Add batch processing pipeline with xmap_reader.

deep_speech_2/data_utils/data.py

@@ -186,13 +186,18 @@ class DataGenerator(object):
             for instance in instance_reader():
                 batch.append(instance)
                 if len(batch) == batch_size:
-                    yield self._padding_batch(batch, padding_to, flatten)
+                    yield batch
                     batch = []
             if len(batch) >= min_batch_size:
-                yield self._padding_batch(batch, padding_to, flatten)
+                yield batch
             self._epoch += 1
 
-        return batch_reader
+        return paddle.reader.xmap_readers(
+            lambda batch: self._padding_batch(batch, padding_to, flatten),
+            batch_reader,
+            process_num=1,
+            buffer_size=8,
+            order=True)
 
     @property
     def feeding(self):

deep_speech_2/train.py

@@ -101,7 +101,7 @@ def train():
         rnn_layer_size=args.rnn_layer_size,
         use_gru=args.use_gru,
         pretrained_model_path=args.init_model_path,
-        share_rnn_weights=args.share_weights)
+        share_rnn_weights=args.share_rnn_weights)
     ds2_model.train(
         train_batch_reader=train_batch_reader,
         dev_batch_reader=dev_batch_reader,