Table 12. CochCNN9 architecture.

Input (cochleagram)	(1,211,390)
BatchNorm2d_1 [1]	(1,211,390)
Conv2d_1(1, 96, kernel_size = [7, 14], stride = [3, 3], padding = ’same’)	[96, 71, 130]
ReLU_1	[96, 71, 130]
MaxPool2d_1(kernel_size = [2,5], stride = [2,2], padding = ’same’)	[96, 36, 65]
BatchNorm2d_2 [96]	[96, 36, 65]
Conv2d_2(96, 256, kernel_size = [4,8], stride = [2,2], padding = ’same’)	(256, 18, 33)
ReLU_2	(256, 18, 33)
MaxPool2d_2(kernel_size = [2,5], stride = [2,2], padding = ’same’)	(256, 9, 17)
BatchNorm2d_3 (256)	(256, 9, 17)
Conv2d_3 (512, kernel_size = [2,5], stride = [1,1], padding = ’same’)	(512, 9, 17)
ReLU_3	(512, 9, 17)
Conv2d_4 (1024, kernel_size = [2,5], stride = [1,1], padding = ’same’)	(1,024, 9, 17)
ReLU_4	(1,024, 9, 17)
Conv2d_5 (512, kernel_size = [2,5], stride = [1,1], padding = ’same’)	(512, 9, 17)
ReLU_5	(512, 9, 17)
AvgPool_1 (kernel_size = [2,5], stride = [2,2], padding = ’same’)	(512, 5, 9)
Linear_1	(4,096)
ReLU_6	(4,096)
Dropout_1 (p = 0.5)	(4,096)
Linear_2	(num_classes)