Reverse PR

.travis.yml

@@ -21,13 +21,20 @@ before_install:
   # Useful for debugging any issues with conda
   - conda config --add channels pypi
   - conda info -a
-  - deps='pip numpy scipy cython nose pytorch'
+  - deps='pip numpy scipy cython nose pytorch flake8'
   - conda create -q -n test-environment "python=$TRAVIS_PYTHON_VERSION" $deps -c pytorch
   - source activate test-environment
 
 install:
   - pip install -e ".[test]"
   - python -c "import nltk; nltk.download('cmudict')"
 
+before_script:
+  # stop the build if there are Python syntax errors or undefined names
+  - flake8 . --count --select=E901,E999,F821,F822,F823 --show-source --statistics
+  # exit-zero treats all errors as warnings.  The GitHub editor is 127 chars wide
+  - flake8 . --count --exit-zero --max-complexity=10 --max-line-length=127 --statistics
+
 script:
   - nosetests -v -w tests/ -a '!local_only'
+

README.md

@@ -5,7 +5,7 @@
 
 PyTorch implementation of convolutional networks-based text-to-speech synthesis models:
 
-1. [arXiv:1710.07654](https://arxiv.org/abs/1710.07654): Deep Voice 3: 2000-Speaker Neural Text-to-Speech.
+1. [arXiv:1710.07654](https://arxiv.org/abs/1710.07654): Deep Voice 3: Scaling Text-to-Speech with Convolutional Sequence Learning.
 2. [arXiv:1710.08969](https://arxiv.org/abs/1710.08969): Efficiently Trainable Text-to-Speech System Based on Deep Convolutional Networks with Guided Attention.
 
 Audio samples are available at https://r9y9.github.io/deepvoice3_pytorch/.

deepvoice3_pytorch/conv.py

@@ -40,7 +40,7 @@ def incremental_forward(self, input):
                 self.input_buffer[:, :-1, :] = self.input_buffer[:, 1:, :].clone()
             # append next input
             self.input_buffer[:, -1, :] = input[:, -1, :]
-            input = torch.autograd.Variable(self.input_buffer, volatile=True)
+            input = torch.Tensor(self.input_buffer)
             if dilation > 1:
                 input = input[:, 0::dilation, :].contiguous()
         output = F.linear(input.view(bsz, -1), weight, self.bias)

deepvoice3_pytorch/modules.py

@@ -48,22 +48,18 @@ def forward(self, x, w=1.0):
 
         if isscaler or w.size(0) == 1:
             weight = sinusoidal_encode(self.weight, w)
-            return self._backend.Embedding.apply(
-                x, weight,
-                padding_idx, self.max_norm, self.norm_type,
-                self.scale_grad_by_freq, self.sparse
-            )
+            return F.embedding(
+                x, weight, padding_idx, self.max_norm,
+                self.norm_type, self.scale_grad_by_freq, self.sparse)
         else:
             # TODO: cannot simply apply for batch
             # better to implement efficient function
             pe = []
             for batch_idx, we in enumerate(w):
                 weight = sinusoidal_encode(self.weight, we)
-                pe.append(self._backend.Embedding.apply(
-                    x[batch_idx], weight,
-                    padding_idx, self.max_norm, self.norm_type,
-                    self.scale_grad_by_freq, self.sparse
-                ))
+                pe.append(F.embedding(
+                    x[batch_idx], weight, padding_idx, self.max_norm,
+                    self.norm_type, self.scale_grad_by_freq, self.sparse))
             pe = torch.stack(pe)
             return pe
 

docs/content/index.md

@@ -392,5 +392,5 @@ Your browser does not support the audio element.
 
