diff --git a/modelscope/models/audio/tts/frontend/generic_text_to_speech_frontend.py b/modelscope/models/audio/tts/frontend/generic_text_to_speech_frontend.py
index ed34143f..c6aabf75 100644
--- a/modelscope/models/audio/tts/frontend/generic_text_to_speech_frontend.py
+++ b/modelscope/models/audio/tts/frontend/generic_text_to_speech_frontend.py
@@ -2,8 +2,6 @@ import os
 import zipfile
 from typing import Any, Dict, List
 
-import ttsfrd
-
 from modelscope.models.base import Model
 from modelscope.models.builder import MODELS
 from modelscope.utils.audio.tts_exceptions import (
@@ -20,6 +18,8 @@ class GenericTtsFrontend(Model):
 
     def __init__(self, model_dir='.', lang_type='pinyin', *args, **kwargs):
         super().__init__(model_dir, *args, **kwargs)
+        import ttsfrd
+
         frontend = ttsfrd.TtsFrontendEngine()
         zip_file = os.path.join(model_dir, 'resource.zip')
         self._res_path = os.path.join(model_dir, 'resource')
diff --git a/modelscope/models/audio/tts/vocoder/models/models.py b/modelscope/models/audio/tts/vocoder/models/models.py
index 83fc7dc2..c46a9204 100755
--- a/modelscope/models/audio/tts/vocoder/models/models.py
+++ b/modelscope/models/audio/tts/vocoder/models/models.py
@@ -3,7 +3,6 @@ from distutils.version import LooseVersion
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
-from pytorch_wavelets import DWT1DForward
 from torch.nn import AvgPool1d, Conv1d, Conv2d, ConvTranspose1d
 from torch.nn.utils import remove_weight_norm, spectral_norm, weight_norm
 
@@ -357,6 +356,7 @@ class MultiScaleDiscriminator(torch.nn.Module):
             DiscriminatorS(),
             DiscriminatorS(),
         ])
+        from pytorch_wavelets import DWT1DForward
         self.meanpools = nn.ModuleList(
             [DWT1DForward(wave='db3', J=1),
              DWT1DForward(wave='db3', J=1)])
diff --git a/modelscope/preprocessors/text_to_speech.py b/modelscope/preprocessors/text_to_speech.py
index fd41b752..8b8dae14 100644
--- a/modelscope/preprocessors/text_to_speech.py
+++ b/modelscope/preprocessors/text_to_speech.py
@@ -2,8 +2,6 @@
 import io
 from typing import Any, Dict, Union
 
-import ttsfrd
-
 from modelscope.fileio import File
 from modelscope.models.audio.tts.frontend import GenericTtsFrontend
 from modelscope.models.base import Model