Merge branch 'normal-attention+masked-loss'

config.json

@@ -7,25 +7,24 @@
   "preemphasis": 0.97,
   "min_level_db": -100,
   "ref_level_db": 20,
-  "hidden_size": 128,
   "embedding_size": 256,
   "text_cleaner": "english_cleaners",
 
   "epochs": 2000,
   "lr": 0.001,
   "warmup_steps": 4000,
-  "batch_size": 32,
+  "batch_size": 128,
-  "eval_batch_size": 32,
+  "eval_batch_size":32,
   "r": 5,
 
   "griffin_lim_iters": 60,
   "power": 1.5,
 
-  "num_loader_workers": 12,
+  "num_loader_workers": 8,
 
-  "checkpoint": false,
+  "checkpoint": true,
-  "save_step": 69,
+  "save_step": 378,
   "data_path": "/run/shm/erogol/LJSpeech-1.0",
   "min_seq_len": 0, 
-  "output_path": "result"
+  "output_path": "/data/shared/erogol_models/"
 }

datasets/LJSpeech.py

@@ -7,7 +7,8 @@ from torch.utils.data import Dataset
 
 from TTS.utils.text import text_to_sequence
 from TTS.utils.audio import AudioProcessor
-from TTS.utils.data import prepare_data, pad_data, pad_per_step
+from TTS.utils.data import (prepare_data, pad_per_step,
+                            prepare_tensor, prepare_stop_target)
 
 
 class LJSpeechDataset(Dataset):
@@ -93,26 +94,27 @@ class LJSpeechDataset(Dataset):
             text_lenghts = np.array([len(x) for x in text])
             max_text_len = np.max(text_lenghts)
 
+            linear = [self.ap.spectrogram(w).astype('float32') for w in wav]
+            mel = [self.ap.melspectrogram(w).astype('float32') for w in wav]
+            mel_lengths = [m.shape[1] + 1 for m in mel]  # +1 for zero-frame 
+            
+            # compute 'stop token' targets
+            stop_targets = [np.array([0.]*(mel_len-1)) for mel_len in mel_lengths]
+            
+            # PAD stop targets
+            stop_targets = prepare_stop_target(stop_targets, self.outputs_per_step)
+
             # PAD sequences with largest length of the batch
             text = prepare_data(text).astype(np.int32)
             wav = prepare_data(wav)
 
-            linear = np.array([self.ap.spectrogram(w).astype('float32') for w in wav])
+            # PAD features with largest length + a zero frame
-            mel = np.array([self.ap.melspectrogram(w).astype('float32') for w in wav])
+            linear = prepare_tensor(linear, self.outputs_per_step)
+            mel = prepare_tensor(mel, self.outputs_per_step)
             assert mel.shape[2] == linear.shape[2]
-            timesteps = mel.shape[2]
+            timesteps = mel.shape[2]            
 
-            # PAD with zeros that can be divided by outputs per step
+            # B x T x D
-            if (timesteps + 1) % self.outputs_per_step != 0:
-                pad_len = self.outputs_per_step - \
-                        ((timesteps + 1) % self.outputs_per_step)
-                pad_len += 1
-            else:
-                pad_len = 1
-            linear = pad_per_step(linear, pad_len)
-            mel = pad_per_step(mel, pad_len)
-
-            # reshape jombo
             linear = linear.transpose(0, 2, 1)
             mel = mel.transpose(0, 2, 1)
 
@@ -121,7 +123,10 @@ class LJSpeechDataset(Dataset):
             text = torch.LongTensor(text)
             linear = torch.FloatTensor(linear)
             mel = torch.FloatTensor(mel)
-            return text, text_lenghts, linear, mel, item_idxs[0]
+            mel_lengths = torch.LongTensor(mel_lengths)
+            stop_targets = torch.FloatTensor(stop_targets)
+            
+            return text, text_lenghts, linear, mel, mel_lengths, stop_targets, item_idxs[0]
 
         raise TypeError(("batch must contain tensors, numbers, dicts or lists;\
                          found {}"

layers/attention.py

@@ -48,7 +48,7 @@ class AttentionRNN(nn.Module):
     def __init__(self, out_dim, annot_dim, memory_dim,
                  score_mask_value=-float("inf")):
         super(AttentionRNN, self).__init__()
-        self.rnn_cell = nn.GRUCell(annot_dim + memory_dim, out_dim)
+        self.rnn_cell = nn.GRUCell(out_dim + memory_dim, out_dim)
         self.alignment_model = BahdanauAttention(annot_dim, out_dim, out_dim)
         self.score_mask_value = score_mask_value
 
@@ -57,11 +57,19 @@ class AttentionRNN(nn.Module):
 
         if annotations_lengths is not None and mask is None:
             mask = get_mask_from_lengths(annotations, annotations_lengths)
+            
+        # Concat input query and previous context context
+        rnn_input = torch.cat((memory, context), -1)
+        #rnn_input = rnn_input.unsqueeze(1)
+        
+        # Feed it to RNN
+        # s_i = f(y_{i-1}, c_{i}, s_{i-1})
+        rnn_output = self.rnn_cell(rnn_input, rnn_state)
 
         # Alignment
         # (batch, max_time)
         # e_{ij} = a(s_{i-1}, h_j)
-        alignment = self.alignment_model(annotations, rnn_state)
+        alignment = self.alignment_model(annotations, rnn_output)
 
         # TODO: needs recheck.
         if mask is not None:
@@ -75,16 +83,6 @@ class AttentionRNN(nn.Module):
         # (batch, 1, dim)
         # c_i = \sum_{j=1}^{T_x} \alpha_{ij} h_j
         context = torch.bmm(alignment.unsqueeze(1), annotations)
-        context = context.squeeze(1)
-
-        # Concat input query and previous context context
-        rnn_input = torch.cat((memory, context), -1)
-        #rnn_input = rnn_input.unsqueeze(1)
-
-        # Feed it to RNN
-        # s_i = f(y_{i-1}, c_{i}, s_{i-1})
-        rnn_output = self.rnn_cell(rnn_input, rnn_state)
-
         context = context.squeeze(1)
         return rnn_output, context, alignment
 

layers/custom_layers.py

@@ -0,0 +1,26 @@
+# coding: utf-8
+import torch
+from torch.autograd import Variable
+from torch import nn
+
+
+# class StopProjection(nn.Module):
+#     r""" Simple projection layer to predict the "stop token"
+
+#     Args:
+#         in_features (int): size of the input vector
+#         out_features (int or list): size of each output vector. aka number
+#             of predicted frames.
+#     """  
+
+#     def __init__(self, in_features, out_features):
+#         super(StopProjection, self).__init__()
+#         self.linear = nn.Linear(in_features, out_features)
+#         self.dropout = nn.Dropout(0.5)
+#         self.sigmoid = nn.Sigmoid()
+    
+#     def forward(self, inputs):
+#         out = self.dropout(inputs)
+#         out = self.linear(out)
+#         out = self.sigmoid(out)
+#         return out

