Merge pull request #779 from PaddlePaddle/aug

fix specaug config

deepspeech/frontend/augmentor/spec_augment.py

@@ -245,6 +245,8 @@ class SpecAugmentor(AugmentorBase):
         Returns:
             x (np.ndarray): `[T, F]`
         """
+        assert isinstance(x, np.ndarray)
+        assert x.ndim == 2
         x = self.time_warp(x, self.mode)
         x = self.mask_freq(x, self.replace_with_zero)
         x = self.mask_time(x, self.replace_with_zero)

examples/aishell/s0/conf/augmentation.json

@@ -19,16 +19,17 @@
   {
     "type": "specaug",
     "params": {
-      "F": 10,
+      "W": 5,
-      "T": 50,
+      "warp_mode": "PIL",
+      "F": 30,
       "n_freq_masks": 2,
+      "T": 40,
       "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
+      "replace_with_zero": false
     },
     "prob": 1.0
   }

examples/aishell/s1/conf/augmentation.json

@@ -28,7 +28,8 @@
       "adaptive_number_ratio": 0,
       "adaptive_size_ratio": 0,
       "max_n_time_masks": 20,
-      "replace_with_zero": true
+      "replace_with_zero": true,
+      "warp_mode": "PIL"
     },
     "prob": 1.0
   }

examples/augmentation/augmentation.json

@@ -52,17 +52,18 @@
   {
     "type": "specaug",
     "params": {
+      "W": 80,
+      "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
+      "replace_with_zero": false
     },
-    "prob": 0.0
+    "prob": 1.0
   }
 ]

examples/librispeech/s0/conf/augmentation.json

@@ -28,7 +28,8 @@
       "adaptive_number_ratio": 0,
       "adaptive_size_ratio": 0,
       "max_n_time_masks": 20,
-      "replace_with_zero": true
+      "replace_with_zero": true,
+      "warp_mode": "PIL"
     },
     "prob": 1.0
   }

examples/librispeech/s1/conf/augmentation.json

@@ -28,7 +28,8 @@
       "adaptive_number_ratio": 0,
       "adaptive_size_ratio": 0,
       "max_n_time_masks": 20,
-      "replace_with_zero": true
+      "replace_with_zero": true,
+      "warp_mode": "PIL"
     },
     "prob": 1.0
   }

examples/librispeech/s2/conf/augmentation.json

@@ -11,7 +11,8 @@
       "adaptive_number_ratio": 0,
       "adaptive_size_ratio": 0,
       "max_n_time_masks": 20,
-      "replace_with_zero": true
+      "replace_with_zero": true,
+      "warp_mode": "PIL"
     },
     "prob": 1.0
   }

examples/timit/s1/conf/augmentation.json

@@ -28,7 +28,8 @@
       "adaptive_number_ratio": 0,
       "adaptive_size_ratio": 0,
       "max_n_time_masks": 20,
-      "replace_with_zero": true
+      "replace_with_zero": true,
+      "warp_mode": "PIL"
     },
     "prob": 1.0
   }

examples/tiny/s0/conf/augmentation.json

@@ -6,7 +6,7 @@
       "max_speed_rate": 1.1,
       "num_rates": 3
     },
-    "prob": 1.0
+    "prob": 0.0
   },
   {
     "type": "shift",
@@ -19,16 +19,18 @@
   {
     "type": "specaug",
     "params": {
-      "F": 10,
+      "W": 5,
-      "T": 50,
+      "warp_mode": "PIL",
+      "F": 30,
       "n_freq_masks": 2,
+      "T": 40,
       "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
+      "replace_with_zero": true,
+      "warp_mode": "PIL"
     },
     "prob": 1.0
   }

examples/tiny/s1/conf/augmentation.json

@@ -28,7 +28,8 @@
       "adaptive_number_ratio": 0,
       "adaptive_size_ratio": 0,
       "max_n_time_masks": 20,
-      "replace_with_zero": true
+      "replace_with_zero": true,
+      "warp_mode": "PIL"
     },
     "prob": 1.0
   }