diff --git a/modelscope/outputs/outputs.py b/modelscope/outputs/outputs.py
index 8bff70da..268556ef 100644
--- a/modelscope/outputs/outputs.py
+++ b/modelscope/outputs/outputs.py
@@ -723,6 +723,10 @@ TASK_OUTPUTS = {
     # { "text": "每一天都要快乐喔"}
     Tasks.auto_speech_recognition: [OutputKeys.TEXT],
 
+    # itn result for single sample
+    # {"text": "123"}
+    Tasks.inverse_text_processing: [OutputKeys.TEXT],
+
     # speaker verification for single compare task
     # {'score': 84.2332}
     Tasks.speaker_verification: [OutputKeys.SCORES],
diff --git a/modelscope/pipelines/audio/inverse_text_processing_pipeline.py b/modelscope/pipelines/audio/inverse_text_processing_pipeline.py
index 8e75d548..f5282691 100644
--- a/modelscope/pipelines/audio/inverse_text_processing_pipeline.py
+++ b/modelscope/pipelines/audio/inverse_text_processing_pipeline.py
@@ -44,14 +44,16 @@ class InverseTextProcessingPipeline(Pipeline):
         super().__init__(model=model, **kwargs)
         self.model_cfg = self.model.forward()
 
-    def __call__(self, text_in: str = None) -> str:
+    def __call__(self, text_in: str = None) -> Dict[str, Any]:
 
         if len(text_in) == 0:
             raise ValueError('The input of ITN should not be null.')
         else:
             self.text_in = text_in
+        output = {}
+        itn_result = self.forward(self.text_in)
+        output['text'] = itn_result
 
-        output = self.forward(self.text_in)
         return output
 
     def postprocess(self, inputs: Dict[str, Any]) -> Dict[str, Any]: