diff --git a/audio/README.md b/audio/README.md
index f336ac9ae1a1ab2f8241a8cc72eaa5541a32c03a..bfd8625f0b6268489b1ea69b8aa334c978f7165c 100644
--- a/audio/README.md
+++ b/audio/README.md
@@ -29,7 +29,6 @@ MAC：test build whl envrioment：
 * gcc/g++/gfortran 12.2.0
 * cpu Intel Xeon E5 x86_64
 
-
 Windows：
 not support： paddleaudio C++ extension lib (sox io, kaldi native fbank)
-python setup.py bdist_wheel
\ No newline at end of file
+python setup.py bdist_wheel
diff --git a/audio/paddleaudio/src/pybind/pybind.cpp b/audio/paddleaudio/src/pybind/pybind.cpp
index c4dfa8d519d80af4b3b2906c43d7f34a2e1f3dcf..692e809954eae888749cd7a7400cab116f678258 100644
--- a/audio/paddleaudio/src/pybind/pybind.cpp
+++ b/audio/paddleaudio/src/pybind/pybind.cpp
@@ -1,7 +1,9 @@
 // Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
 
+#ifdef INCLUDE_KALDI
 #include "paddleaudio/src/pybind/kaldi/kaldi_feature.h"
 #include "paddleaudio/third_party/kaldi/feat/feature-fbank.h"
+#endif
 
 #ifdef INCLUDE_SOX
 #include "paddleaudio/src/pybind/sox/io.h"
diff --git a/paddlespeech/s2t/models/u2_st/u2_st.py b/paddlespeech/s2t/models/u2_st/u2_st.py
index 016087d682079495974718a92c03c90c14aa21b4..31defbbaf1b751ac6c6331cc6aff5d52277ca833 100644
--- a/paddlespeech/s2t/models/u2_st/u2_st.py
+++ b/paddlespeech/s2t/models/u2_st/u2_st.py
@@ -24,9 +24,9 @@ from typing import Tuple
 import paddle
 from paddle import jit
 from paddle import nn
-from paddleaudio.utils.tensor_utils import add_sos_eos
-from paddleaudio.utils.tensor_utils import th_accuracy
 
+from paddlespeech.audio.utils.tensor_utils import add_sos_eos
+from paddlespeech.audio.utils.tensor_utils import th_accuracy
 from paddlespeech.s2t.frontend.utility import IGNORE_ID
 from paddlespeech.s2t.frontend.utility import load_cmvn
 from paddlespeech.s2t.modules.cmvn import GlobalCMVN