| { | |
| "format_version": 1, | |
| "type": "vocoder", | |
| "model": { | |
| "encoder": { | |
| "latent_channels": 32, | |
| "hidden_channels": 192, | |
| "encoder_kernel_size": 5, | |
| "encoder_dilation_rate": 1, | |
| "encoder_layers": 16, | |
| "stft_hop_length": 320, | |
| "stft_win_length": 1280, | |
| "m2l_last_time_downsample_factor": 2 | |
| }, | |
| "decoder": { | |
| "latent_channels": 32, | |
| "stft_hop_length": 320, | |
| "stft_win_length": 1280, | |
| "m2l_last_time_downsample_factor": 2 | |
| } | |
| } | |
| } | |