Fork of https://github.com/alokprasad/fastspeech_squeezewave to also fix denoising in squeezewave
You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
39 lines
1.5 KiB
39 lines
1.5 KiB
import sys
|
|
import torch
|
|
from stft import STFT
|
|
|
|
|
|
class Denoiser(torch.nn.Module):
|
|
""" Removes model bias from audio produced with squeezewave"""
|
|
|
|
def __init__(self, squeezewave, 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)
|
|
if mode == 'zeros':
|
|
mel_input = torch.zeros(
|
|
(1, 80, 88),
|
|
dtype=squeezewave.upsample.weight.dtype,
|
|
device=squeezewave.upsample.weight.device)
|
|
elif mode == 'normal':
|
|
mel_input = torch.randn(
|
|
(1, 80, 88),
|
|
dtype=squeezewave.upsample.weight.dtype,
|
|
device=squeezewave.upsample.weight.device)
|
|
else:
|
|
raise Exception("Mode {} if not supported".format(mode))
|
|
|
|
with torch.no_grad():
|
|
bias_audio = squeezewave.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.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
|