diff --git a/codes/data/audio/nv_tacotron_dataset.py b/codes/data/audio/nv_tacotron_dataset.py
index c6d509c6..3b2902d8 100644
--- a/codes/data/audio/nv_tacotron_dataset.py
+++ b/codes/data/audio/nv_tacotron_dataset.py
@@ -84,7 +84,11 @@ class TextWavLoader(torch.utils.data.Dataset):
         return text_norm
 
     def __getitem__(self, index):
-        tseq, wav, text, path = self.get_wav_text_pair(self.audiopaths_and_text[index])
+        try:
+            tseq, wav, text, path = self.get_wav_text_pair(self.audiopaths_and_text[index])
+        except:
+            print(f"error loadding {self.audiopaths_and_text[index][0]")
+            return self[index+1]
         if wav is None or \
             (self.max_wav_len is not None and wav.shape[-1] > self.max_wav_len) or \
             (self.max_text_len is not None and tseq.shape[0] > self.max_text_len):