Analysis and Optimization of Flute Playing and Teaching System Based on Convolutional Neural Network

Table 2

Performance of recognition model when two-layer network and filter bank are used to extract the primary features of audio, respectively.

Primary feature extraction method	Frame length	Accuracy p (%)	Recall rate R (%)	F1 score (%)
Double-layer ReLu network	2047	60.92	69.95	65.11
	8190	65.74	84.94	74.12
Logarithmic frequency domain filter bank	2047	73.45	72.96	73.22
	8191	75.88	79.81	77.82
Commercial software melody	—	66.72	71.77	69.12