diff --git a/asteroid/scripts/asteroid_cli.py b/asteroid/scripts/asteroid_cli.py
index bb6ee6c47..32ea2358c 100644
--- a/asteroid/scripts/asteroid_cli.py
+++ b/asteroid/scripts/asteroid_cli.py
@@ -1,5 +1,6 @@
 import os
 import argparse
+import torch
 import yaml
 import itertools
 import glob
@@ -87,7 +88,7 @@ def upload():
         )
 
 
-def infer():
+def infer(argv=None):
     """CLI function to run pretrained model inference on wav files."""
     parser = argparse.ArgumentParser()
     parser.add_argument("url_or_path", type=str, help="Path to the pretrained model.")
@@ -140,7 +141,20 @@ def infer():
     parser.add_argument(
         "-o", "--output-dir", default=None, type=str, help="Output directory to save files."
     )
-    args = parser.parse_args()
+    parser.add_argument(
+        "-d",
+        "--device",
+        default=None,
+        type=str,
+        help="Device to run the model on, eg. 'cuda:0'."
+        "Defaults to 'cuda' if CUDA is available, else 'cpu'.",
+    )
+    args = parser.parse_args(argv)
+
+    if args.device is None:
+        device = "cuda" if torch.cuda.is_available() else "cpu"
+    else:
+        device = args.device
 
     model = BaseModel.from_pretrained(pretrained_model_conf_or_path=args.url_or_path)
     if args.ola_window is not None:
@@ -152,8 +166,9 @@ def infer():
             window=args.ola_window_type,
             reorder_chunks=not args.ola_no_reorder,
         )
-    file_list = _process_files_as_list(args.files)
+    model = model.to(device)
 
+    file_list = _process_files_as_list(args.files)
     for f in file_list:
         separate(
             model,
diff --git a/tests/cli_test.py b/tests/cli_test.py
index 849069f08..3bfdeaaa9 100644
--- a/tests/cli_test.py
+++ b/tests/cli_test.py
@@ -1,4 +1,4 @@
-from asteroid.scripts import asteroid_versions
+from asteroid.scripts import asteroid_versions, asteroid_cli
 
 
 def test_asteroid_versions():
@@ -15,3 +15,31 @@ def test_print_versions():
 def test_asteroid_versions_without_git(monkeypatch):
     monkeypatch.setenv("PATH", "")
     asteroid_versions.asteroid_versions()
+
+
+def test_infer_device(monkeypatch):
+    """Test that inference is performed on the PyTorch device given by '--device'.
+
+    We can't properly test this in environments with only CPU device available.
+    As an approximation we test that the '.to()' method of the model is called
+    with the device given by '--device'.
+    """
+    # We can't use a real model to test this because calling .to() with a fake device
+    # on a real model will fail.
+    class FakeModel:
+        def to(self, device):
+            self.device = device
+
+    fake_model = FakeModel()
+
+    # Monkeypatch 'from_pretrained' to load our fake model.
+    from asteroid.models import BaseModel
+
+    monkeypatch.setattr(BaseModel, "from_pretrained", lambda *args, **kwargs: fake_model)
+
+    # Note that this will issue a warning about the missing file.
+    asteroid_cli.infer(
+        ["--device", "cuda:42", "somemodel", "--files", "file_that_does_not_exist.wav"]
+    )
+
+    assert fake_model.device == "cuda:42"