layers/losses.py

@@ -0,0 +1,57 @@
+import torch 
+from torch.nn import functional
+from torch.autograd import Variable
+from torch import nn
+
+
+# from https://gist.github.com/jihunchoi/f1434a77df9db1bb337417854b398df1
+def _sequence_mask(sequence_length, max_len=None):
+    if max_len is None:
+        max_len = sequence_length.data.max()
+    batch_size = sequence_length.size(0)
+    seq_range = torch.arange(0, max_len).long()
+    seq_range_expand = seq_range.unsqueeze(0).expand(batch_size, max_len)
+    seq_range_expand = Variable(seq_range_expand)
+    if sequence_length.is_cuda:
+        seq_range_expand = seq_range_expand.cuda()
+    seq_length_expand = (sequence_length.unsqueeze(1)
+                         .expand_as(seq_range_expand))
+    return seq_range_expand < seq_length_expand
+
+
+class L1LossMasked(nn.Module):
+    
+    def __init__(self):
+        super(L1LossMasked, self).__init__()
+    
+    def forward(self, input, target, length):
+        """
+        Args:
+            logits: A Variable containing a FloatTensor of size
+                (batch, max_len, num_classes) which contains the
+                unnormalized probability for each class.
+            target: A Variable containing a LongTensor of size
+                (batch, max_len) which contains the index of the true
+                class for each corresponding step.
+            length: A Variable containing a LongTensor of size (batch,)
+                which contains the length of each data in a batch.
+        Returns:
+            loss: An average loss value masked by the length.
+        """
+        input = input.contiguous()
+        target = target.contiguous()
+
+        # logits_flat: (batch * max_len, dim)
+        input = input.view(-1, input.size(-1))
+        # target_flat: (batch * max_len, dim)
+        target_flat = target.view(-1, 1)
+        # losses_flat: (batch * max_len, dim)
+        losses_flat = functional.l1_loss(input, target, size_average=False,
+                             reduce=False)
+        # losses: (batch, max_len, dim)
+        losses = losses_flat.view(*target.size())
+        # mask: (batch, max_len, 1)
+        mask = _sequence_mask(sequence_length=length, max_len=target.size(1)).unsqueeze(2)
+        losses = losses * mask.float()
+        loss = losses.sum() / (length.float().sum() * float(target.shape[2]))
+        return loss

layers/tacotron.py

@@ -48,6 +48,7 @@ class BatchNormConv1d(nn.Module):
         - input: batch x dims
         - output: batch x dims
     """
+
     def __init__(self, in_channels, out_channels, kernel_size, stride, padding,
                  activation=None):
         super(BatchNormConv1d, self).__init__()
@@ -213,8 +214,9 @@ class Decoder(nn.Module):
         r (int): number of outputs per time step.
         eps (float): threshold for detecting the end of a sentence.
     """
-    def __init__(self, in_features, memory_dim, r, eps=0.05):
+    def __init__(self, in_features, memory_dim, r, eps=0.05, mode='train'):
         super(Decoder, self).__init__()
+        self.mode = mode
         self.max_decoder_steps = 200
         self.memory_dim = memory_dim
         self.eps = eps
@@ -241,7 +243,8 @@ class Decoder(nn.Module):
         Args:
             inputs: Encoder outputs.
             memory (None): Decoder memory (autoregression. If None (at eval-time),
-              decoder outputs are used as decoder inputs.
+              decoder outputs are used as decoder inputs. If None, it uses the last
+              output as the input.
 
         Shapes:
             - inputs: batch x time x encoder_out_dim
@@ -250,14 +253,13 @@ class Decoder(nn.Module):
         B = inputs.size(0)
 
         # Run greedy decoding if memory is None
-        greedy = memory is None
+        greedy = not self.training
 
         if memory is not None:
-
+            
             # Grouping multiple frames if necessary
             if memory.size(-1) == self.memory_dim:
                 memory = memory.view(B, memory.size(1) // self.r, -1)
-            assert memory.size(-1) == self.memory_dim * self.r,\
                 " !! Dimension mismatch {} vs {} * {}".format(memory.size(-1),
                                                          self.memory_dim, self.r)
             T_decoder = memory.size(1)
@@ -286,15 +288,23 @@ class Decoder(nn.Module):
         memory_input = initial_memory
         while True:
             if t > 0:
-                memory_input = outputs[-1] if greedy else memory[t - 1]
+                if greedy:
+                    memory_input = outputs[-1]
+                else:
+                    # combine prev. model output and prev. real target
+                    # memory_input = torch.div(outputs[-1] + memory[t-1], 2.0)
+                    # add a random noise
+                    # noise = torch.autograd.Variable(
+                        # memory_input.data.new(memory_input.size()).normal_(0.0, 0.5))
+                    # memory_input = memory_input + noise
+                    memory_input = memory[t-1]
 
             # Prenet
             processed_memory = self.prenet(memory_input)
 
             # Attention RNN
             attention_rnn_hidden, current_context_vec, alignment = self.attention_rnn(
-                processed_memory, current_context_vec, attention_rnn_hidden,
+                processed_memory, current_context_vec, attention_rnn_hidden, inputs)
-                inputs)
 
