotional condition. The number of classes originally used was eight, but some researches extend it through additional classes, reaching a total of thirteen [28]. This model is frequently used in music psychology assessments. In Figure 1, a visual representation of the original eight classes is provided, as proposed by [26].
Figure 1: The list of adjectives with 8 classes ac- cording to [26, 27] The list of adjectives proposed by [26] represents an alter- native discrete emotions modeling approach. Emotions are not referred as single words, like in the case of basic emo- tions. Instead, a number of word groups is used, with most words in the same group to be synonyms [27]. These groups serve as a way to refer to and model emotions or emotional states. Each group is denoted as a class and consists of words capable for describing a similar emotional condition. The number of classes originally used was eight, but some researches extend it through additional classes, reaching a total of thirteen [28]. This model is frequently used in music psychology assessments. In Figure 1, a visual representation of the original eight classes is provided, as proposed by [26].
Figure 3: The AV plane with four emotions pre- sented as clusters
Figure 3: The AV plane with four emotions pre- sented as clusters
Figure 2: AV plane representation with indication to corresponding emotions The center of the axes in a dimensional model can be mapped to the neutral affective state or to the extremes of each di- mension. In the proposal of the circumplex model of affect [31] for example, which served as the basis for the dimen- sional approach, the center of axes was defined as the neutral state. Figure 2 illustrates the AV plane with emotions de- fined as specific points in the corresponding plane [19, 31, 32, 33]. These points can serve as center points during the AV values clustering process. On the other hand, Figure 3 provides a graphical representation of the AV plane with four emotions mapped to clusters of AV values [19, 33].
Figure 2: AV plane representation with indication to corresponding emotions The center of the axes in a dimensional model can be mapped to the neutral affective state or to the extremes of each di- mension. In the proposal of the circumplex model of affect [31] for example, which served as the basis for the dimen- sional approach, the center of axes was defined as the neutral state. Figure 2 illustrates the AV plane with emotions de- fined as specific points in the corresponding plane [19, 31, 32, 33]. These points can serve as center points during the AV values clustering process. On the other hand, Figure 3 provides a graphical representation of the AV plane with four emotions mapped to clusters of AV values [19, 33].
for arousal and valence groups are shown in Figures 4 and 5 respectively.
for arousal and valence groups are shown in Figures 4 and 5 respectively.
Figure 7: Illustration of the emotion recognition process followed testing set against the remaining 90% which was used as training dataset. Both training and testing was carried out using Weka [44]. Figure 7 graphically illustrates the above process.
Figure 7: Illustration of the emotion recognition process followed testing set against the remaining 90% which was used as training dataset. Both training and testing was carried out using Weka [44]. Figure 7 graphically illustrates the above process.
Prior to emotion recognition from the considered sound ev- ents, a pre-processing stage was carried out. All available audio files were resampled at a 16 kHz sampling frequency. From the annotation values, arousal and valence were used and these were clustered in four classes. The correspondence of each class with the annotated arousal and valence values is presented in Figure 6. Figure 6: Mapping of AV values to the classes used in the emotion recognition process
Prior to emotion recognition from the considered sound ev- ents, a pre-processing stage was carried out. All available audio files were resampled at a 16 kHz sampling frequency. From the annotation values, arousal and valence were used and these were clustered in four classes. The correspondence of each class with the annotated arousal and valence values is presented in Figure 6. Figure 6: Mapping of AV values to the classes used in the emotion recognition process
Table 1: 10-Fold cross validation results for each of the algorithms used Figure 8: Arousal Vs Normalised energy values for the IADS database
Table 1: 10-Fold cross validation results for each of the algorithms used Figure 8: Arousal Vs Normalised energy values for the IADS database
Related topics:
Music and EmotionsAcoustic EcologySound and EmotionsSound EventsSound Emotion RecognitionAudio Emotion RecognitionGeneral Sounds
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved