|
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, |
|
sampling_rate=22050, |
|
**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 |
|
self.sampling_rate = sampling_rate |
|
super().__init__(**kwargs) |
|
|