'_' + lang + '.pkl', 'wb'))
pickle.dump(kmeans, open(str(num_phones) + '_' + lang + \
'/kmeans_' + str(num_phones) + \
'_' + lang + '.pkl', 'wb'))
all_features = {}
# get bottleneck features of all .wav files (stimuli)
for root, dirs, files in os.walk(WAV_FOLDER):
for wav_file in files:
if wav_file.endswith(".wav"):
audio = Audio.load(root + wav_file)
all_features[wav_file] = audio
processor = BottleneckProcessor(weights='BabelMulti')
corpus_features = processor.process_all(all_features)
open_feats = []
for key in corpus_features:
# access every features object
feats = corpus_features[key].data
# put them all together
open_feats.append(feats)
unlisted_feats = np.asarray(open_feats)
#flattened_feats = np.concatenate(all_corpus_features, axis=0)
flattened_feats = np.concatenate(unlisted_feats)
# idx = (3,6,9)
# flattened_feats = np.insert(flattened_feats, idx, 0, axis=1)
print flattened_feats.shape
for NUM_PHONES in list(NUM_PHONES_LIST):
# clustered = kmeans_corpus.fit_predict(stim_pqcode)
np.save(to_folder + utterance_n, softmax_stim)
#dist_df = pd.DataFrame(softmax_stim)
#dist_df.to_csv(to_folder + utterance_n + '.csv')
all_features = {}
# get bottleneck features of all .wav files (stimuli)
for root, dirs, files in os.walk(WAV_FOLDER):
for wav_file in files:
if wav_file.endswith(".wav"):
audio = Audio.load(root + wav_file)
all_features[wav_file] = audio
processor = BottleneckProcessor(weights='BabelMulti')
stim_features = processor.process_all(all_features)
# dict_feats = {}
# for key in stim_features:
# # access every features object
# feats = stim_features[key].data
# # put them all together
# dict_feats[key] = feats
for root_p, dirs_p, files_ in pickles_folder:
for dir_p in dirs_p:
if dir_p == 'softmax_dist':
pass
else:
PATH = root_p + dir_p
NUM = dir_p.split('_')[0]