 ## References
 
-- [Wei Ping, Kainan Peng, Andrew Gibiansky, et al, "Deep Voice 3: 2000-Speaker Neural Text-to-Speech", arXiv:1710.07654, Oct. 2017.](https://arxiv.org/abs/1710.07654)
+- [Wei Ping, Kainan Peng, Andrew Gibiansky, et al, "Deep Voice 3: Scaling Text-to-Speech with Convolutional Sequence Learning", arXiv:1710.07654, Oct. 2017.](https://arxiv.org/abs/1710.07654)
 - [Hideyuki Tachibana, Katsuya Uenoyama, Shunsuke Aihara, "Efficiently Trainable Text-to-Speech System Based on Deep Convolutional Networks with Guided Attention". arXiv:1710.08969, Oct 2017.](https://arxiv.org/abs/1710.08969)

nikl_m.py

@@ -5,6 +5,8 @@
 import audio
 import re
 
+from hparams import hparams
+
 
 def build_from_path(in_dir, out_dir, num_workers=1, tqdm=lambda x: x):
     '''Preprocesses the LJ Speech dataset from a given input path into a given output directory.

nikl_s.py

@@ -5,6 +5,8 @@
 import audio
 import re
 
+from hparams import hparams
+
 
 def build_from_path(in_dir, out_dir, num_workers=1, tqdm=lambda x: x):
     '''Preprocesses the LJ Speech dataset from a given input path into a given output directory.

tests/test_deepvoice3.py

@@ -255,7 +255,7 @@ def test_incremental_forward():
         max_input_len = np.max(input_lengths) + 10  # manuall padding
         seqs = np.array([_pad(x, max_input_len) for x in seqs], dtype=np.int)
         input_lengths = torch.LongTensor(input_lengths)
-        input_lengths = input_lengths.cuda() if use_cuda else input_lenghts
+        input_lengths = input_lengths.cuda() if use_cuda else input_lengths
     else:
         input_lengths = None
 

train.py

@@ -728,30 +728,30 @@ def train(model, data_loader, optimizer, writer,
             # Update
             loss.backward()
             if clip_thresh > 0:
-                grad_norm = torch.nn.utils.clip_grad_norm(
+                grad_norm = torch.nn.utils.clip_grad_norm_(
                     model.get_trainable_parameters(), clip_thresh)
             optimizer.step()
 
             # Logs
-            writer.add_scalar("loss", float(loss.data[0]), global_step)
+            writer.add_scalar("loss", float(loss.item()), global_step)
             if train_seq2seq:
-                writer.add_scalar("done_loss", float(done_loss.data[0]), global_step)
-                writer.add_scalar("mel loss", float(mel_loss.data[0]), global_step)
-                writer.add_scalar("mel_l1_loss", float(mel_l1_loss.data[0]), global_step)
-                writer.add_scalar("mel_binary_div_loss", float(mel_binary_div.data[0]), global_step)
+                writer.add_scalar("done_loss", float(done_loss.item()), global_step)
+                writer.add_scalar("mel loss", float(mel_loss.item()), global_step)
+                writer.add_scalar("mel_l1_loss", float(mel_l1_loss.item()), global_step)
+                writer.add_scalar("mel_binary_div_loss", float(mel_binary_div.item()), global_step)
             if train_postnet:
-                writer.add_scalar("linear_loss", float(linear_loss.data[0]), global_step)
-                writer.add_scalar("linear_l1_loss", float(linear_l1_loss.data[0]), global_step)
+                writer.add_scalar("linear_loss", float(linear_loss.item()), global_step)
+                writer.add_scalar("linear_l1_loss", float(linear_l1_loss.item()), global_step)
                 writer.add_scalar("linear_binary_div_loss", float(
-                    linear_binary_div.data[0]), global_step)
+                    linear_binary_div.item()), global_step)
             if train_seq2seq and hparams.use_guided_attention:
-                writer.add_scalar("attn_loss", float(attn_loss.data[0]), global_step)
+                writer.add_scalar("attn_loss", float(attn_loss.item()), global_step)
             if clip_thresh > 0:
                 writer.add_scalar("gradient norm", grad_norm, global_step)
             writer.add_scalar("learning rate", current_lr, global_step)
 
             global_step += 1
-            running_loss += loss.data[0]
+            running_loss += loss.item()
 
         averaged_loss = running_loss / (len(data_loader))
         writer.add_scalar("loss (per epoch)", averaged_loss, global_epoch)