Merge branch 'develop' into network_structure

cloud/pcloud_submit.sh

@@ -1,6 +1,6 @@
 TRAIN_MANIFEST="cloud/cloud.manifest.train"
 DEV_MANIFEST="cloud/cloud.manifest.dev"
-CLOUD_MODEL_DIR="/pfs/dlnel/home/USERNAME/deepspeech2/model"
+CLOUD_MODEL_DIR="./checkpoints"
 BATCH_SIZE=256
 NUM_GPU=8
 NUM_NODE=1
@@ -11,7 +11,7 @@ DS2_PATH=${PWD%/*}
 cp -f  pcloud_train.sh ${DS2_PATH}
 
 paddlecloud submit \
--image bootstrapper:5000/wanghaoshuang/pcloud_ds2:latest \
+-image bootstrapper:5000/paddlepaddle/pcloud_ds2:latest \
 -jobname ${JOB_NAME} \
 -cpu ${NUM_GPU} \
 -gpu ${NUM_GPU} \

data_utils/data.py

@@ -85,9 +85,9 @@ class DataGenerator(object):
         self._rng = random.Random(random_seed)
         self._epoch = 0
         # for caching tar files info
-        self.local_data = local()
+        self._local_data = local()
-        self.local_data.tar2info = {}
+        self._local_data.tar2info = {}
-        self.local_data.tar2object = {}
+        self._local_data.tar2object = {}
 
     def process_utterance(self, filename, transcript):
         """Load, augment, featurize and normalize for speech data.
@@ -240,16 +240,16 @@ class DataGenerator(object):
         """
         if file.startswith('tar:'):
             tarpath, filename = file.split(':', 1)[1].split('#', 1)
-            if 'tar2info' not in self.local_data.__dict__:
+            if 'tar2info' not in self._local_data.__dict__:
-                self.local_data.tar2info = {}
+                self._local_data.tar2info = {}
-            if 'tar2object' not in self.local_data.__dict__:
+            if 'tar2object' not in self._local_data.__dict__:
-                self.local_data.tar2object = {}
+                self._local_data.tar2object = {}
-            if tarpath not in self.local_data.tar2info:
+            if tarpath not in self._local_data.tar2info:
                 object, infoes = self._parse_tar(tarpath)
-                self.local_data.tar2info[tarpath] = infoes
+                self._local_data.tar2info[tarpath] = infoes
-                self.local_data.tar2object[tarpath] = object
+                self._local_data.tar2object[tarpath] = object
-            return self.local_data.tar2object[tarpath].extractfile(
+            return self._local_data.tar2object[tarpath].extractfile(
-                self.local_data.tar2info[tarpath][filename])
+                self._local_data.tar2info[tarpath][filename])
         else:
             return open(file, 'r')
 

data_utils/featurizer/audio_featurizer.py

@@ -57,7 +57,7 @@ class AudioFeaturizer(object):
     def featurize(self,
                   audio_segment,
                   allow_downsampling=True,
-                  allow_upsamplling=True):
+                  allow_upsampling=True):
         """Extract audio features from AudioSegment or SpeechSegment.
 
         :param audio_segment: Audio/speech segment to extract features from.

layer.py

@@ -55,16 +55,20 @@ def bidirectional_simple_rnn_bn_layer(name, input, size, act):
     :rtype: LayerOutput
     """
     # input-hidden weights shared across bi-direcitonal rnn.
-    input_proj = paddle.layer.fc(
+    input_proj_forward = paddle.layer.fc(
+        input=input, size=size, act=paddle.activation.Linear(), bias_attr=False)
+    input_proj_backward = paddle.layer.fc(
         input=input, size=size, act=paddle.activation.Linear(), bias_attr=False)
     # batch norm is only performed on input-state projection
-    input_proj_bn = paddle.layer.batch_norm(
+    input_proj_bn_forward = paddle.layer.batch_norm(
-        input=input_proj, act=paddle.activation.Linear())
+        input=input_proj_forward, act=paddle.activation.Linear())
+    input_proj_bn_backward = paddle.layer.batch_norm(
+        input=input_proj_backward, act=paddle.activation.Linear())
     # forward and backward in time
     forward_simple_rnn = paddle.layer.recurrent(
-        input=input_proj_bn, act=act, reverse=False)
+        input=input_proj_bn_forward, act=act, reverse=False)
     backward_simple_rnn = paddle.layer.recurrent(
-        input=input_proj_bn, act=act, reverse=True)
+        input=input_proj_bn_backward, act=act, reverse=True)
     return paddle.layer.concat(input=[forward_simple_rnn, backward_simple_rnn])