diff --git a/src/python/larynx_train/__main__.py b/src/python/larynx_train/__main__.py
index f6153e4..e27e5e9 100644
--- a/src/python/larynx_train/__main__.py
+++ b/src/python/larynx_train/__main__.py
@@ -5,6 +5,7 @@ from pathlib import Path
 
 import torch
 from pytorch_lightning import Trainer
+from pytorch_lightning.callbacks import ModelCheckpoint
 
 from .vits.lightning import VitsModel
 
@@ -18,6 +19,11 @@ def main():
     parser.add_argument(
         "--dataset-dir", required=True, help="Path to pre-processed dataset directory"
     )
+    parser.add_argument(
+        "--checkpoint-epochs",
+        type=int,
+        help="Save checkpoint every N epochs (default: 1)",
+    )
     Trainer.add_argparse_args(parser)
     VitsModel.add_model_specific_args(parser)
     parser.add_argument("--seed", type=int, default=1234)
@@ -42,6 +48,12 @@ def main():
         sample_rate = int(config["audio"]["sample_rate"])
 
     trainer = Trainer.from_argparse_args(args)
+    if args.checkpoint_epochs is not None:
+        trainer.callbacks = [ModelCheckpoint(every_n_epochs=args.checkpoint_epochs)]
+        _LOGGER.debug(
+            "Checkpoints will be saved every %s epoch(s)", args.checkpoint_epochs
+        )
+
     dict_args = vars(args)
     model = VitsModel(
         num_symbols=num_symbols,