표 2 / Table 2 실험에 사용된 음향모델의 세부구조 / Detailed structure of acoustic models for experiments
Model
Input
Convolution
Pooling
FCs
CNN
40×11
-
-
2,048*6
VDCNN
64×17
[3×3]*10
2,048*4
DenseNet
64×17
[3×3]*30[1×1]*35
[2×1]*2[2×2]*3
-
MR-DenseNet
64×17
[3×3]*12 [1×1]*15
[3×3]*18 [1×1]*20
[2×1]*2[2×2]*1
[2×2]*2
-
64×8
[3×3]*12 [1×1]*15
[2×1]*3
CNN, convolutional neural network; VDCNN, very deep convolutional neural network; MR-DenseNet, multi-resolution DenseNet