def test_save_load(self):
dir_out = self._get_test_dir()
theta_start = 0.01
theta_stop = 0.055
theta_step = 0.005
thetas = np.arange(theta_start, theta_stop, theta_step)
k = 6
frame_size_ms = 5
atom_gen = AtomVUVDistPosLabelGen(self.dir_wcad_root, dir_out,
dir_out, thetas, frame_size_ms)
label_dict = atom_gen.gen_data(self.dir_wav, None,
id_list=self.id_list, return_dict=True)
# Call this once before starting the pre-processing.
atom_gen.get_normalisation_params(dir_out)
test_label = label_dict[self.id_list[1]]
test_label_pre = atom_gen.preprocess_sample(test_label)
# TODO: Figure out what to test here.
# self.assertTrue(np.isclose(test_label_pre,
# atom_gen[self.id_list[1]]).all())
test_label_post = atom_gen.postprocess_sample(test_label_pre)
# self.assertTrue(np.isclose(test_label, test_label_post).all())
self.assertTrue(np.isclose(-3.4898, test_label_post.sum(),
atol=0.0001))
os.remove(os.path.join(self.dir_database, "wcad_.txt"))
shutil.rmtree(dir_out)
sp_type="mcep")
world_generator.gen_data(dir_wav, dir_world, id_list=id_list)
world_generator = WorldFeatLabelGen(dir_world,
add_deltas=True,
num_coded_sps=20,
sp_type="mcep")
world_generator.gen_data(dir_wav, dir_world, id_list=id_list)
# Generate atoms.
logging.info("Create atom files.")
shutil.rmtree(dir_atoms)
makedirs_safe(dir_atoms)
atom_generator = AtomVUVDistPosLabelGen(
os.path.join(os.path.dirname(os.environ["IDIAPTTS_ROOT"]), "tools",
"wcad"), dir_atoms, dir_world, thetas)
atom_generator.gen_data(dir_wav, dir_atoms, id_list=id_list)
if retrain_models:
raise NotImplementedError("Did not yet implemented retraining of models.")
elif save_models:
from idiaptts.src.neural_networks.pytorch.ModelHandlerPyTorch import ModelHandlerPyTorch
from idiaptts.src.model_trainers.wcad.AtomVUVDistPosModelTrainer import AtomVUVDistPosModelTrainer
hparams = AtomVUVDistPosModelTrainer.create_hparams()
hparams.model_name = "test_model_in409_out7.nn"
model_handler = ModelHandlerPyTorch()
# The following code uses the load_model method and saves it back as a checkpoint.
# model, model_type, dim_in, dim_out = model_handler.load_model(hparams.model_name, hparams)
# model_handler.model_type = "RNNDYN-1_RELU_32-1_FC_7"
# model_handler.dim_in = model.dim_in
# model_handler.dim_out = model.dim_out