Table 1.

Configuration of ResNet-BiLSTM architectures

Layers	ResNet-BiLSTM
Input layer	mag spectrogram (N X 257)
ResNet Block(1)	$\left\{\begin{array}{c}Conv2D-filters=16,strides=(1,1)\hfill \\ BatchNormalization()\hfill \\ Conv2D-filters=16,strides=(1,1)\hfill \\ Conv2D-filters=16,strides=(1,3)\hfill \end{array}\right)$
ResNet Block(2)	$\left\{\begin{array}{c}Conv2D-filters=32,strides=(1,1)\hfill \\ BatchNormalization()\hfill \\ Conv2D-filters=32,strides=(1,1)\hfill \\ Conv2D-filters=32,strides=(1,3)\hfill \end{array}\right)$
ResNet Block(3)	$\left\{\begin{array}{c}Conv2D-filters=64,strides=(1,1)\hfill \\ BatchNormalization()\hfill \\ Conv2D-filters=64,strides=(1,1)\hfill \\ Conv2D-filters=64,strides=(1,3)\hfill \end{array}\right)$
ResNet Block(4)	$\left\{\begin{array}{c}Conv2D-filters=128,strides=(1,1)\hfill \\ BatchNormalization()\hfill \\ Conv2D-filters=128,strides=(1,1)\hfill \\ Conv2D-filters=128,strides=(1,3)\hfill \end{array}\right)$
Recurrent layer	BiLSTM-128
	Attention
FC layer	FC-128
	ReLU
	dropout(0.3)
	FC-1
Output layer	Average pool

The number of frames is denoted by N