Automatic Synthesis Technology of Music Teaching Melodies Based on Recurrent Neural Network

Table 6

Prediction errors of different models in fundamental frequency and spectrum.

Model		DCNN	LSTM	Our model
F0 RMSE	Natural melody	35.21	21.79	20.45
	Score notes	34.68	19.36	8.59
F0 correlation	Natural melody	0.85	0.94	0.93
	Score notes	0.86	0.95	0.98
LE			2.34	2.38
		4.34	3.59	3.53