small enhanecements (#1322)

- add extra check of 'x' and 'x_lens' to earlier point in Transducer model
- specify 'utf' encoding when opening text files for writing (recogs,
  errs)

egs/librispeech/ASR/pruned_transducer_stateless7/model.py

@@ -114,6 +114,9 @@ class Transducer(nn.Module):
 
         assert x.size(0) == x_lens.size(0) == y.dim0
 
+        # x.T_dim == max(x_len)
+        assert x.size(1) == x_lens.max().item(), (x.shape, x_lens, x_lens.max())
+
         encoder_out, x_lens = self.encoder(x, x_lens)
         assert torch.all(x_lens > 0)
 

icefall/utils.py

@@ -498,7 +498,7 @@ def store_transcripts(
     Returns:
       Return None.
     """
-    with open(filename, "w") as f:
+    with open(filename, "w", encoding="utf8") as f:
         for cut_id, ref, hyp in texts:
             if char_level:
                 ref = list("".join(ref))
@@ -523,7 +523,7 @@ def store_transcripts_and_timestamps(
     Returns:
       Return None.
     """
-    with open(filename, "w") as f:
+    with open(filename, "w", encoding="utf8") as f:
         for cut_id, ref, hyp, time_ref, time_hyp in texts:
             print(f"{cut_id}:\tref={ref}", file=f)
             print(f"{cut_id}:\thyp={hyp}", file=f)