update

funasr/datasets/large_datasets/utils/padding.py

@@ -13,16 +13,16 @@ def padding(data, float_pad_value=0.0, int_pad_value=-1):
     batch = {}
     data_names = data[0].keys()
     for data_name in data_names:
-        if data_name == "key" or data_name == "sampling_rate" or data_name == 'hotword_indxs':
-            batch[data_name] = data[0][data_name]
+        if data_name == "key" or data_name == "sampling_rate":
             continue
         else:
-            if data[0][data_name].dtype.kind == "i":
-                pad_value = int_pad_value
-                tensor_type = torch.int64
-            else:
-                pad_value = float_pad_value
-                tensor_type = torch.float32
+            if data_name != 'hotword_indxs':
+                if data[0][data_name].dtype.kind == "i":
+                    pad_value = int_pad_value
+                    tensor_type = torch.int64
+                else:
+                    pad_value = float_pad_value
+                    tensor_type = torch.float32
 
             tensor_list = [torch.tensor(np.copy(d[data_name]), dtype=tensor_type) for d in data]
             tensor_lengths = torch.tensor([len(d[data_name]) for d in data], dtype=torch.int32)