Table 3. Size of representation at each layer for best-performing architecture of each network type (spatial × temporal × filter dimensions).

Layer	Dimension	Layer	Dimension	Layer	Dimension
Input	25 × 320 × 2	Input	25 × 320 × 2	Input	25 × 320 × 2
SC0	13 × 320 × 8	STC0	13 × 160 × 8	SC0	25 × 320 × 8
SC1	7 × 320 × 16	STC1	7 × 80 × 8	SC1	25 × 320 × 16
SC2	4 × 320 × 16	STC2	4 × 40 × 32	SC2	25 × 320 × 16
SC3	2 × 320 × 32	STC3	2 × 20 × 64	R	256 × 320
TC0	2 × 107 × 32
TC1	2 × 36 × 32
TC2	2 × 12 × 64
TC3	2 × 4 × 64