             # Concat RNN output and attention context vector
             decoder_input = self.project_to_decoder_in(
@@ -306,8 +316,9 @@ class Decoder(nn.Module):
                     decoder_input, decoder_rnn_hiddens[idx])
                 # Residual connectinon
                 decoder_input = decoder_rnn_hiddens[idx] + decoder_input
-
+            
             output = decoder_input
+            
 
             # predict mel vectors from decoder vectors
             output = self.proj_to_mel(output)
@@ -317,17 +328,17 @@ class Decoder(nn.Module):
 
             t += 1
 
-            if greedy:
+            if (not greedy and self.training) or (greedy and memory is not None):
+                if t >= T_decoder:
+                    break
+            else:
                 if t > 1 and is_end_of_frames(output, self.eps):
                     break
                 elif t > self.max_decoder_steps:
                     print(" !! Decoder stopped with 'max_decoder_steps'. \
                           Something is probably wrong.")
                     break
-            else:
+                           
-                if t >= T_decoder:
-                    break
-
         assert greedy or len(outputs) == T_decoder
 
         # Back to batch first
@@ -338,4 +349,4 @@ class Decoder(nn.Module):
 
 
 def is_end_of_frames(output, eps=0.2): #0.2
-    return (output.data <= eps).all()
+    return (output.data <= eps).all()

models/tacotron.py

@@ -8,9 +8,10 @@ from TTS.layers.tacotron import Prenet, Encoder, Decoder, CBHG
 
 class Tacotron(nn.Module):
     def __init__(self, embedding_dim=256, linear_dim=1025, mel_dim=80,
-                 freq_dim=1025, r=5, padding_idx=None):
+                 r=5, padding_idx=None):
                  
         super(Tacotron, self).__init__()
+        self.r = r
         self.mel_dim = mel_dim
         self.linear_dim = linear_dim
         self.embedding = nn.Embedding(len(symbols), embedding_dim,
@@ -23,9 +24,10 @@ class Tacotron(nn.Module):
         self.decoder = Decoder(256, mel_dim, r)
 
         self.postnet = CBHG(mel_dim, K=8, projections=[256, mel_dim])
-        self.last_linear = nn.Linear(mel_dim * 2, freq_dim)
+        self.last_linear = nn.Linear(mel_dim * 2, linear_dim)
 
     def forward(self, characters, mel_specs=None):
+        
         B = characters.size(0)
 
         inputs = self.embedding(characters)

tests/layers_tests.py

@@ -2,6 +2,7 @@ import unittest
 import torch as T
 
 from TTS.layers.tacotron import Prenet, CBHG, Decoder, Encoder
+from layers.losses import L1LossMasked, _sequence_mask
 
 
 class PrenetTests(unittest.TestCase):
@@ -32,23 +33,22 @@ class CBHGTests(unittest.TestCase):
 class DecoderTests(unittest.TestCase):
 
     def test_in_out(self):
-        layer = Decoder(in_features=128, memory_dim=32, r=5)
+        layer = Decoder(in_features=256, memory_dim=80, r=2)
-        dummy_input = T.autograd.Variable(T.rand(4, 8, 128))
+        dummy_input = T.autograd.Variable(T.rand(4, 8, 256))
-        dummy_memory = T.autograd.Variable(T.rand(4, 120, 32))
+        dummy_memory = T.autograd.Variable(T.rand(4, 2, 80))
 
-        print(layer)
         output, alignment = layer(dummy_input, dummy_memory)
-        print(output.shape)
+        
         assert output.shape[0] == 4
-        assert output.shape[1] == 120 / 5
+        assert output.shape[1] == 1, "size not {}".format(output.shape[1])
-        assert output.shape[2] == 32 * 5
+        assert output.shape[2] == 80 * 2, "size not {}".format(output.shape[2])
-
+        
 
 class EncoderTests(unittest.TestCase):
 
     def test_in_out(self):
         layer = Encoder(128)
-       dummy_input = T.autograd.Variable(T.rand(4, 8, 128))
+        dummy_input = T.autograd.Variable(T.rand(4, 8, 128))
 
         print(layer)
         output = layer(dummy_input)
@@ -56,4 +56,29 @@ class EncoderTests(unittest.TestCase):
         assert output.shape[0] == 4
         assert output.shape[1] == 8
         assert output.shape[2] == 256  # 128 * 2 BiRNN
+        
 
