def main():
args = check_argv()
pairs_fn = args.pairs_fn
features_fn = args.features_fn
distances_fn = args.distances_fn
normalize_feats = args.normalize_feats
only = args.only
if args.metric == "cosine":
dtw_cost_func = _dtw.multivariate_dtw_cost_cosine
elif args.metric == "euclidean":
dtw_cost_func = _dtw.multivariate_dtw_cost_euclidean
# normalize_feats = False
elif args.metric == "euclidean_squared":
dtw_cost_func = _dtw.multivariate_dtw_cost_euclidean_squared
# Read the pairs and the archive
print "Start time: " + str(datetime.datetime.now())
print "Reading pairs from:", pairs_fn
pairs = read_pairs(pairs_fn, only)
print "Reading features from:", features_fn
if args.input_fmt == "kaldi_txt":
ark = read_kaldi_ark(features_fn)
elif args.input_fmt == "npz":
ark = np.load(features_fn)
ark = dict(ark)
# sys.stdout.flush()
# Normalize features per frame
if normalize_feats:
print "Normalizing features"
for utt_id in ark:
N = ark[utt_id].shape[0]
for i in range(N):
ark[utt_id][i, :] = ark[utt_id][i, :] / np.linalg.norm(
ark[utt_id][i, :])
# Calculate distances
print "Calculating distances"
costs = np.zeros(len(pairs))
for i_pair, pair in enumerate(pairs):
utt_id_1, utt_id_2 = pair
costs[i_pair] = dtw_cost_func(np.array(ark[utt_id_1], dtype=np.double),
np.array(ark[utt_id_2], dtype=np.double),
True)
# Write to file
if args.binary_dists:
print "Writing distances to binary file:", distances_fn
np.asarray(costs, dtype=np.float32).tofile(distances_fn)
else:
print "Writing distances to text file:", distances_fn
np.asarray(costs, dtype=np.float32).tofile(distances_fn, "\n")
open(distances_fn, "a").write("\n") # add final newline
print "End time: " + str(datetime.datetime.now())
def main():
args = check_argv()
pairs_fn = args.pairs_fn
features_fn = args.features_fn
distances_fn = args.distances_fn
normalize_feats = args.normalize_feats
if args.metric == "cosine":
dtw_cost_func = _dtw.multivariate_dtw_cost_cosine
elif args.metric == "euclidean":
dtw_cost_func = _dtw.multivariate_dtw_cost_euclidean
# normalize_feats = False
elif args.metric == "euclidean_squared":
dtw_cost_func = _dtw.multivariate_dtw_cost_euclidean_squared
# Read the pairs and the archive
print "Start time: " + str(datetime.datetime.now())
print "Reading pairs from:", pairs_fn
pairs = read_pairs(pairs_fn)
print "Reading features from:", features_fn
if args.input_fmt == "kaldi_txt":
ark = read_kaldi_ark(features_fn)
elif args.input_fmt == "npz":
ark = np.load(features_fn)
ark = dict(ark)
# sys.stdout.flush()
# Normalize features per frame
if normalize_feats:
print "Normalizing features"
for utt_id in ark:
N = ark[utt_id].shape[0]
for i in range(N):
ark[utt_id][i, :] = ark[utt_id][i, :]/np.linalg.norm(ark[utt_id][i, :])
# Calculate distances
print "Calculating distances"
costs = np.zeros(len(pairs))
for i_pair, pair in enumerate(pairs):
utt_id_1, utt_id_2 = pair
costs[i_pair] = dtw_cost_func(
np.array(ark[utt_id_1], dtype=np.double), np.array(ark[utt_id_2], dtype=np.double), True
)
# Write to file
if args.binary_dists:
print "Writing distances to binary file:", distances_fn
np.asarray(costs, dtype=np.float32).tofile(distances_fn)
else:
print "Writing distances to text file:", distances_fn
np.asarray(costs, dtype=np.float32).tofile(distances_fn, "\n")
open(distances_fn, "a").write("\n") # add final newline
print "End time: " + str(datetime.datetime.now())
def main():
args = check_argv()
print str(datetime.datetime.now())
print "Reading Kaldi archive:", args.kaldi_ark_fn
kaldi_ark = read_kaldi_ark(args.kaldi_ark_fn)
print "Number of keys in archive:", len(kaldi_ark.keys())
# all_features = np.asarray(np.concatenate(kaldi_ark.values(), axis=0), dtype=np.float32)
# print "Number of feature vectors:", all_features.shape[0]
# print "Feature vector dimensions:", all_features.shape[1]
print "Writing feature vectors to file:", args.npz_fn
np.savez(args.npz_fn, **kaldi_ark)
print str(datetime.datetime.now())
def main():
args = check_argv()
print str(datetime.datetime.now())
print "Reading Kaldi archive:", args.kaldi_ark_fn
kaldi_ark = read_kaldi_ark(args.kaldi_ark_fn)
print "Number of keys in archive:", len(kaldi_ark.keys())
# all_features = np.asarray(np.concatenate(kaldi_ark.values(), axis=0), dtype=np.float32)
# print "Number of feature vectors:", all_features.shape[0]
# print "Feature vector dimensions:", all_features.shape[1]
print "Writing feature vectors to file:", args.npz_fn
np.savez(args.npz_fn, **kaldi_ark)
print str(datetime.datetime.now())
def main():
args = check_argv()
pairs_fn = args.pairs_fn
features_fn = args.features_fn
paths_pkl_fn = args.paths_pkl_fn
normalize_feats = args.normalize_feats
n_paths_notify = 100
# if args.metric == "cosine":
# dtw_cost_func = _dtw.multivariate_dtw_cost_cosine
# elif args.metric == "euclidean":
# dtw_cost_func = _dtw.multivariate_dtw_cost_euclidean
# normalize_feats = False
# Read the pairs and the archive
print "Start time: " + str(datetime.datetime.now())
print "Reading pairs from:", pairs_fn
pairs = read_pairs(pairs_fn)
print "Reading features from:", features_fn
if args.input_fmt == "kaldi_txt":
ark = read_kaldi_ark(features_fn)
elif args.input_fmt == "npz":
ark = np.load(features_fn)
ark = dict(ark)
print "Number of entries loaded:", len(ark)
# sys.stdout.flush()
# Normalize features per frame
if normalize_feats:
print "Normalizing features"
for utt_id in ark:
N = ark[utt_id].shape[0]
for i in range(N):
ark[utt_id][i, :] = ark[utt_id][i, :]/np.linalg.norm(ark[utt_id][i, :])
# Calculate distances
print "Calculating paths: ",
paths = []
i_paths = 0
for i_pair, pair in enumerate(pairs):
utt_id_1, utt_id_2 = pair
paths.append(_dtw.multivariate_dtw(
np.array(ark[utt_id_1], dtype=np.double), np.array(ark[utt_id_2], dtype=np.double)
)[0])
i_paths += 1
if i_paths % n_paths_notify == 0:
sys.stdout.write('.')
print
# costs = np.zeros(len(pairs))
# for i_pair, pair in enumerate(pairs):
# utt_id_1, utt_id_2 = pair
# costs[i_pair] = dtw_cost_func(ark[utt_id_1], ark[utt_id_2], True)
# Write to file
print "Pickling paths:", paths_pkl_fn
f = open(paths_pkl_fn, "wb")
cPickle.dump(paths, f, -1)
f.close()
# if args.binary_dists:
# print "Writing distances to binary file:", distances_fn
# np.asarray(costs, dtype=np.float32).tofile(distances_fn)
# else:
# print "Writing distances to text file:", distances_fn
# np.asarray(costs, dtype=np.float32).tofile(distances_fn, "\n")
# open(distances_fn, "a").write("\n") # add final newline
print "End time: " + str(datetime.datetime.now())
def main():
args = check_argv()
pairs_fn = args.pairs_fn
features_fn = args.features_fn
paths_pkl_fn = args.paths_pkl_fn
normalize_feats = args.normalize_feats
n_paths_notify = 100
# if args.metric == "cosine":
# dtw_cost_func = _dtw.multivariate_dtw_cost_cosine
# elif args.metric == "euclidean":
# dtw_cost_func = _dtw.multivariate_dtw_cost_euclidean
# normalize_feats = False
# Read the pairs and the archive
print("Start time: " + str(datetime.datetime.now()))
print("Reading pairs from:", pairs_fn)
pairs = read_pairs(pairs_fn)
print("Reading features from:", features_fn)
if args.input_fmt == "kaldi_txt":
ark = read_kaldi_ark(features_fn)
elif args.input_fmt == "npz":
ark = np.load(features_fn)
ark = dict(ark)
print("Number of entries loaded:", len(ark))
# sys.stdout.flush()
# Normalize features per frame
if normalize_feats:
print("Normalizing features")
for utt_id in ark:
N = ark[utt_id].shape[0]
for i in range(N):
ark[utt_id][i, :] = ark[utt_id][i, :] / np.linalg.norm(
ark[utt_id][i, :])
# Calculate distances
print("Calculating paths: ", )
paths = []
i_paths = 0
for i_pair, pair in enumerate(pairs):
utt_id_1, utt_id_2 = pair
paths.append(
_dtw.multivariate_dtw(np.array(ark[utt_id_1], dtype=np.double),
np.array(ark[utt_id_2], dtype=np.double))[0])
i_paths += 1
if i_paths % n_paths_notify == 0:
sys.stdout.write('.')
print
# costs = np.zeros(len(pairs))
# for i_pair, pair in enumerate(pairs):
# utt_id_1, utt_id_2 = pair
# costs[i_pair] = dtw_cost_func(ark[utt_id_1], ark[utt_id_2], True)
# Write to file
print("Pickling paths:", paths_pkl_fn)
f = open(paths_pkl_fn, "wb")
cPickle.dump(paths, f, -1)
f.close()
# if args.binary_dists:
# print("Writing distances to binary file:", distances_fn)
# np.asarray(costs, dtype=np.float32).tofile(distances_fn)
# else:
# print("Writing distances to text file:", distances_fn)
# np.asarray(costs, dtype=np.float32).tofile(distances_fn, "\n")
# open(distances_fn, "a").write("\n") # add final newline
print("End time: " + str(datetime.datetime.now()))