diff --git a/models/tacotron.py b/models/tacotron.py
index 9f432a6c..5b4509c7 100644
--- a/models/tacotron.py
+++ b/models/tacotron.py
@@ -77,5 +77,5 @@ class Tacotron(nn.Module):
             speaker_embeddings = speaker_embeddings.expand(encoder_outputs.size(0),
                                                            encoder_outputs.size(1),
                                                            -1)
-            encoder_outputs += speaker_embeddings
+            encoder_outputs = encoder_outputs + speaker_embeddings
         return encoder_outputs
diff --git a/models/tacotron2.py b/models/tacotron2.py
index 499ec28a..c892824a 100644
--- a/models/tacotron2.py
+++ b/models/tacotron2.py
@@ -96,5 +96,5 @@ class Tacotron2(nn.Module):
             speaker_embeddings = speaker_embeddings.expand(encoder_outputs.size(0),
                                                            encoder_outputs.size(1),
                                                            -1)
-            encoder_outputs += speaker_embeddings
+            encoder_outputs = encoder_outputs + speaker_embeddings
         return encoder_outputs
diff --git a/models/tacotrongst.py b/models/tacotrongst.py
index b0a14c54..5176e051 100644
--- a/models/tacotrongst.py
+++ b/models/tacotrongst.py
@@ -86,5 +86,5 @@ class TacotronGST(nn.Module):
             speaker_embeddings = speaker_embeddings.expand(encoder_outputs.size(0),
                                                            encoder_outputs.size(1),
                                                            -1)
-            encoder_outputs += speaker_embeddings
+            encoder_outputs = encoder_outputs + speaker_embeddings
         return encoder_outputs