StarMirror/MockingBird
1
0
Fork 0
mirror of https://github.com/babysor/MockingBird.git synced 2024-03-22 13:11:31 +08:00
Code Issues Projects Releases Wiki Activity

fixed the issues #372 (#379)

Browse Source 
修复了一些参数传递造成的问题，把过时的torch.nn.functional.tanh()改成了torch.tanh()
This commit is contained in:
AyahaShirane 2022-02-27 11:02:01 +08:00 committed by GitHub
parent 9e072c2619
commit ad22997614
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
5 changed files with 7 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
synthesizer/audio.py
View File

@ -167,7 +167,7 @@ def _mel_to_linear(mel_spectrogram, hparams):
def _build_mel_basis(hparams):
assert hparams.fmax <= hparams.sample_rate // 2
return librosa.filters.mel(hparams.sample_rate, hparams.n_fft, n_mels=hparams.num_mels,
return librosa.filters.mel(sr=hparams.sample_rate, n_fft=hparams.n_fft, n_mels=hparams.num_mels,
fmin=hparams.fmin, fmax=hparams.fmax)
def _amp_to_db(x, hparams):

2
synthesizer/inference.py
View File

@ -149,7 +149,7 @@ class Synthesizer:
Loads and preprocesses an audio file under the same conditions the audio files were used to
train the synthesizer.
"""
wav = librosa.load(str(fpath), hparams.sample_rate)[0]
wav = librosa.load(path=str(fpath), sr=hparams.sample_rate)[0]
if hparams.rescale:
wav = wav / np.abs(wav).max() * hparams.rescaling_max
# denoise

2
synthesizer/models/global_style_token.py
View File

@ -97,7 +97,7 @@ class STL(nn.Module):
def forward(self, inputs):
N = inputs.size(0)
query = inputs.unsqueeze(1) # [N, 1, E//2]
keys = tFunctional.tanh(self.embed).unsqueeze(0).expand(N, -1, -1) # [N, token_num, E // num_heads]
keys = torch.tanh(self.embed).unsqueeze(0).expand(N, -1, -1) # [N, token_num, E // num_heads]
style_embed = self.attention(query, keys)
return style_embed

2
synthesizer/preprocess_speaker.py
View File

@ -63,7 +63,7 @@ def _process_utterance(wav: np.ndarray, text: str, out_dir: Path, basename: str,
def _split_on_silences(wav_fpath, words, hparams):
# Load the audio waveform
wav, _ = librosa.load(wav_fpath, hparams.sample_rate)
wav, _ = librosa.load(wav_fpath, sr= hparams.sample_rate)
wav = librosa.effects.trim(wav, top_db= 40, frame_length=2048, hop_length=512)[0]
if hparams.rescale:
wav = wav / np.abs(wav).max() * hparams.rescaling_max

6
vocoder/wavernn/models/deepmind_version.py
View File

@ -59,7 +59,7 @@ class WaveRNN(nn.Module) :
# Compute all gates for coarse and fine
u = F.sigmoid(R_u + I_u + self.bias_u)
r = F.sigmoid(R_r + I_r + self.bias_r)
e = F.tanh(r * R_e + I_e + self.bias_e)
e = torch.tanh(r * R_e + I_e + self.bias_e)
hidden = u * prev_hidden + (1. - u) * e
# Split the hidden state
@ -118,7 +118,7 @@ class WaveRNN(nn.Module) :
# Compute the coarse gates
u = F.sigmoid(R_coarse_u + I_coarse_u + b_coarse_u)
r = F.sigmoid(R_coarse_r + I_coarse_r + b_coarse_r)
e = F.tanh(r * R_coarse_e + I_coarse_e + b_coarse_e)
e = torch.tanh(r * R_coarse_e + I_coarse_e + b_coarse_e)
hidden_coarse = u * hidden_coarse + (1. - u) * e
# Compute the coarse output
@ -138,7 +138,7 @@ class WaveRNN(nn.Module) :
# Compute the fine gates
u = F.sigmoid(R_fine_u + I_fine_u + b_fine_u)
r = F.sigmoid(R_fine_r + I_fine_r + b_fine_r)
e = F.tanh(r * R_fine_e + I_fine_e + b_fine_e)
e = torch.tanh(r * R_fine_e + I_fine_e + b_fine_e)
hidden_fine = u * hidden_fine + (1. - u) * e
# Compute the fine output