diff --git a/paddlespeech/t2s/exps/inference.py b/paddlespeech/t2s/exps/inference.py
index 8a526982..21d105ad 100644
--- a/paddlespeech/t2s/exps/inference.py
+++ b/paddlespeech/t2s/exps/inference.py
@@ -112,7 +112,7 @@ def parse_args():
     parser.add_argument(
         "--device",
         default="gpu",
-        choices=["gpu", "cpu", "xpu"],
+        choices=["gpu", "cpu", "xpu", "npu"],
         help="Device selected for inference.", )
     parser.add_argument('--cpu_threads', type=int, default=1)
 
diff --git a/paddlespeech/t2s/exps/syn_utils.py b/paddlespeech/t2s/exps/syn_utils.py
index 9a07df64..75b98f1e 100644
--- a/paddlespeech/t2s/exps/syn_utils.py
+++ b/paddlespeech/t2s/exps/syn_utils.py
@@ -591,7 +591,8 @@ def get_predictor(
 
     config = inference.Config(
         str(Path(model_dir) / model_file), str(Path(model_dir) / params_file))
-    config.enable_memory_optim()
+    if device != "npu":
+        config.enable_memory_optim()
     config.switch_ir_optim(True)
     if device == "gpu":
         config.enable_use_gpu(100, device_id)