def test_tflite_inference(self, feature_inputs):
test_dir = 'non_semantic_speech_benchmark/data_prep/testdata'
if feature_inputs:
test_file = 'model1_woutfrontend.tflite'
else:
test_file = 'model1_wfrontend.tflite'
tflite_model_path = os.path.join(absltest.get_default_test_srcdir(),
test_dir, test_file)
output_key = '0'
interpreter = audio_to_embeddings_beam_utils._build_tflite_interpreter(
tflite_model_path=tflite_model_path)
model_input = np.zeros([32000], dtype=np.float32)
sample_rate = 16000
if feature_inputs:
model_input = audio_to_embeddings_beam_utils._default_feature_fn(
model_input, sample_rate)
audio_to_embeddings_beam_utils._samples_to_embedding_tflite(
model_input, sample_rate, interpreter, output_key)
def test_tflite_inference(self, feature_inputs):
if feature_inputs:
test_file = 'model1_woutfrontend.tflite'
else:
test_file = 'model1_wfrontend.tflite'
tflite_model_path = os.path.join(absltest.get_default_test_srcdir(),
TEST_DIR, test_file)
output_key = '0'
interpreter = audio_to_embeddings_beam_utils.build_tflite_interpreter(
tflite_model_path=tflite_model_path)
model_input = np.zeros([32000], dtype=np.float32)
sample_rate = 16000
if feature_inputs:
model_input = audio_to_embeddings_beam_utils._default_feature_fn(
model_input, sample_rate)
else:
model_input = np.expand_dims(model_input, axis=0)
audio_to_embeddings_beam_utils.samples_to_embedding_tflite(
model_input, sample_rate, interpreter, output_key, 'name')