diff --git a/paddlespeech/s2t/models/wav2vec2/wav2vec2_ASR.py b/paddlespeech/s2t/models/wav2vec2/wav2vec2_ASR.py
index 64195defc..3fb91cb86 100755
--- a/paddlespeech/s2t/models/wav2vec2/wav2vec2_ASR.py
+++ b/paddlespeech/s2t/models/wav2vec2/wav2vec2_ASR.py
@@ -80,6 +80,24 @@ class Wav2vec2ASR(nn.Layer):
 
         return ctc_loss
 
+    @paddle.no_grad()
+    def extract_features(self, wav):
+        if self.normalize_wav:
+            wav = F.layer_norm(wav, wav.shape[-1])
+ 
+        # Extract wav2vec output
+        out = self.wav2vec2(wav)[0]
+        # We normalize the output if required
+        if self.output_norm:
+            out = F.layer_norm(out, out.shape[-1])
+ 
+        if self.training and hasattr(self.config, 'spec_augment'):
+            feats = self.spec_augment(out)
+        else:
+            feats = out
+ 
+        return feats
+
     @paddle.no_grad()
     def decode(self,
                feats: paddle.Tensor,