diff --git a/datasets/TTSDataset.py b/datasets/TTSDataset.py
index 0305da7f..dea6cb8c 100644
--- a/datasets/TTSDataset.py
+++ b/datasets/TTSDataset.py
@@ -103,7 +103,6 @@ class MyDataset(Dataset):
         if self.enable_eos_bos:
             phonemes = pad_with_eos_bos(phonemes)
             phonemes = np.asarray(phonemes, dtype=np.int32)
-
         return phonemes
 
     def load_data(self, idx):
diff --git a/layers/common_layers.py b/layers/common_layers.py
index 98fc70ae..d5836a9f 100644
--- a/layers/common_layers.py
+++ b/layers/common_layers.py
@@ -234,7 +234,7 @@ class Attention(nn.Module):
                 query, processed_inputs)
         # apply masking
         if mask is not None:
-            attention.data.masked_fill_(torch.bitwise_not(mask), self._mask_value)
+            attention.data.masked_fill_(~mask, self._mask_value)
         # apply windowing - only in eval mode
         if not self.training and self.windowing:
             attention = self.apply_windowing(attention, inputs)
diff --git a/layers/tacotron.py b/layers/tacotron.py
index 329fdb11..788e5230 100644
--- a/layers/tacotron.py
+++ b/layers/tacotron.py
@@ -315,7 +315,7 @@ class Decoder(nn.Module):
         # learn init values instead of zero init.
         self.stopnet = StopNet(256 + memory_dim * self.r_init)
 
-    def _set_r(self, new_r):
+    def set_r(self, new_r):
         self.r = new_r
 
     def _reshape_memory(self, memory):