Sound Classification Based on Multihead Attention and Support Vector Machine

Table 2

Classification accuracy on UrbanSound8K compared across different numbers of heads and layers with Feature 1 and Feature 2 individually.

Feature	Head (#)	L (#)	MhaNN accu. (%)	MhaNN-SVM accu. (%)	MhaNN-LR accu. (%)	MhaNN-KNN accu. (%)
Feature 1	2	1	91.6	92.1	92.3	91.5
		2	92.2	93.3	93.0	92.9
		3	91.6	93.3	91.7	92.2
	4	1	91.8	92.7	91.6	92.1
		2	92.1	93.6	92.8	93.2
		3	92.1	94.6	92.3	93.0
	8	1	91.4	93.2	91.0	92.9
		2	90.9	92.1	91.7	91.0
		3	90.5	91.0	90.8	91.2
Feature 2	2	1	83.7	84.8	86.1	85.2
		2	89.1	90.3	87.8	88.1
		3	86.2	87.4	86.1	86.8
	4	1	85.5	86.7	85.9	85.1
		2	87.1	89.7	87.2	88.4
		3	83.0	84.1	82.7	83.0