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

Table 2

Errors of different model configurations in predicting phoneme length.

	RMSE			RMSE + cross entropy
Number of layers	DCNN	LSTM	Our model	DCNN	LSTM	Our model
1	6.35	5.81	5.82	5.31	4.82	4.76
2	5.93	5.46	5.48	5.13	4.51	4.34
3	5.74	5.23	5.24	4.86	4.35	4.23
4	5.89	5.65	5.19	4.92	4.53	4.18
5	5.82	5.67	5.81	4.89	4.59	4.47