diff --git a/paddlespeech/s2t/__init__.py b/paddlespeech/s2t/__init__.py
index b67322cdc2f4ac305b42b399ea5feb87207e0375..4507365d6e7e9fabfd45c1f1467b8fb654c3c2bc 100644
--- a/paddlespeech/s2t/__init__.py
+++ b/paddlespeech/s2t/__init__.py
@@ -166,7 +166,6 @@ def broadcast_shape(shp1, shp2):
 def masked_fill(xs: paddle.Tensor,
                 mask: paddle.Tensor,
                 value: Union[float, int]):
-    mask.stop_gradient = True
     mask = mask.astype(xs.dtype)
     return xs * (1.0 - mask) + mask * value