def dump_pseudo_label_mfcc(km_path, task, sample_rate, nj):
apply_kmeans = ApplyKmeans(km_path)
reader = MfccFeatureReader(sample_rate)
generator, num = get_path_iterator(f"{task}/wav.scp", 1.0)
iterator = generator()
if nj > 1:
feats = joblib.Parallel(n_jobs=nj)(
joblib.delayed(reader.get_feats)(path)
for utt_id, path in
tqdm.tqdm(iterator, total=num))
p_labs = joblib.Parallel(n_jobs=nj)(
joblib.delayed(apply_kmeans)(feat)
for feat in
tqdm.tqdm(feats, total=num))
iterator = generator()
utt_ids = [utt_id for utt_id, _ in iterator]
else:
utt_ids, p_labs = [], []
for utt_id, path in tqdm.tqdm(iterator, total=num):
feat = reader.get_feats(path)
p_lab = apply_kmeans(feat).tolist()
p_labs.append(p_lab)
utt_ids.append(utt_id)
return utt_ids, p_labs
def dump_pseudo_label_hubert(km_path, task, sample_rate, url, dir, layer):
apply_kmeans = ApplyKmeans(km_path)
reader = HubertFeatureReader(sample_rate, url, dir, layer)
generator, num = get_path_iterator(f"{task}/wav.scp", 1.0)
iterator = generator()
utt_ids, p_labs = [], []
for utt_id, path in tqdm.tqdm(iterator, total=num):
feat = reader.get_feats(path)
p_lab = apply_kmeans(feat).tolist()
p_labs.append(p_lab)
utt_ids.append(utt_id)
return utt_ids, p_labs