Merge branch 'develop' of https://github.com/PaddlePaddle/DeepSpeech into ds2_online

deepspeech/exps/u2_kaldi/model.py

@@ -228,7 +228,7 @@ class U2Trainer(Trainer):
             maxlen_in=float('inf'),
             maxlen_out=float('inf'),
             minibatches=0,
-            mini_batch_size=1,
+            mini_batch_size=self.args.nprocs,
             batch_count='auto',
             batch_bins=0,
             batch_frames_in=0,
@@ -247,7 +247,7 @@ class U2Trainer(Trainer):
             maxlen_in=float('inf'),
             maxlen_out=float('inf'),
             minibatches=0,
-            mini_batch_size=1,
+            mini_batch_size=self.args.nprocs,
             batch_count='auto',
             batch_bins=0,
             batch_frames_in=0,
@@ -263,7 +263,7 @@ class U2Trainer(Trainer):
             json_file=config.data.test_manifest,
             train_mode=False,
             sortagrad=False,
-            batch_size=config.collator.batch_size,
+            batch_size=config.decoding.batch_size,
             maxlen_in=float('inf'),
             maxlen_out=float('inf'),
             minibatches=0,
@@ -282,7 +282,7 @@ class U2Trainer(Trainer):
             json_file=config.data.test_manifest,
             train_mode=False,
             sortagrad=False,
-            batch_size=config.collator.batch_size,
+            batch_size=config.decoding.batch_size,
             maxlen_in=float('inf'),
             maxlen_out=float('inf'),
             minibatches=0,

deepspeech/frontend/augmentor/spec_augment.py

@@ -151,6 +151,9 @@ class SpecAugmentor(AugmentorBase):
             np.ndarray: time warped spectrogram (time, freq)
         """
         window = max_time_warp = self.W
+        if window == 0:
+            return x
+        
         if mode == "PIL":
             t = x.shape[0]
             if t - window <= window:

deepspeech/frontend/utility.py

@@ -46,7 +46,7 @@ def load_dict(dict_path: Optional[Text], maskctc=False) -> Optional[List[Text]]:
 
     with open(dict_path, "r") as f:
         dictionary = f.readlines()
-    char_list = [entry.split(" ")[0] for entry in dictionary]
+    char_list = [entry.strip().split(" ")[0] for entry in dictionary]
     if BLANK not in char_list:
         char_list.insert(0, BLANK)
     if EOS not in char_list:

examples/aishell/s0/README.md

@@ -1,10 +1,18 @@
 # Aishell-1
 
+## Data
+| Data Subset | Duration in Seconds |
+| data/manifest.train |  1.23 ~ 14.53125 |
+| data/manifest.dev | 1.645 ~ 12.533 |  
+| data/manifest.test | 1.859125 ~ 14.6999375 |
+
+`jq '.feat_shape[0]' data/manifest.train | sort -un`
+
 ## Deepspeech2
 
 | Model | Params | Release | Config | Test set | Loss | CER |  
 | --- | --- | --- | --- | --- | --- | --- |  
-| DeepSpeech2 | 58.4M | 2.2.0 | conf/deepspeech2.yaml + spec aug + new datapipe | test | 6.396368026733398 | 0.068382,0.073507 |  
+| DeepSpeech2 | 58.4M | 2.2.0 | conf/deepspeech2.yaml + spec aug + new datapipe | test | 6.396368026733398 | 0.068382 |  
 | DeepSpeech2 | 58.4M | 2.1.0 | conf/deepspeech2.yaml + spec aug | test | 7.483316898345947 | 0.077860 |  
 | DeepSpeech2 | 58.4M | 2.1.0 | conf/deepspeech2.yaml | test | 7.299022197723389 | 0.078671 |
 | DeepSpeech2 | 58.4M | 2.0.0 | conf/deepspeech2.yaml | test | - | 0.078977 |  

examples/aishell/s0/conf/augmentation.json

@@ -19,17 +19,17 @@
   {
     "type": "specaug",
     "params": {
-      "W": 5,
+      "W": 0,
       "warp_mode": "PIL",
-      "F": 30,
+      "F": 10,
       "n_freq_masks": 2,
-      "T": 40,
+      "T": 50,
       "n_time_masks": 2,
       "p": 1.0,
       "adaptive_number_ratio": 0,
       "adaptive_size_ratio": 0,
       "max_n_time_masks": 20,
-      "replace_with_zero": false
+      "replace_with_zero": true 
     },
     "prob": 1.0
   }

examples/librispeech/s0/conf/augmentation.json

@@ -19,17 +19,17 @@
   {
     "type": "specaug",
     "params": {
+      "W": 0,
+      "warp_mode": "PIL",
       "F": 10,
-      "T": 50,
       "n_freq_masks": 2,
+      "T": 50,
       "n_time_masks": 2,
       "p": 1.0,
-      "W": 80,
       "adaptive_number_ratio": 0,
       "adaptive_size_ratio": 0,
       "max_n_time_masks": 20,
-      "replace_with_zero": true,
-      "warp_mode": "PIL"
+      "replace_with_zero": true
     },
     "prob": 1.0
   }