def __init__(self, hparams):
super(Decoder, self).__init__()
self.n_acoustic_feat_dims = hparams.n_acoustic_feat_dims
self.encoder_embedding_dim = hparams.encoder_embedding_dim
self.attention_rnn_dim = hparams.attention_rnn_dim
self.decoder_rnn_dim = hparams.decoder_rnn_dim
self.prenet_dim = hparams.prenet_dim
self.max_decoder_steps = hparams.max_decoder_steps
self.gate_threshold = hparams.gate_threshold
self.p_attention_dropout = hparams.p_attention_dropout
self.p_decoder_dropout = hparams.p_decoder_dropout
self.attention_window_size = hparams.attention_window_size
self.prenet = Prenet(hparams.n_acoustic_feat_dims,
[hparams.prenet_dim, hparams.prenet_dim])
self.attention_rnn = nn.LSTMCell(
hparams.prenet_dim + hparams.encoder_embedding_dim,
hparams.attention_rnn_dim)
self.attention_layer = Attention(
hparams.attention_rnn_dim, hparams.encoder_embedding_dim,
hparams.attention_dim, hparams.attention_location_n_filters,
hparams.attention_location_kernel_size)
self.decoder_rnn = nn.LSTMCell(
hparams.attention_rnn_dim + hparams.encoder_embedding_dim,
hparams.decoder_rnn_dim, 1)
self.linear_projection = LinearNorm(
hparams.decoder_rnn_dim + hparams.encoder_embedding_dim,
hparams.n_acoustic_feat_dims)
self.gate_layer = LinearNorm(
hparams.decoder_rnn_dim + hparams.encoder_embedding_dim, 1,
bias=True, w_init_gain='sigmoid')
def __init__(self, vocab_size, decoder_dim, hidden_size, dropout=0.5):
super(MIEsitmator, self).__init__()
self.proj = nn.Sequential(
LinearNorm(decoder_dim, hidden_size, bias=True,
w_init_gain='relu'), nn.ReLU(), nn.Dropout(p=dropout))
self.ctc_proj = LinearNorm(hidden_size, vocab_size + 1, bias=True)
self.ctc = nn.CTCLoss(blank=vocab_size, reduction='none')
def __init__(self, n_mel_channels, n_frames_per_step,
encoder_embedding_dim, attention_dim,
attention_location_n_filters, attention_location_kernel_size,
prenet_dim, decoder_rnn_dim, max_decoder_steps,
gate_threshold, decoder_n_lstms, p_decoder_dropout):
super(Decoder, self).__init__()
self.n_mel_channels = n_mel_channels
self.n_frames_per_step = n_frames_per_step
self.encoder_embedding_dim = encoder_embedding_dim
self.decoder_rnn_dim = decoder_rnn_dim
self.prenet_dim = prenet_dim
self.max_decoder_steps = max_decoder_steps
self.gate_threshold = gate_threshold
self.decoder_n_lstms = decoder_n_lstms
self.p_decoder_dropout = p_decoder_dropout
self.prenet = Prenet(n_mel_channels, [prenet_dim, prenet_dim])
self.lstm0 = nn.LSTMCell(prenet_dim + encoder_embedding_dim,
decoder_rnn_dim)
self.lstm1 = nn.LSTMCell(decoder_rnn_dim + encoder_embedding_dim,
decoder_rnn_dim)
self.attention_layer = Attention(decoder_rnn_dim,
encoder_embedding_dim, attention_dim,
attention_location_n_filters,
attention_location_kernel_size)
self.linear_projection = LinearNorm(
decoder_rnn_dim + encoder_embedding_dim,
n_mel_channels * n_frames_per_step)
self.gate_layer = LinearNorm(decoder_rnn_dim + encoder_embedding_dim,
n_frames_per_step,
w_init_gain='sigmoid')
def __init__(self, query_dim, memory_dim, attention_dim,
attention_location_n_filters, attention_location_kernel_size):
super(Attention, self).__init__()
self.query_layer = LinearNorm(query_dim, attention_dim, w_init_gain='tanh')
self.memory_layer = LinearNorm(memory_dim, attention_dim, w_init_gain='tanh')
self.v = LinearNorm(attention_dim, 1)
self.location_layer = LocationLayer(attention_location_n_filters,
attention_location_kernel_size,
attention_dim)
self.score_mask_value = -float("inf")
def __init__(self, attention_rnn_dim, embedding_dim, attention_dim,
attention_location_n_filters, attention_location_kernel_size):
super(Attention, self).__init__()
self.query_layer = LinearNorm(attention_rnn_dim, attention_dim,
bias=False, w_init_gain='tanh')
self.memory_layer = LinearNorm(embedding_dim, attention_dim, bias=False,
w_init_gain='tanh')
self.v = LinearNorm(attention_dim, 1, bias=False)
self.location_layer = LocationLayer(attention_location_n_filters,
attention_location_kernel_size,
attention_dim)
self.score_mask_value = -float("inf")
self.not_so_small_mask = -1000
