Skip to main content
EURASIP Journal on Audio, Speech, and Music Processing
Fig. 16 | EURASIP Journal on Audio, Speech, and Music Processing

Fig. 16

From: Piano multipitch estimation using sparse coding embedded deep learning

Fig. 16

Structure of the recurrent network used in AMT experiments. The inputs are 16 time steps before and after current label. The number of hidden state of GRUs is 200, the output numbers of InnerProducts are all 200 except the last one is 88. Before concatenation, we only keep the last step’s output of each GRU because they accumulate all the information through time. Legends can be referred to Fig. 1

Back to article page