+class L1LossMaskedTests(unittest.TestCase):
+    
+    def test_in_out(self):
+        layer = L1LossMasked()
+        dummy_input = T.autograd.Variable(T.ones(4, 8, 128).float())
+        dummy_target = T.autograd.Variable(T.ones(4, 8, 128).float())
+        dummy_length = T.autograd.Variable((T.ones(4) * 8).long())
+        output = layer(dummy_input, dummy_target, dummy_length)
+        assert output.shape[0] == 1
+        assert len(output.shape) == 1
+        assert output.data[0] == 0.0
+        
+        dummy_input = T.autograd.Variable(T.ones(4, 8, 128).float())
+        dummy_target = T.autograd.Variable(T.zeros(4, 8, 128).float())
+        dummy_length = T.autograd.Variable((T.ones(4) * 8).long())
+        output = layer(dummy_input, dummy_target, dummy_length)
+        assert output.data[0] == 1.0, "1.0 vs {}".format(output.data[0])
+
+        dummy_input = T.autograd.Variable(T.ones(4, 8, 128).float())
+        dummy_target = T.autograd.Variable(T.zeros(4, 8, 128).float())
+        dummy_length = T.autograd.Variable((T.arange(5,9)).long())
+        mask = ((_sequence_mask(dummy_length).float() - 1.0) * 100.0).unsqueeze(2)
+        output = layer(dummy_input + mask, dummy_target, dummy_length)
+        assert output.data[0] == 1.0, "1.0 vs {}".format(output.data[0])

tests/loader_tests.py

@@ -32,7 +32,7 @@ class TestDataset(unittest.TestCase):
                                   c.power
                                 )
 
-        dataloader = DataLoader(dataset, batch_size=c.batch_size,
+        dataloader = DataLoader(dataset, batch_size=2,
                                 shuffle=True, collate_fn=dataset.collate_fn,
                                 drop_last=True, num_workers=c.num_loader_workers)
 
@@ -43,8 +43,10 @@ class TestDataset(unittest.TestCase):
             text_lengths = data[1]
             linear_input = data[2]
             mel_input = data[3]
-            item_idx = data[4]
+            mel_lengths = data[4]
-
+            stop_target = data[5]
+            item_idx = data[6]
+            
             neg_values = text_input[text_input < 0]
             check_count = len(neg_values)
             assert check_count == 0, \
@@ -70,8 +72,9 @@ class TestDataset(unittest.TestCase):
                                   c.power
                                 )
 
+        # Test for batch size 1
         dataloader = DataLoader(dataset, batch_size=1,
-                                shuffle=True, collate_fn=dataset.collate_fn,
+                                shuffle=False, collate_fn=dataset.collate_fn,
                                 drop_last=True, num_workers=c.num_loader_workers)
 
         for i, data in enumerate(dataloader):
@@ -81,13 +84,63 @@ class TestDataset(unittest.TestCase):
             text_lengths = data[1]
             linear_input = data[2]
             mel_input = data[3]
-            item_idx = data[4]
+            mel_lengths = data[4]
+            stop_target = data[5]
+            item_idx = data[6]
 
