Deep Learning-Based Amplitude Fusion for Speech Dereverberation

Table 1

Parameters setting up for DNN.

Parameter	DNN
Type	MLP
Number of hidden layers	3
Number of nodes in hidden layer	3072
Input feature	LMS
Input context	Contextual 7 frames
Output context	Contextual 7 frames
Activation function for hidden layer	ReLU
Activation function for output layer	Linear for MAP, IAM, and DCC sigmoid for IRM
Loss function	MSE
Optimizer	Adam
Learning rate	0.0002
Batch size	200
Number of training epochs	80