def patch_step(self, x, x_tilde, is_dis=True):
D_real, real_logits = self.PatchDiscriminator(x, classify=True)
D_fake, fake_logits = self.PatchDiscriminator(x_tilde, classify=True)
if is_dis:
w_dis = torch.mean(D_real - D_fake)
gp = calculate_gradients_penalty(self.PatchDiscriminator, x, x_tilde)
return w_dis, real_logits, gp
else:
return -torch.mean(D_fake), fake_logits
def patch_discriminate_step(self, x, x_tilde, cal_gp=True):
# w-distance
D_real, real_logits = self.PatchDiscriminator(x, classify=True)
D_fake, fake_logits = self.PatchDiscriminator(x_tilde, classify=True)
w_dis = torch.mean(D_real - D_fake)
if cal_gp:
gp = calculate_gradients_penalty(self.PatchDiscriminator, x,
x_tilde)
return w_dis, real_logits, fake_logits, gp
else:
return w_dis, real_logits, fake_logits
def latent_discriminate_step(self, enc_i_t, enc_i_tk, enc_i_prime, enc_j, is_dis=True):
same_pair = torch.cat([enc_i_t, enc_i_tk], dim=1)
diff_pair = torch.cat([enc_i_prime, enc_j], dim=1)
if is_dis:
same_val = self.LatentDiscriminator(same_pair)
diff_val = self.LatentDiscriminator(diff_pair)
w_dis = torch.mean(same_val - diff_val)
gp = calculate_gradients_penalty(self.LatentDiscriminator, same_pair, diff_pair)
return w_dis, gp
else:
diff_val = self.LatentDiscriminator(diff_pair)
loss_adv = -torch.mean(diff_val)
return loss_adv
w_dis = torch.mean(same_val - diff_val)
gp = calculate_gradients_penalty(self.LatentDiscriminator, same_pair, diff_pair)
return w_dis, gp
else:
diff_val = self.LatentDiscriminator(diff_pair)
loss_adv = -torch.mean(diff_val)
return loss_adv
def patch_discriminate_step(self, x, x_tilde, cal_gp=True):
# w-distance
>>>>>>> 4b2f701ba47b7c326213842f90ded5a8f429ec15
D_real, real_logits = self.PatchDiscriminator(x, classify=True)
D_fake, fake_logits = self.PatchDiscriminator(x_tilde, classify=True)
if is_dis:
w_dis = torch.mean(D_real - D_fake)
gp = calculate_gradients_penalty(self.PatchDiscriminator, x, x_tilde)
return w_dis, real_logits, gp
else:
return -torch.mean(D_fake), fake_logits
def gen_step(self, enc, c):
x_gen = self.Decoder(enc, c) + self.Generator(enc, c)
return x_gen
def clf_step(self, enc):
logits = self.SpeakerClassifier(enc)
return logits
def cal_loss(self, logits, y_true):
# calculate loss
criterion = nn.CrossEntropyLoss()