diff --git a/egs/librispeech/ASR/pruned_transducer_stateless7/zipformer.py b/egs/librispeech/ASR/pruned_transducer_stateless7/zipformer.py
index ed1e2efa2..71f12e44a 100644
--- a/egs/librispeech/ASR/pruned_transducer_stateless7/zipformer.py
+++ b/egs/librispeech/ASR/pruned_transducer_stateless7/zipformer.py
@@ -741,7 +741,7 @@ class DownsampledZipformerEncoder(nn.Module):
             src,
             feature_mask=feature_mask,
             mask=mask,
-            src_key_padding_mask=mask,
+            src_key_padding_mask=src_key_padding_mask,
         )
         src = self.upsample(src)
         # remove any extra frames that are not a multiple of downsample_factor