Academia.eduAcademia.edu

Figure 1 – uploaded by Rajani Muraleedharan

See full PDFdownloadDownload figure
Table 1. Hybrid kernel selection based on classifier-level recall (%). Table | shows the CL-recall values for the different ker- nels. For each OAA classifier, we choose the kernel with the highest CL-recall (numbers in bold in Table 1). For exam- ple, we choose the polynomial kernel for the ‘Happy or Not’ OAA classifier in our hybrid kernel approach. Note that when training individual SVM OAA classifiers, we make sure that there are a comparable number of training utterances for both classes, otherwise, the trained model will provide biased clas- sification results. For example, when training the ‘Happy or Not’ OAA classifier, ‘not happy’ utterances consist of all of the other five emotions. Therefore, we need to sample utter- ances from the five emotions with the same number as that of happy utterances. The selected kernel functions for the indi- vidual OAA classifiers are shown in Table 1.

Table 1 Hybrid kernel selection based on classifier-level recall (%). Table | shows the CL-recall values for the different ker- nels. For each OAA classifier, we choose the kernel with the highest CL-recall (numbers in bold in Table 1). For exam- ple, we choose the polynomial kernel for the ‘Happy or Not’ OAA classifier in our hybrid kernel approach. Note that when training individual SVM OAA classifiers, we make sure that there are a comparable number of training utterances for both classes, otherwise, the trained model will provide biased clas- sification results. For example, when training the ‘Happy or Not’ OAA classifier, ‘not happy’ utterances consist of all of the other five emotions. Therefore, we need to sample utter- ances from the five emotions with the same number as that of happy utterances. The selected kernel functions for the indi- vidual OAA classifiers are shown in Table 1.

Related Figures (3)

Fig. 1. The proposed emotion classification approach using OAA SVM with hybrid kernel and thresholding fusion.
Fig. 1. The proposed emotion classification approach using OAA SVM with hybrid kernel and thresholding fusion.
Table 2. a) Comparison of classifier-level accuracy (%) of the results in [2] and the proposed approach; b) decision-level recall (%) of the pro- posed approach.
Table 2. a) Comparison of classifier-level accuracy (%) of the results in [2] and the proposed approach; b) decision-level recall (%) of the pro- posed approach.
Related topics:
Computer ScienceArtificial IntelligenceAutomatic Speech Recognitionemotion classification
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved