def _sample_to_features(x, frontend_args, tflite):
if frontend_args is None:
frontend_args = {}
return tf_frontend.compute_frontend_features(x,
16000,
tflite=tflite,
**frontend_args)
def _sample_to_features(x):
return tf_frontend.compute_frontend_features(x, 16000, overlap_seconds=79)
def test_multibatch_sanity(self):
for shape in [[16000], [1, 16000], [3, 20000]]:
tf_frontend.compute_frontend_features(tf.zeros(shape, tf.float32),
sr=16000,
frame_hop=5)
def _sample_to_features(x, export_tflite=False):
return tf_frontend.compute_frontend_features(x,
16000,
overlap_seconds=79,
tflite=export_tflite)
def _feature_fn(x, s):
return tf.expand_dims(tf_frontend.compute_frontend_features(
x, s, overlap_seconds=79),
axis=-1).numpy().astype(np.float32)
def default_feature_fn(samples, sample_rate):
frontend_args = tf_frontend.frontend_args_from_flags()
feats = tf_frontend.compute_frontend_features(samples, sample_rate,
**frontend_args)
logging.info('Feats shape: %s', feats.shape)
return tf.expand_dims(feats, axis=-1).numpy().astype(np.float32)
def _default_feature_fn(samples, sample_rate):
return tf.expand_dims(tf_frontend.compute_frontend_features(
samples, sample_rate, overlap_seconds=79),
axis=-1).numpy().astype(np.float32)
def _sample_to_features(x, export_tflite=False):
frontend_args = tf_frontend.frontend_args_from_flags()
return tf_frontend.compute_frontend_features(x,
16000,
tflite=export_tflite,
**frontend_args)
def _default_feature_fn(samples, sample_rate):
frontend_args = tf_frontend.frontend_args_from_flags()
feats = tf_frontend.compute_frontend_features(samples, sample_rate,
**frontend_args)
return tf.expand_dims(feats, axis=-1).numpy().astype(np.float32)
def _feature_fn(x, s):
return tf.expand_dims(tf_frontend.compute_frontend_features(
x, s, frame_hop=17),
axis=-1).numpy().astype(np.float32)
