|
--- |
|
tags: |
|
- espnet |
|
- audio |
|
- audio-to-audio |
|
- vocoder |
|
language: |
|
- en |
|
datasets: |
|
- vctk |
|
license: cc-by-4.0 |
|
inference: false |
|
--- |
|
|
|
## Vocoder model - HifiGAN - English |
|
|
|
https://github.com/kan-bayashi/ParallelWaveGAN |
|
|
|
**No support given.** |
|
|
|
### Details |
|
|
|
``` |
|
sampling_rate: 44100 # Sampling rate. |
|
fft_size: 2048 # FFT size. |
|
hop_size: 512 # Hop size. |
|
win_length: 2048 # Window length. |
|
# If set to null, it will be the same as fft_size. |
|
window: "hann" # Window function. |
|
num_mels: 80 # Number of mel basis. |
|
fmin: 0 # Minimum freq in mel basis calculation. |
|
fmax: 22050 # Maximum frequency in mel basis calculation. |
|
generator_type: HiFiGANGenerator |
|
generator_params: |
|
in_channels: 80 # Number of input channels. |
|
out_channels: 1 # Number of output channels. |
|
channels: 512 # Number of initial channels. |
|
kernel_size: 7 # Kernel size of initial and final conv layers. |
|
upsample_scales: [8, 8, 2, 2, 2] # Upsampling scales. |
|
upsample_kernel_sizes: [16, 16, 4, 4, 4] # Kernel size for upsampling layers. |
|
``` |
|
|
