diff --git a/codes/data/audio/unsupervised_audio_dataset.py b/codes/data/audio/unsupervised_audio_dataset.py
index ce73c440..5118329a 100644
--- a/codes/data/audio/unsupervised_audio_dataset.py
+++ b/codes/data/audio/unsupervised_audio_dataset.py
@@ -168,7 +168,7 @@ class UnsupervisedAudioDataset(torch.utils.data.Dataset):
         output = {
             'prepad_length': prepad_length,
             'clip': clips[0],
-            'clip_lengths': torch.tensor(audio_norm.shape[-1]),
+            'clip_lengths': torch.tensor(clips[0].shape[-1]),
             'path': filename,
         }
         if self.should_resample_clip: