diff --git a/nn_fingerprint.py b/nn_fingerprint.py
index cef8c08..58b9205 100644
--- a/nn_fingerprint.py
+++ b/nn_fingerprint.py
@@ -60,7 +60,10 @@ class NNFingerprint(NNOperator):
             model_path = os.path.join(path, 'saved_model', 'pfann_fma_m.pt')
         if model_path.endswith('.onnx'):
             log.warning('Using onnx.')
-            self.model = onnxruntime.InferenceSession(model_path)
+            self.model = onnxruntime.InferenceSession(
+            model_path,
+            providers=['TensorrtExecutionProvider', 'CUDAExecutionProvider', 'CPUExecutionProvider']
+            )
         else:
             state_dict = torch.load(model_path, map_location=self.device)
             if isinstance(state_dict, torch.nn.Module):
diff --git a/test.py b/test.py
index b230402..87cd27d 100644
--- a/test.py
+++ b/test.py
@@ -39,7 +39,10 @@ op = ops.audio_embedding.nnfp()
 onnx_model = onnx.load('./saved/onnx/nnfp.onnx')
 onnx.checker.check_model(onnx_model)
 
-ort_session = onnxruntime.InferenceSession('./saved/onnx/nnfp.onnx')
+ort_session = onnxruntime.InferenceSession(
+'./saved/onnx/nnfp.onnx',
+providers=['TensorrtExecutionProvider', 'CUDAExecutionProvider', 'CPUExecutionProvider'])
+
 ort_inputs = {ort_session.get_inputs()[0].name: to_numpy(audio)}
 ort_outs = ort_session.run(None, ort_inputs)
 out3 = ort_outs[0]