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

Table 1

Results before and after the addition of singing melodies.

		BAP distortion	F0 root mean square error (RMSE)	Labeling error of voiced/nonvoiced sound (LE)	Mel cepstral distance (MCD)
DCNN	Preaddition	0.248	12.375	5.459	5.135
	Postaddition	0.239	11.892	5.374	4.913
LSTM	Preaddition	0.241	12.841	5.621	5.092
	Postaddition	0.244	12.163	5.548	4.836
Our model	Preaddition	0.246	12.715	5.539	5.051
	Postaddition	0.248	12.734	5.568	4.993