Sound Classification Based on Multihead Attention and Support Vector Machine

Table 8

Classification accuracy on IEMOCAP compared across different numbers of heads and layers with Feature1 and Feature2 individually.

Feature	Head (#)	L (#)	MhaNN accu. (%)	MhaNN-SVM accu. (%)	MhaNN-LR accu. (%)	MhaNN-KNN accu. (%)
Feature 1	2	1	53.3	55.3	54.0	52.2
		2	54.8	56.8	55.0	55.5
		3	51.9	53.3	52.3	49.8
	4	1	56.5	58.1	55.7	51.5
		2	54.9	56.8	55.1	54.6
		3	52.7	54.9	52.0	53.3
	8	1	55.6	57.3	56.0	54.9
		2	56.1	58.2	57.4	56.7
		3	52.1	54.2	55.0	53.7
Feature 2	2	1	56.1	58.8	56.1	53.9
		2	58.5	61.1	58.7	54.8
		3	57.8	60.1	57.7	55.5
	4	1	60.5	62.8	58.0	56.2
		2	58.3	60.4	58.2	56.9
		3	57.7	59.5	58.1	55.6