diff --git a/src/batdetect2/preprocess/types.py b/src/batdetect2/preprocess/types.py
index 39485e9..dbf3758 100644
--- a/src/batdetect2/preprocess/types.py
+++ b/src/batdetect2/preprocess/types.py
@@ -29,3 +29,6 @@ class PreprocessorProtocol(Protocol):
 
     def process_numpy(self, wav: np.ndarray) -> np.ndarray:
         return self(torch.tensor(wav)).numpy()
+
+    def cpu(self) -> "PreprocessorProtocol":
+        return self
diff --git a/src/batdetect2/train/dataset.py b/src/batdetect2/train/dataset.py
index e9e1883..570106d 100644
--- a/src/batdetect2/train/dataset.py
+++ b/src/batdetect2/train/dataset.py
@@ -272,7 +272,7 @@ def build_train_dataset(
         audio_loader=audio_loader,
         labeller=labeller,
         clipper=clipper,
-        preprocessor=preprocessor,
+        preprocessor=preprocessor.cpu(),
         audio_augmentation=audio_augmentation,
         spectrogram_augmentation=spectrogram_augmentation,
     )
@@ -305,7 +305,7 @@ def build_val_dataset(
         clip_annotations,
         audio_loader=audio_loader,
         labeller=labeller,
-        preprocessor=preprocessor,
+        preprocessor=preprocessor.cpu(),
         clipper=clipper,
     )
 
diff --git a/src/batdetect2/train/train.py b/src/batdetect2/train/train.py
index 8845fa8..1f1745a 100644
--- a/src/batdetect2/train/train.py
+++ b/src/batdetect2/train/train.py
@@ -73,8 +73,9 @@ def run_train(
 
     audio_loader = audio_loader or build_audio_loader(config=audio_config)
 
-    # NOTE: Create a new preprocessor instead of using the one from the model
-    #       to avoid issues with device placement.
+    if model is not None:
+        preprocessor = preprocessor or model.preprocessor
+
     preprocessor = preprocessor or build_preprocessor(
         input_samplerate=audio_loader.samplerate,
         config=model_config.preprocess,