Spaces:
Sleeping
Sleeping
File size: 1,610 Bytes
7e4b981 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 |
# import sys
# sys.path.append('tacotron2')
import torch
from .layers import STFT
class Denoiser(torch.nn.Module):
""" Removes model bias from audio produced with waveglow """
def __init__(self, waveglow, filter_length=1024, n_overlap=4,
win_length=1024, mode='zeros'):
super(Denoiser, self).__init__()
self.stft = STFT(filter_length=filter_length,
hop_length=int(filter_length/n_overlap),
win_length=win_length).cuda()
if mode == 'zeros':
mel_input = torch.zeros(
(1, 80, 88),
dtype=waveglow.upsample.weight.dtype,
device=waveglow.upsample.weight.device)
elif mode == 'normal':
mel_input = torch.randn(
(1, 80, 88),
dtype=waveglow.upsample.weight.dtype,
device=waveglow.upsample.weight.device)
else:
raise Exception("Mode {} if not supported".format(mode))
with torch.no_grad():
bias_audio = waveglow.infer(mel_input, sigma=0.0).float()
bias_spec, _ = self.stft.transform(bias_audio)
self.register_buffer('bias_spec', bias_spec[:, :, 0][:, :, None])
def forward(self, audio, strength=0.1):
audio_spec, audio_angles = self.stft.transform(audio.cuda().float())
audio_spec_denoised = audio_spec - self.bias_spec * strength
audio_spec_denoised = torch.clamp(audio_spec_denoised, 0.0)
audio_denoised = self.stft.inverse(audio_spec_denoised, audio_angles)
return audio_denoised
|