Sound Classification Based on Multihead Attention and Support Vector Machine

Table 5

Classification accuracy on GTZAN 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	81.8	82.9	81.6	82.2
		2	82.9	84.0	81.3	83.4
		3	81.2	81.7	79.2	82.0
	4	1	82.3	83.1	82.0	83.3
		2	85.4	88.4	84.7	86.7
		3	84.2	86.1	85.5	84.8
	8	1	82.7	84.8	82.7	83.1
		2	83.6	85.1	83.3	84.1
		3	81.2	83.2	78.7	80.1
Feature 2	2	1	70.1	72.2	70.3	72.0
		2	76.5	78.7	77.0	74.8
		3	72.5	74.6	70.8	72.2
	4	1	71.0	73.3	72.7	70.9
		2	75.1	78.0	76.2	75.8
		3	73.7	75.3	73.6	72.1