| from transformers import PretrainedConfig | |
| class HiFiGANConfig(PretrainedConfig): | |
| model_type = "hifigan" | |
| def __init__( | |
| self, | |
| resblock_kernel_sizes=[3, 7, 11], | |
| resblock_dilation_sizes=[[1, 3, 5], [1, 3, 5], [1, 3, 5]], | |
| upsample_rates=[8, 8, 2, 2], | |
| upsample_initial_channel=512, | |
| upsample_kernel_sizes=[16, 16, 4, 4], | |
| model_in_dim=80, | |
| **kwargs, | |
| ): | |
| self.resblock_kernel_sizes = resblock_kernel_sizes | |
| self.resblock_dilation_sizes = resblock_dilation_sizes | |
| self.upsample_rates = upsample_rates | |
| self.model_in_dim = model_in_dim | |
| self.upsample_initial_channel = upsample_initial_channel | |
| self.upsample_kernel_sizes = upsample_kernel_sizes | |
| super().__init__(**kwargs) | |