mirror of https://github.com/malarinv/tacotron2
model.py: attending to full mel instead of prenet and dropout mel
parent
da30fd8709
commit
977cb37cea
11
model.py
11
model.py
|
|
@ -221,7 +221,7 @@ class Decoder(nn.Module):
|
|||
[hparams.prenet_dim, hparams.prenet_dim])
|
||||
|
||||
self.attention_rnn = nn.LSTMCell(
|
||||
hparams.prenet_dim + hparams.encoder_embedding_dim,
|
||||
hparams.decoder_rnn_dim + hparams.encoder_embedding_dim,
|
||||
hparams.attention_rnn_dim)
|
||||
|
||||
self.attention_layer = Attention(
|
||||
|
|
@ -230,7 +230,7 @@ class Decoder(nn.Module):
|
|||
hparams.attention_location_kernel_size)
|
||||
|
||||
self.decoder_rnn = nn.LSTMCell(
|
||||
hparams.attention_rnn_dim + hparams.encoder_embedding_dim,
|
||||
hparams.prenet_dim + hparams.encoder_embedding_dim,
|
||||
hparams.decoder_rnn_dim, 1)
|
||||
|
||||
self.linear_projection = LinearNorm(
|
||||
|
|
@ -351,8 +351,8 @@ class Decoder(nn.Module):
|
|||
attention_weights:
|
||||
"""
|
||||
|
||||
decoder_input = self.prenet(decoder_input)
|
||||
cell_input = torch.cat((decoder_input, self.attention_context), -1)
|
||||
prenet_output = self.prenet(decoder_input)
|
||||
cell_input = torch.cat((self.decoder_hidden, self.attention_context), -1)
|
||||
self.attention_hidden, self.attention_cell = self.attention_rnn(
|
||||
cell_input, (self.attention_hidden, self.attention_cell))
|
||||
|
||||
|
|
@ -364,8 +364,7 @@ class Decoder(nn.Module):
|
|||
attention_weights_cat, self.mask)
|
||||
|
||||
self.attention_weights_cum += self.attention_weights
|
||||
decoder_input = torch.cat(
|
||||
(self.attention_hidden, self.attention_context), -1)
|
||||
decoder_input = torch.cat((prenet_output, self.attention_context), -1)
|
||||
self.decoder_hidden, self.decoder_cell = self.decoder_rnn(
|
||||
decoder_input, (self.decoder_hidden, self.decoder_cell))
|
||||
|
||||
|
|
|
|||
Loading…
Reference in New Issue