diff --git a/g2p_seq2seq/g2p.py b/g2p_seq2seq/g2p.py
index a7e76ac0..2cd16356 100644
--- a/g2p_seq2seq/g2p.py
+++ b/g2p_seq2seq/g2p.py
@@ -294,7 +294,7 @@ def decode(self, output_file_path):
       # If path to the output file pointed out, dump decoding results to the file
       if output_file_path:
         tf.logging.info("Writing decodes into %s" % output_file_path)
-        outfile = tf.gfile.Open(output_file_path, "w")
+        outfile = tf.gfile.Open(output_file_path, "wb")
 
       inputs, decodes = self.__decode_from_file(self.test_path, outfile)
 
@@ -542,7 +542,7 @@ def _get_inputs(filename, delimiters="\t "):
   delimiters_regex = re.compile("[" + delimiters + "]+")
 
   inputs = []
-  with tf.gfile.Open(filename) as input_file:
+  with tf.gfile.Open(filename, "rb") as input_file:
     lines = input_file.readlines()
     for line in lines:
       if set("[" + delimiters + "]+$").intersection(line):
diff --git a/g2p_seq2seq/g2p_encoder.py b/g2p_seq2seq/g2p_encoder.py
index a35cf128..13a6b384 100644
--- a/g2p_seq2seq/g2p_encoder.py
+++ b/g2p_seq2seq/g2p_encoder.py
@@ -95,7 +95,7 @@ def _init_vocab_from_file(self, filename):
     """
     def sym_gen():
       """Symbols generator for vocab initializer from file."""
-      with tf.gfile.Open(filename) as vocab_file:
+      with tf.gfile.Open(filename, "rb") as vocab_file:
         for line in vocab_file:
           sym = line.strip()
           yield sym
@@ -144,7 +144,7 @@ def store_to_file(self, filename):
     Args:
       filename: Full path of the file to store the vocab to.
     """
-    with tf.gfile.Open(filename, "w") as vocab_file:
+    with tf.gfile.Open(filename, "wb") as vocab_file:
       for i in range(len(self._id_to_sym)):
         vocab_file.write(self._id_to_sym[i] + "\n")
 
diff --git a/g2p_seq2seq/params.py b/g2p_seq2seq/params.py
index e087c66a..bcabbd91 100644
--- a/g2p_seq2seq/params.py
+++ b/g2p_seq2seq/params.py
@@ -48,7 +48,7 @@ def __init__(self, model_dir, data_path, flags=None):
       self.iterations_per_loop = min(1000, max(10, int(self.batch_size/10)))
       if flags.max_epochs > 0:
         self.train_steps = max(10000, 
-                               int(len(open(data_path).readlines()) /\
+                               int(len(open(data_path, encoding="utf-8").readlines()) /\
                                    self.batch_size) *\
                                self.iterations_per_loop *\
                                flags.max_epochs)