def __init__(self, n_mel_channels, n_frames_per_step,
encoder_embedding_dim, attention_dim,
attention_location_n_filters, attention_location_kernel_size,
attention_rnn_dim, decoder_rnn_dim, prenet_dim,
max_decoder_steps, gate_threshold, p_attention_dropout,
p_decoder_dropout, early_stopping):
super(Decoder, self).__init__()
self.n_mel_channels = n_mel_channels
self.n_frames_per_step = n_frames_per_step
self.encoder_embedding_dim = encoder_embedding_dim
self.attention_rnn_dim = attention_rnn_dim
self.decoder_rnn_dim = decoder_rnn_dim
self.prenet_dim = prenet_dim
self.max_decoder_steps = max_decoder_steps
self.gate_threshold = gate_threshold
self.p_attention_dropout = p_attention_dropout
self.p_decoder_dropout = p_decoder_dropout
self.prenet = Prenet(n_mel_channels * n_frames_per_step,
[prenet_dim, prenet_dim])
self.attention_rnn = nn.LSTMCell(prenet_dim + encoder_embedding_dim,
attention_rnn_dim)
self.attention_layer = Attention(attention_rnn_dim,
encoder_embedding_dim, attention_dim,
attention_location_n_filters,
attention_location_kernel_size)
self.decoder_rnn = nn.LSTMCell(
attention_rnn_dim + encoder_embedding_dim, decoder_rnn_dim, 1)
self.linear_projection = nn.Sequential(
LinearNorm(decoder_rnn_dim + encoder_embedding_dim,
decoder_rnn_dim,
bias=True,
w_init_gain='relu'), nn.ReLU(), nn.Dropout(p=0.5))
self.mel_layer = nn.Sequential(
LinearNorm(decoder_rnn_dim,
decoder_rnn_dim,
bias=True,
w_init_gain='relu'), nn.ReLU(), nn.Dropout(p=0.5),
LinearNorm(decoder_rnn_dim, n_mel_channels * n_frames_per_step))
self.gate_layer = LinearNorm(decoder_rnn_dim,
1,
bias=True,
w_init_gain='sigmoid')
def __init__(self, in_dim, sizes):
super(Prenet, self).__init__()
in_sizes = [in_dim] + sizes[:-1]
self.layers = nn.ModuleList([
LinearNorm(in_size, out_size, bias=False)
for (in_size, out_size) in zip(in_sizes, sizes)
])
def __init__(self, attention_n_filters, attention_kernel_size, attention_dim):
super(LocationLayer, self).__init__()
self.location_conv = ConvNorm(1, attention_n_filters,
kernel_size=attention_kernel_size,
padding=int((attention_kernel_size - 1) / 2),
stride=1, dilation=1)
self.location_dense = LinearNorm(attention_n_filters, attention_dim,
bias=False, w_init_gain='tanh')
def __init__(self, n_mel_channels, n_frames_per_step,
encoder_embedding_dim, attention_dim,
attention_location_n_filters, attention_location_kernel_size,
attention_rnn_dim, decoder_rnn_dim, prenet_dim,
max_decoder_steps, gate_threshold, p_attention_dropout,
p_decoder_dropout, early_stopping, E, speaker_embedding_dim):
super(Decoder, self).__init__()
self.n_mel_channels = n_mel_channels
self.n_frames_per_step = n_frames_per_step
self.encoder_embedding_dim = encoder_embedding_dim + E # + speaker_embedding_dim # speaker_id 쓸때만 이거 값 더해주기!
self.attention_rnn_dim = attention_rnn_dim
self.decoder_rnn_dim = decoder_rnn_dim
self.prenet_dim = prenet_dim
self.max_decoder_steps = max_decoder_steps
self.gate_threshold = gate_threshold
self.p_attention_dropout = p_attention_dropout
self.p_decoder_dropout = p_decoder_dropout
self.early_stopping = early_stopping
self.prenet = Prenet(n_mel_channels * n_frames_per_step,
[prenet_dim, prenet_dim])
self.attention_rnn = nn.LSTMCell(
prenet_dim + self.encoder_embedding_dim, attention_rnn_dim)
self.attention_layer = Attention(attention_rnn_dim,
self.encoder_embedding_dim,
attention_dim,
attention_location_n_filters,
attention_location_kernel_size)
self.decoder_rnn = nn.LSTMCell(
attention_rnn_dim + self.encoder_embedding_dim, decoder_rnn_dim, 1)
self.linear_projection = LinearNorm(
decoder_rnn_dim + self.encoder_embedding_dim,
n_mel_channels * n_frames_per_step)
self.gate_layer = LinearNorm(decoder_rnn_dim +
self.encoder_embedding_dim,
1,
bias=True,
w_init_gain='sigmoid')
def __init__(self, in_dim, sizes):
"""
Args:
in_dim ([type]): n_mel_channels * n_frames_per_step
sizes ([type]): [prenet_dim, prenet_dim]
"""
super(Prenet, self).__init__()
in_sizes = [in_dim] + sizes[:-1]
self.layers = nn.ModuleList([
LinearNorm(in_size, out_size, bias=False)
for (in_size, out_size) in zip(in_sizes, sizes)
])
