Merge pull request #1449 from Jackwaterveg/fix

[ASR] fix sortagrad

paddlespeech/s2t/io/batchfy.py

@@ -419,7 +419,7 @@ def make_batchset(
         # sort it by input lengths (long to short)
         sorted_data = sorted(
             d.items(),
-            key=lambda data: int(data[1][batch_sort_key][batch_sort_axis]["shape"][0]),
+            key=lambda data: float(data[1][batch_sort_key][batch_sort_axis]["shape"][0]),
             reverse=not shortest_first, )
         logger.info("# utts: " + str(len(sorted_data)))
 

paddlespeech/s2t/io/dataloader.py

@@ -61,7 +61,7 @@ class BatchDataLoader():
     def __init__(self,
                  json_file: str,
                  train_mode: bool,
-                 sortagrad: bool=False,
+                 sortagrad: int=0,
                  batch_size: int=0,
                  maxlen_in: float=float('inf'),
                  maxlen_out: float=float('inf'),