Merge pull request #813 from PaddlePaddle/sh

relax librispeech audio max len to 30 second

deepspeech/io/dataloader.py

@@ -44,7 +44,7 @@ def feat_dim_and_vocab_size(data_json: List[Dict[Text, Any]],
 
 
 def batch_collate(x):
-    """de-tuple.
+    """de-minibatch, since user compose batch.
 
     Args:
         x (List[Tuple]): [(utts, xs, ilens, ys, olens)]

examples/librispeech/s0/conf/deepspeech2.yaml

@@ -4,7 +4,7 @@ data:
   dev_manifest: data/manifest.dev-clean
   test_manifest: data/manifest.test-clean
   min_input_len: 0.0
-  max_input_len: 27.0 # second
+  max_input_len: 30.0 # second
   min_output_len: 0.0
   max_output_len: .inf
   min_output_input_ratio: 0.00

examples/librispeech/s0/conf/deepspeech2_online.yaml

@@ -4,7 +4,7 @@ data:
   dev_manifest: data/manifest.dev-clean
   test_manifest: data/manifest.test-clean
   min_input_len: 0.0
-  max_input_len: 27.0 # second
+  max_input_len: 30.0 # second
   min_output_len: 0.0
   max_output_len: .inf
   min_output_input_ratio: 0.00

examples/librispeech/s2/conf/transformer.yaml

@@ -22,7 +22,7 @@ collator:
   batch_frames_out: 0
   batch_frames_inout: 0
   augmentation_config: conf/augmentation.json
-  num_workers: 2
+  num_workers: 0
   subsampling_factor: 1
   num_encs: 1
 

examples/tiny/s0/conf/deepspeech2.yaml

@@ -4,7 +4,7 @@ data:
   dev_manifest: data/manifest.tiny
   test_manifest: data/manifest.tiny 
   min_input_len: 0.0
-  max_input_len: 27.0
+  max_input_len: 30.0
   min_output_len: 0.0
   max_output_len: 400.0
   min_output_input_ratio: 0.05

examples/tiny/s1/conf/chunk_confermer.yaml

@@ -4,7 +4,7 @@ data:
   dev_manifest: data/manifest.tiny
   test_manifest: data/manifest.tiny
   min_input_len: 0.5  # second
-  max_input_len: 20.0 # second
+  max_input_len: 30.0 # second
   min_output_len: 0.0 # tokens
   max_output_len: 400.0 # tokens
   min_output_input_ratio: 0.05