             # check the last time step to be zero padded
             assert mel_input[0, -1].sum() == 0
             assert mel_input[0, -2].sum() != 0
             assert linear_input[0, -1].sum() == 0
             assert linear_input[0, -2].sum() != 0
+            assert stop_target[0, -1] == 1
+            assert stop_target[0, -2] == 0
+            assert stop_target.sum() == 1
+            assert len(mel_lengths.shape) == 1
+            assert mel_lengths[0] == mel_input[0].shape[0]
+            
+        # Test for batch size 2    
+        dataloader = DataLoader(dataset, batch_size=2,
+                                shuffle=False, collate_fn=dataset.collate_fn,
+                                drop_last=False, num_workers=c.num_loader_workers)
+
+        for i, data in enumerate(dataloader):
+            if i == self.max_loader_iter:
+                break
+            text_input = data[0]
+            text_lengths = data[1]
+            linear_input = data[2]
+            mel_input = data[3]
+            mel_lengths = data[4]
+            stop_target = data[5]
+            item_idx = data[6]
+
+            if mel_lengths[0] >  mel_lengths[1]:
+                idx = 0
+            else:
+                idx = 1
+                
+            # check the first item in the batch
+            assert mel_input[idx, -1].sum() == 0
+            assert mel_input[idx, -2].sum() != 0, mel_input
+            assert linear_input[idx, -1].sum() == 0
+            assert linear_input[idx, -2].sum() != 0
+            assert stop_target[idx, -1] == 1
+            assert stop_target[idx, -2] == 0
+            assert stop_target[idx].sum() == 1
+            assert len(mel_lengths.shape) == 1
+            assert mel_lengths[idx] == mel_input[idx].shape[0]
+            
+            # check the second itme in the batch
+            assert mel_input[1-idx, -1].sum() == 0
+            assert linear_input[1-idx, -1].sum() == 0
+            assert stop_target[1-idx, -1] == 1
+            assert len(mel_lengths.shape) == 1
+            
+            # check batch conditions
+            assert (mel_input * stop_target.unsqueeze(2)).sum() == 0
+            assert (linear_input * stop_target.unsqueeze(2)).sum() == 0
+
 
 
 

train.py

@@ -26,6 +26,7 @@ from utils.model import get_param_size
 from utils.visual import plot_alignment, plot_spectrogram
 from datasets.LJSpeech import LJSpeechDataset
 from models.tacotron import Tacotron
+from layers.losses import L1LossMasked
 
 
 use_cuda = torch.cuda.is_available()
@@ -80,7 +81,8 @@ def train(model, criterion, data_loader, optimizer, epoch):
         text_lengths = data[1]
         linear_input = data[2]
         mel_input = data[3]
-
+        mel_lengths = data[4]
+        
         current_step = num_iter + args.restore_step + epoch * len(data_loader) + 1
 
         # setup lr
@@ -93,21 +95,14 @@ def train(model, criterion, data_loader, optimizer, epoch):
         # convert inputs to variables
         text_input_var = Variable(text_input)
         mel_spec_var = Variable(mel_input)
+        mel_lengths_var = Variable(mel_lengths)
         linear_spec_var = Variable(linear_input, volatile=True)
 
-        # sort sequence by length for curriculum learning
-        # TODO: might be unnecessary
-        sorted_lengths, indices = torch.sort(
-                 text_lengths.view(-1), dim=0, descending=True)
-        sorted_lengths = sorted_lengths.long().numpy()
-        text_input_var = text_input_var[indices]
-        mel_spec_var = mel_spec_var[indices]
-        linear_spec_var = linear_spec_var[indices]
-
         # dispatch data to GPU
         if use_cuda:
             text_input_var = text_input_var.cuda()
             mel_spec_var = mel_spec_var.cuda()
+            mel_lengths_var = mel_lengths_var.cuda()
             linear_spec_var = linear_spec_var.cuda()
 
         # forward pass
@@ -115,10 +110,11 @@ def train(model, criterion, data_loader, optimizer, epoch):
             model.forward(text_input_var, mel_spec_var)
         
         # loss computation
-        mel_loss = criterion(mel_output, mel_spec_var)
+        mel_loss = criterion(mel_output, mel_spec_var, mel_lengths_var)
-        linear_loss = 0.5 * criterion(linear_output, linear_spec_var) \
+        linear_loss = 0.5 * criterion(linear_output, linear_spec_var, mel_lengths_var) \
                 + 0.5 * criterion(linear_output[:, :, :n_priority_freq],
-                                  linear_spec_var[: ,: ,:n_priority_freq])
+                                  linear_spec_var[: ,: ,:n_priority_freq],
+                                  mel_lengths_var)
         loss = mel_loss + linear_loss
 
