out_feat_dir = os.path.join(
work_dir, 'Data/inter-module/blzpilot/label_features/' +
str(decomposition_unit) + '_baseline_592/binary_label_' + str(out_dim))
if not os.path.exists(out_feat_dir):
os.makedirs(out_feat_dir)
feat_switch = {}
feat_switch['binary'] = 1
feat_switch['wordEmbed'] = 0
feat_switch['identity'] = 0
feat_switch['bottleneck'] = 0
feat_dir_path = {}
feat_dir_path['input_binary'] = binary_label_dir
feat_dir_path['input_labfile'] = label_align_dir
feat_dir_path['input_txt'] = text_dir
feat_dir_path['output_feat'] = out_feat_dir
DFP = 1
if DFP:
extract_base_feats = True
if extract_base_feats:
filelist = os.path.join(work_dir, 'Data/fileList/blzpilot.scp')
word_embed_file = os.path.join(
work_dir, 'Data/word_embeddings/turian-embeddings-50.txt')
list_of_files = io_funcs.load_file_list(filelist)
ip_feats.load_word_embeddings(word_embed_file)
ip_feats.extract_base_features(feat_dir_path, feat_switch,
list_of_files, decomposition_unit)
extractDur = False
perform_CTC_classes = False
analysisDur = False
denormDur = True
writeDur = False
modifyLabels = False
skip_pause = True
calcRMSE = True
if extractDur:
if not os.path.exists(out_dir):
os.makedirs(out_dir)
filelist = os.path.join(work_dir, 'Data/fileList/'+speaker+'.scp')
stat_fname = feat_dir_path + '.txt'
stats_file = os.path.join(work_dir, 'Data/inter-module/'+speaker+'/misc/', stat_fname)
list_of_files = io_funcs.load_file_list(filelist)
all_files_ph = []
dur_features = []; flens = []
for i in range(len(list_of_files)):
filename = list_of_files[i]
print filename
in_lab_file = os.path.join(label_align_dir, filename + '.lab')
[phone, st_arr, ph_arr, mean_f0_arr] = htsclass.read_state_align_label_file(in_lab_file)
flens.append((len(phone)-2)*out_dim)
for j in range(len(phone)):
if(phone[j] == '#'):
continue;
DFP = True
if DFP:
prosodydecomp = True
hierarcluster = True
templatefeats = False
analyseTrainData = False
prosodyrecons = False
computermse = False
analyseresults = False
if prosodydecomp:
filelist = os.path.join(work_dir,
'Data/fileList/' + speaker + '_train.scp')
list_arr = io_funcs.load_file_list(filelist)
prosody_feats = []
flens = []
dct_features_all_files = []
stat_features_all_files = []
for k in xrange(100, 500):
#if(k<1180):
# continue
filename = list_arr[k]
print filename
f0_file = os.path.join(f0_dir, filename + '.f0')
lab_file = os.path.join(lab_dir, filename + '.lab')
### processing lf0 file ###
# ori_f0_data, frame_number = io_funcs.load_binary_file_frame(f0_file, 1)
else:
in_feat_dir_path = in_unit + '_template_' + str(num_of_clusters)
out_feat_dir_path = out_unit + '_template_' + str(num_of_clusters)
lab_dir = os.path.join(work_dir, 'Data/inter-module/'+speaker+'/label_state_align/')
### Directory of files processing ###
DFP = True;
if DFP:
templatefeats = True;
if templatefeats:
filelist = os.path.join(work_dir, 'Data/fileList/'+speaker+'.scp')
list_arr = io_funcs.load_file_list(filelist)
in_template_dir = os.path.join(work_dir, 'Data/inter-module/'+speaker+'/template_features/' + in_feat_dir_path + '/')
out_template_dir = os.path.join(work_dir, 'Data/inter-module/'+speaker+'/template_features/' + out_feat_dir_path + '/')
if not os.path.exists(out_template_dir):
os.makedirs(out_template_dir)
removesilence=False
for k in range(1):
filename = list_arr[k]
print filename
in_file = os.path.join(in_template_dir, filename + '.cmp')
out_file = os.path.join(out_template_dir, filename + '.cmp')
in_lab_file = os.path.join(lab_dir, filename + '.lab')