Long Short-Term Memory Projection Recurrent Neural Network Architectures for Piano’s Continuous Note Recognition

Table 1

The number of iterations and the recognition rates of different models.

Model	Number of iterations	Recognition rate
LSTM-20	300	77.2%
LSTM-40	300	87.1%
LSTM-80	300	90.7%
LSTM-160	400	82.5%
DLSTM-two layers	61	85.7%
DLSTM-three layers	71	99.5%
DLSTM-four layers	109	100.0%
DLSTM-five layers	76	97.5%
DLSTM-six layers	103	100.0%
LSTMP-10 to 20	43	56%
LSTMP-30 to 20	35	94.2%
LSTMP-40 to 20	45	96.0%
LSTMP-40 to 30	30	94.0%
LSTMP-60 to 30	22	98.0%
LSTMP-80 to 20	58	99.8%
LSTMP-80 to 30	27	97.3%
LSTMP-80 to 40	29	97.3%
LSTMP-160 to 40	39	99.0%
LSTMP-160 to 80	50	95.0%
LSTMP-320 to 160	93	93.4%