Spoken Language Identification Using Deep Learning

Table 2

Architecture of 2D ConvNet model.

Layers	Filters, kernels, and stride	Padding	Output	No. of parameters
First block
Conv2D	(32, 7, 7)	Valid	(None, 994, 34, 32)	1600
BatchNorm			(None, 994, 34, 32)	128
MaxPool2D	(3, 3) s = 2	Same	(None, 497, 17, 32)	0
Second block
Conv2D	(64, 5, 5)	Same	(None, 497, 17, 64)	51264
BatchNorm			(None, 497, 17, 64)	256
MaxPool2D	(3, 3) s = 2	Same	(None, 249, 9, 64)	0
Third block
Conv2D	(128, 3, 3)	Same	(None, 249, 9, 128)	73856
BatchNorm			(None, 249, 9, 128)	512
MaxPool2D	(3, 3) s = 2	Same	(None, 125, 5, 128)	0
Fourth block
Conv2D	(256, 3, 3)	Same	(None, 125, 5, 256)	295168
BatchNorm			(None, 125, 5, 256)	1024
MaxPool2D	(3, 3) s = 2	Same	(None, 63, 3, 256)	0
Fifth block
Conv2D	(512, 3, 3)	Same	(None, 63, 3, 512)	1180160
BatchNorm			(None, 63, 3, 512)	2048
MaxPool2D	(3, 3) s = 2	Same	(None, 32, 2, 512)	0
Flatten layer			(None, 32768)	0
BatchNorm			(None, 32768)	131072
Dense layer	256		(None, 256)	8388864
BatchNorm			(None, 256)	1024
Dropout	0.5		(None, 256)	0
Dense layer	3		(None, 3)	771