diff --git a/pyproject.toml b/pyproject.toml
index e23e7a04..24698370 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -23,10 +23,10 @@ dependencies = [
 	"numpy>=1.18",
 	"onnxruntime; platform_system == 'Windows' and python_version < '3.11'",
 	"pretty_midi>=0.2.9",
-	"resampy>=0.2.2",
+	"resampy>=0.2.2,<0.4.3",
 	"scikit-learn",
 	"scipy>=1.4.1",
-	"tensorflow>=2.4.1; platform_system != 'Darwin' and python_version >= '3.11'",
+	"tensorflow>=2.4.1,<2.16; platform_system != 'Darwin' and python_version >= '3.11'",
 	"tensorflow-macos>=2.4.1; platform_system == 'Darwin' and python_version >= '3.11'",
 	"tflite-runtime; platform_system == 'Linux' and python_version < '3.11'",
 	"typing_extensions",
@@ -56,8 +56,8 @@ test = [
 	"pytest-mock",
 ]
 tf = [
-	"tensorflow>=2.4.1; platform_system != 'Darwin'",
-	"tensorflow-macos>=2.4.1; platform_system == 'Darwin' and python_version > '3.7'",
+	"tensorflow>=2.4.1,<2.16; platform_system != 'Darwin'",
+	"tensorflow-macos>=2.4.1,<2.16; platform_system == 'Darwin' and python_version > '3.7'",
 ]
 coreml = ["coremltools"] 
 onnx = ["onnxruntime"]
diff --git a/tests/resources/vocadito_10/model_output.npz b/tests/resources/vocadito_10/model_output.npz
index 8719f8a8..9b2a4045 100644
Binary files a/tests/resources/vocadito_10/model_output.npz and b/tests/resources/vocadito_10/model_output.npz differ
diff --git a/tests/resources/vocadito_10/note_events.npz b/tests/resources/vocadito_10/note_events.npz
index ffd860c5..a5e7d51b 100644
Binary files a/tests/resources/vocadito_10/note_events.npz and b/tests/resources/vocadito_10/note_events.npz differ
diff --git a/tests/test_inference.py b/tests/test_inference.py
index 1934239e..89665855 100644
--- a/tests/test_inference.py
+++ b/tests/test_inference.py
@@ -55,12 +55,11 @@ def test_predict() -> None:
     assert all(note_pitch_max)
     assert isinstance(note_events, list)
 
-    expected_model_output = np.load("tests/resources/vocadito_10/model_output.npz")
+    expected_model_output = np.load("tests/resources/vocadito_10/model_output.npz", allow_pickle=True)["arr_0"].item()
     for k in expected_model_output.keys():
         np.testing.assert_allclose(expected_model_output[k], model_output[k], atol=1e-4, rtol=0)
 
-    expected_note_events = np.load("tests/resources/vocadito_10/note_events.npz", allow_pickle=True)
-    expected_note_events = expected_note_events.get("arr_0")
+    expected_note_events = np.load("tests/resources/vocadito_10/note_events.npz", allow_pickle=True)["arr_0"]
 
     assert len(expected_note_events) == len(note_events)
     for expected, calculated in zip(expected_note_events, note_events):
diff --git a/tox.ini b/tox.ini
index 24a1df56..38a93806 100644
--- a/tox.ini
+++ b/tox.ini
@@ -11,7 +11,7 @@ requires =
 [testenv]
 deps = -e .[test]
 commands =
-    pytest tests --ignore tests/test_nn.py -s {posargs}
+    pytest tests --ignore tests/test_nn.py {posargs}
 setenv =
     SOURCE = {toxinidir}/basic_pitch
     TEST_SOURCE = {toxinidir}/tests