diff --git a/codes/models/audio/mel2vec.py b/codes/models/audio/mel2vec.py
index 93bf59f2..098fafc0 100644
--- a/codes/models/audio/mel2vec.py
+++ b/codes/models/audio/mel2vec.py
@@ -663,6 +663,7 @@ class ContrastiveTrainingWrapper(nn.Module):
     def forward(self, mel, inp_lengths=None):
         mel = mel[:, :, :-1]  # The MEL computation always pads with 1, throwing off optimal tensor math.
         features_shape = (mel.shape[0], mel.shape[-1]//self.m2v.dim_reduction_mult)
+        orig_mel = mel
 
         # Frequency masking
         freq_mask_width = int(random.random() * self.freq_mask_percent * mel.shape[1])
@@ -731,7 +732,7 @@ class ContrastiveTrainingWrapper(nn.Module):
 
         if self.reconstruction:
             reconstruction = self.reconstruction_net(quantized_features.permute(0,2,1))
-            reconstruction_loss = F.mse_loss(reconstruction, mel)
+            reconstruction_loss = F.mse_loss(reconstruction, orig_mel)
             return contrastive_loss, diversity_loss, reconstruction_loss
 
         return contrastive_loss, diversity_loss