         # backpass and check the grad norm 
@@ -215,28 +211,31 @@ def evaluate(model, criterion, data_loader, current_step):
         text_lengths = data[1]
         linear_input = data[2]
         mel_input = data[3]
+        mel_lengths = data[4]
 
         # convert inputs to variables
         text_input_var = Variable(text_input)
         mel_spec_var = Variable(mel_input)
+        mel_lengths_var = Variable(mel_lengths)
         linear_spec_var = Variable(linear_input, volatile=True)
 
         # dispatch data to GPU
         if use_cuda:
             text_input_var = text_input_var.cuda()
             mel_spec_var = mel_spec_var.cuda()
+            mel_lengths_var = mel_lengths_var.cuda()
             linear_spec_var = linear_spec_var.cuda()
 
         # forward pass
-        mel_output, linear_output, alignments =\
+        mel_output, linear_output, alignments = model.forward(text_input_var, mel_spec_var)
-            model.forward(text_input_var, mel_spec_var)
         
         # loss computation
-        mel_loss = criterion(mel_output, mel_spec_var)
+        mel_loss = criterion(mel_output, mel_spec_var, mel_lengths_var)
-        linear_loss = 0.5 * criterion(linear_output, linear_spec_var) \
+        linear_loss = 0.5 * criterion(linear_output, linear_spec_var, mel_lengths_var) \
                 + 0.5 * criterion(linear_output[:, :, :n_priority_freq],
-                                  linear_spec_var[: ,: ,:n_priority_freq])
+                                  linear_spec_var[: ,: ,:n_priority_freq],
-        loss = mel_loss + linear_loss
+                                  mel_lengths_var)
+        loss = mel_loss + linear_loss 
 
         step_time = time.time() - start_time
         epoch_time += step_time
@@ -333,17 +332,16 @@ def main(args):
                             pin_memory=True)
 
     model = Tacotron(c.embedding_size,
-                     c.hidden_size,
-                     c.num_mels,
                      c.num_freq,
+                     c.num_mels,
                      c.r)
-                     
+
     optimizer = optim.Adam(model.parameters(), lr=c.lr)
     
     if use_cuda:
-        criterion = nn.L1Loss().cuda()
+        criterion = L1LossMasked().cuda()
     else:
-        criterion = nn.L1Loss()
+        criterion = L1LossMasked()   
 
     if args.restore_path:
         checkpoint = torch.load(args.restore_path)

utils/data.py

@@ -1,7 +1,7 @@
 import numpy as np
 
 
-def pad_data(x, length):
+def _pad_data(x, length):
     _pad = 0
     assert x.ndim == 1
     return np.pad(x, (0, length - x.shape[0]),
@@ -11,7 +11,33 @@ def pad_data(x, length):
 
 def prepare_data(inputs):
     max_len = max((len(x) for x in inputs))
-    return np.stack([pad_data(x, max_len) for x in inputs])
+    return np.stack([_pad_data(x, max_len) for x in inputs])
+
+
+def _pad_tensor(x, length):
+    _pad = 0
+    assert x.ndim == 2
+    x = np.pad(x, [[0, 0], [0, length - x.shape[1]]], mode='constant', constant_values=_pad)
+    return x
+
+def prepare_tensor(inputs, out_steps):
+    max_len = max((x.shape[1] for x in inputs)) + 1 # zero-frame
+    remainder = max_len % out_steps
+    pad_len = max_len + (out_steps - remainder) if remainder > 0 else max_len
+    return np.stack([_pad_tensor(x, pad_len) for x in inputs])
+
+
+def _pad_stop_target(x, length):
+    _pad = 1.
+    assert x.ndim == 1
+    return np.pad(x, (0, length - x.shape[0]), mode='constant', constant_values=_pad)
+
+
+def prepare_stop_target(inputs, out_steps):
+    max_len = max((x.shape[0] for x in inputs)) + 1  # zero-frame
+    remainder = max_len % out_steps
+    pad_len = max_len + (out_steps - remainder) if remainder > 0 else max_len
+    return np.stack([_pad_stop_target(x, pad_len) for x in inputs])
 
 
 def pad_per_step(inputs, pad_len):