Speaker Gender Recognition Based on Deep Neural Networks and ResNet50

Table 2

ResNet 34 and ResNet 50 detailed architecture.

Layer name	Output size	34 layers	50 layers
Conv 1		, 64, stride 2
Conv 2.x		max pool, stride 2
		Conv 3.x
Conv 4.x		[3×3,256 3×3,256]×6
Conv 5.x
FLOPs		Average pool, 1000-d fc, softmax