def calculate(cv_path, gt_file_list, conf, spkr_conf):
basename = cv_path.stem
number, orgspk, tarspk = basename.split("_")
tarspk = tarspk.split("-")[-1]
orgspk = orgspk.split("-")[-1]
# get converted features. If mcep, from h5; else waveform
if conf["feat_type"] == "mcep":
cv_mcep = read_feature(cv_path, "feat")
cv_f0 = read_feature(cv_path, "f0")
else:
cv_mcep, cv_f0 = get_world_features(cv_path, tarspk, conf, spkr_conf)
# get ground truth features
gt_mcep = read_feature(gt_file_list[f"{tarspk}_{number}"], "mcep")
gt_f0 = read_feature(gt_file_list[f"{tarspk}_{number}"], "f0")
# non-silence parts
gt_idx = np.where(gt_f0 > 0)[0]
gt_mcep = gt_mcep[gt_idx]
cv_idx = np.where(cv_f0 > 0)[0]
cv_mcep = cv_mcep[cv_idx]
# DTW
_, path = fastdtw(cv_mcep, gt_mcep, dist=scipy.spatial.distance.euclidean)
twf = np.array(path).T
cv_mcep_dtw = cv_mcep[twf[0]]
gt_mcep_dtw = gt_mcep[twf[1]]
# MCD
diff2sum = np.sum((cv_mcep_dtw - gt_mcep_dtw)**2, 1)
mcd = np.mean(10.0 / np.log(10.0) * np.sqrt(2 * diff2sum), 0)
return f"{orgspk}-{tarspk}-{number}", mcd
def main():
parser = argparse.ArgumentParser(
description=
"Convert filter banks to waveform using Griffin-Lim algorithm",
formatter_class=argparse.ArgumentDefaultsHelpFormatter,
)
parser.add_argument("--conf",
type=str,
required=True,
help="Cofiguration file")
parser.add_argument(
"--rootdir",
type=str,
required=True,
help="Root directory of filter bank h5 files",
)
parser.add_argument("--outdir",
type=str,
required=True,
help="Output directory")
args = parser.parse_args()
# logging info
logging.basicConfig(
level=logging.INFO,
stream=sys.stdout,
format="%(asctime)s (%(module)s:%(lineno)d) "
"%(levelname)s: %(message)s",
)
# load configure files
conf = load_yaml(args.conf)
for k, v in conf.items():
logging.info("{}: {}".format(k, v))
# find h5 files
feats_files = sorted(list(Path(args.rootdir).glob("*.h5")))
feats = {
Path(args.outdir) / filename.stem + ".wav":
read_feature(filename, "feats")
for filename in feats_files
}
# Main Griffin-Lim algorithm
Parallel(n_jobs=30)([
delayed(mlfb2wavf)(
feats[wavf],
wavf,
fs=conf["feature"]["fs"],
n_mels=conf["feature"]["mlfb_dim"],
fftl=conf["feature"]["fftl"],
hop_size=conf["feature"]["hop_size"],
plot=False,
) for wavf in list(feats.keys())
])