Table 1 Bold words represent the three parts of our U2-VC as noted in section 3. “I” denotes the input size. “O” denotes the output size. “K” denotes the kernel size. “S” denotes the stride. “D” denotes the dilation. “H” denotes the hidden size. “L” denotes the layer. As stated before, residual U-block has the same structure as original residual U-block except the input layer and the reshaping operation
From: U2-VC: one-shot voice conversion using two-level nested U-structure
Encoder | Decoder | |
|---|---|---|
Conv1D IN I:80 O:256 K:1 ×1 | Conv1D IN I:4 O:256 K:3 ×3 | |
Conv1D OUT I:256 O:4 K:1 ×1 | ||
1-2-1RSU7 | Conv1D IN I:256 O:256 K:3 ×3 | Conv1D IN I:256 O:256 K:3 ×3 |
Conv2D EN1 I:1 O:16 K:3 ×3 | Conv2D EN1 I:1 O:16 K:3 ×3 | |
Conv2D EN2∼6 I:16 O:16 K:3 ×3 | Conv2DEN EN2∼6 I:16 O:16 K:3 ×3 | |
MaxPool2D K:3 ×3 S:2 | MaxPool2D K:3 ×3 S:2 | |
Conv2D EN7 I:16 O:16 K:3 ×3 D:2 | Conv2D EN7 I:16 O:16 K:3 ×3 D:2 | |
Conv2D DE6∼2 I:32 O:32 K:3 ×3 | Conv2D DE6∼2 I:32 O:32 K:3 ×3 | |
Conv2D DE1 I:32 O:1 K:3 ×3 | Conv2D DE1 I:32 O:1 K:3 ×3 | |
1-2-1RSU6 | Conv1D IN I:256 O:256 K:3 ×3 | Conv1D IN I:256 O:256 K:3 ×3 |
Conv2D EN1 I:1 O:16 K:3 ×3 | Conv2D EN1 I:1 O:16 K:3 ×3 | |
MaxPool2D K:3 ×3 S:2 | MaxPool2D K:3 ×3 S:2 | |
Conv2DEN EN2∼5 I:16 O:16 K:3 ×3 | Conv2DEN EN2∼5 I:16 O:16 K:3 ×3 | |
Conv2D EN6 I:16 O:16 K:3 ×3 D:2 | Conv2D EN6 I:16 O:16 K:3 ×3 D:2 | |
Conv2D DE5∼2 I:32 O:32 K:3 ×3 | Conv2D DE5∼2 I:32 O:32 K:3 ×3 | |
Conv2D DE1 I:32 O:1 K:3 ×3 | Conv2D DE1 I:32 O:1 K:3 ×3 | |
1-2-1RSU5 | Conv1D IN I:256 O:256 K:3 ×3 | Conv1D IN I:256 O:256 K:3 ×3 |
Conv2D EN1 I:1 O:16 K:3 ×3 | Conv2D EN1 I:1 O:16 K:3 ×3 | |
MaxPool2D K:3 ×3 S:2 | MaxPool2D K:3 ×3 S:2 | |
Conv2D EN2∼4 I:16 O:16 K:3 ×3 | Conv2D EN2∼4 I:16 O:16 K:3 ×3 | |
Conv2D EN5 I:16 O:16 K:3 ×3 D:2 | Conv2D EN5 I:16 O:16 K:3 ×3 D:2 | |
Conv2D DE4∼2 I:32 O:32 K:3 ×3 | Conv2D DE4∼2 I:32 O:32 K:3 ×3 | |
Conv2D DE1 I:32 O:1 K:3 ×3 | Conv2D DE1 I:32 O:1 K:3 ×3 | |
1-2-1RSU4 | Conv1D IN I:256 O:256 K:3 ×3 | Conv1D IN I:256 O:256 K:3 ×3 |
Conv2D EN1 I:1 O:16 K:3 ×3 | Conv2D EN1 I:1 O:16 K:3 ×3 | |
MaxPool2D K:3 ×3 S:2 | MaxPool2D K:3 ×3 S:2 | |
Conv2D EN2∼3 I:16 O:16 K:3 ×3 | Conv2D EN2∼3 I:16 O:16 K:3 ×3 | |
Conv2D EN4 I:16 O:16 K:3 ×3 D:2 | Conv2D EN4 I:16 O:16 K:3 ×3 D:2 | |
Conv2D DE3∼2 I:32 O:32 K:3 ×3 | Conv2D DE3∼2 I:32 O:32 K:3 ×3 | |
Conv2D DE1 I:32 O:1 K:3 ×3 | Conv2D DE1 I:32 O:1 K:3 ×3 | |
1-2-1RSU4F | Conv1D IN I:256 O:256 K:3 ×3 | Conv1D IN I:256 O:256 K:3 ×3 |
Conv2D EN1 I:1 O:16 K:3 ×3 | Conv2D EN1 I:1 O:16 K:3 ×3 | |
Conv2D EN2 I:16 O:16 K:3 ×3 D:2 | Conv2D EN2 I:16 O:16 K:3 ×3 D:2 | |
Conv2D EN3 I:16 O:16 K:3 ×3 D:4 | Conv2D EN3 I:16 O:16 K:3 ×3 D:4 | |
Conv2D EN4 I:16 O:16 K:3 ×3 D:8 | Conv2D EN4 I:16 O:16 K:3 ×3 D:8 | |
Conv2D DE3 I:32 O:32 K:3 ×3 D:4 | Conv2D DE3 I:32 O:32 K:3 ×3 D:4 | |
Conv2D DE2 I:32 O:32 K:3 ×3 D:2 | Conv2D DE2 I:32 O:32 K:3 ×3 D:2 | |
Conv2D DE1 I:32 O:1 K:3 ×3 | Conv2D DE1 I:32 O:1 K:3 ×3 |