Add audio_loss_multiplier to scale audio loss to larger values if desired.

extensions_built_in/sd_trainer/SDTrainer.py

@@ -863,6 +863,7 @@ class SDTrainer(BaseSDTrainProcess):
         # check for audio loss
         if batch.audio_pred is not None and batch.audio_target is not None:
             audio_loss = torch.nn.functional.mse_loss(batch.audio_pred.float(), batch.audio_target.float(), reduction="mean")
+            audio_loss = audio_loss * self.train_config.audio_loss_multiplier
             loss = loss + audio_loss
 
         # check for additional losses