Merge pull request #640 from zhxfl/fix-639

Augmentation should compute frame_dim

fluid/DeepASR/data_utils/data_reader.py

@@ -121,8 +121,6 @@ class DataReader(object):
                                  corresponding description file.
         label_file_list (str): File containing paths of label data file and 
                                corresponding description file.
-        frame_dim (int): The final feature dimension of one frame after all 
-                         augmentation applied.
         drop_frame_len (int): Samples whose label length above the value will be
                               dropped.
         process_num (int): Number of processes for processing data.
@@ -137,21 +135,18 @@ class DataReader(object):
         random_seed (int): Random seed.
     """
 
-    def __init__(
-            self,
-            feature_file_list,
-            label_file_list,
-            frame_dim=120 * 11,  # @TODO augmentor is responsible for the value
-            drop_frame_len=512,
-            process_num=10,
-            sample_buffer_size=1024,
-            sample_info_buffer_size=1024,
-            batch_buffer_size=1024,
-            shuffle_block_num=1,
-            random_seed=0):
+    def __init__(self,
+                 feature_file_list,
+                 label_file_list,
+                 drop_frame_len=512,
+                 process_num=10,
+                 sample_buffer_size=1024,
+                 sample_info_buffer_size=1024,
+                 batch_buffer_size=1024,
+                 shuffle_block_num=1,
+                 random_seed=0):
         self._feature_file_list = feature_file_list
         self._label_file_list = label_file_list
-        self._frame_dim = frame_dim
         self._drop_frame_len = drop_frame_len
         self._shuffle_block_num = shuffle_block_num
         self._block_info_list = None
@@ -300,8 +295,9 @@ class DataReader(object):
 
     def batch_iterator(self, batch_size, minimum_batch_size):
         def batch_to_ndarray(batch_samples, lod):
-            batch_feature = np.zeros(
-                (lod[-1], self._frame_dim), dtype="float32")
+            assert len(batch_samples)
+            frame_dim = batch_samples[0][0].shape[1]
+            batch_feature = np.zeros((lod[-1], frame_dim), dtype="float32")
             batch_label = np.zeros((lod[-1], 1), dtype="int64")
             start = 0
             for sample in batch_samples: