diff --git a/tensorflow/python/kernel_tests/signal/BUILD b/tensorflow/python/kernel_tests/signal/BUILD
index adb12a5e850..bd893184570 100644
--- a/tensorflow/python/kernel_tests/signal/BUILD
+++ b/tensorflow/python/kernel_tests/signal/BUILD
@@ -149,7 +149,6 @@ cuda_py_tests(
     python_version = "PY3",
     shard_count = 4,
     tags = [
-        "no_oss_py38",  #TODO(b/151631881)
         "no_windows_gpu",
     ],
     deps = [
diff --git a/tensorflow/python/kernel_tests/signal/test_util.py b/tensorflow/python/kernel_tests/signal/test_util.py
index 1e95fe4b28f..e8d477a843b 100644
--- a/tensorflow/python/kernel_tests/signal/test_util.py
+++ b/tensorflow/python/kernel_tests/signal/test_util.py
@@ -50,7 +50,7 @@ def grappler_optimize(graph, fetches=None, config_proto=None):
   return tf_optimizer.OptimizeGraph(config_proto, metagraph)
 
 
-def tflite_convert(fn, input_templates, use_mlir=False):
+def tflite_convert(fn, input_templates):
   """Converts the provided fn to tf.lite model.
 
   Args:
@@ -59,7 +59,6 @@ def tflite_convert(fn, input_templates, use_mlir=False):
     input_templates: A list of Tensors, ndarrays or TensorSpecs describing the
       inputs that fn expects. The actual values of the Tensors or ndarrays are
       unused.
-    use_mlir: Experimental. Whether to use the tf.lite MLIR converter.
 
   Returns:
     The serialized tf.lite model.
@@ -67,7 +66,6 @@ def tflite_convert(fn, input_templates, use_mlir=False):
   fn = def_function.function(fn)
   concrete_func = fn.get_concrete_function(*input_templates)
   converter = lite.TFLiteConverterV2([concrete_func])
-  converter.experimental_new_converter = use_mlir
   return converter.convert()
 
 
diff --git a/tensorflow/python/kernel_tests/signal/window_ops_test.py b/tensorflow/python/kernel_tests/signal/window_ops_test.py
index 9f5fe6f64c7..9432e70c7f2 100644
--- a/tensorflow/python/kernel_tests/signal/window_ops_test.py
+++ b/tensorflow/python/kernel_tests/signal/window_ops_test.py
@@ -156,15 +156,14 @@ class WindowOpsTest(test.TestCase, parameterized.TestCase):
       self.assertLen(rewritten_graph.node, 1)
 
   @parameterized.parameters(
-      # Due to control flow, only MLIR is supported.
       # Only float32 is supported.
-      (window_ops.hann_window, 10, False, dtypes.float32, True),
-      (window_ops.hann_window, 10, True, dtypes.float32, True),
-      (window_ops.hamming_window, 10, False, dtypes.float32, True),
-      (window_ops.hamming_window, 10, True, dtypes.float32, True),
-      (window_ops.vorbis_window, 12, None, dtypes.float32, True))
-  def test_tflite_convert(self, window_fn, window_length, periodic, dtype,
-                          use_mlir):
+      (window_ops.hann_window, 10, False, dtypes.float32),
+      (window_ops.hann_window, 10, True, dtypes.float32),
+      (window_ops.hamming_window, 10, False, dtypes.float32),
+      (window_ops.hamming_window, 10, True, dtypes.float32),
+      (window_ops.vorbis_window, 12, None, dtypes.float32))
+  def test_tflite_convert(self, window_fn, window_length, periodic, dtype):
+
     def fn(window_length):
       try:
         return window_fn(window_length, periodic=periodic, dtype=dtype)
@@ -172,7 +171,7 @@ class WindowOpsTest(test.TestCase, parameterized.TestCase):
         return window_fn(window_length, dtype=dtype)
 
     tflite_model = test_util.tflite_convert(
-        fn, [tensor_spec.TensorSpec(shape=[], dtype=dtypes.int32)], use_mlir)
+        fn, [tensor_spec.TensorSpec(shape=[], dtype=dtypes.int32)])
     window_length = np.array(window_length).astype(np.int32)
     actual_output, = test_util.evaluate_tflite_model(
         tflite_model, [window_length])