def extract_single_acoustic(in_path, out_path):
"""
Extract mfcc features from one audio file
:param in_path: the path to the audio file
:param out_path: the path to save the mfcc features
"""
tmp_input = "tmp.input"
tmp_features = "tmp.features"
tmp_label = "tmp.labels"
zero = 0.01
input_file = open(tmp_dir + tmp_input, 'wb') # open the input file for the feature extraction
features_file = open(tmp_dir + tmp_features, 'wb') # open file for the feature list path
labels_file = open(tmp_dir + tmp_label, 'wb') # open file for the labels
length = utils.get_wav_file_length(in_path)
# write the data
input_file.write(
'"' + in_path + '" ' + str('%.8f' % 0) + ' ' + str(float(length) - zero) + ' ' + str(
'%.8f' % 0) + ' ' + str(
'%.8f' % 0))
features_file.write(out_path)
input_file.close()
features_file.close()
labels_file.close()
command = "sbin/fea_extract %s %s %s" % (tmp_dir + tmp_input, tmp_dir + tmp_features, tmp_dir + tmp_label)
utils.easy_call(command)
# remove leftovers
os.remove(tmp_dir + tmp_input)
os.remove(tmp_dir + tmp_features)
os.remove(tmp_dir + tmp_label)
def run(features_path, output_path):
f_abs_path = os.path.abspath(features_path)
o_abs_path = os.path.abspath(output_path)
os.chdir("lua_scripts/")
cmd = 'th classify_multi_class.lua -input_file %s -output_file %s' % (
f_abs_path, o_abs_path)
utils.easy_call(cmd)
os.chdir("..")
def extract_features(wav_filename, output_path, start_extract, end_extract):
# defines
temp_input_filename = utils.generate_tmp_filename('input')
temp_label_filename = utils.generate_tmp_filename('labels')
temp_features_filename = utils.generate_tmp_filename('features')
temp_wav16_filename = utils.generate_tmp_filename('wav')
# validation
if not os.path.exists(wav_filename):
print >> sys.stderr, 'Error: input path %s does not exists.' % wav_filename
return
# loop over all the files in the input dir
if wav_filename.endswith('.wav'):
try:
# convert to 16K 16bit
cmd = 'sox %s -r 16000 -b 16 %s' % (wav_filename, temp_wav16_filename)
utils.easy_call(cmd)
onset = (float(start_extract) + float(end_extract)) / 2
offset = (float(start_extract) + float(end_extract)) / 2
# =================== ACOUSTIC FEATURES =================== #
# # write labels
# label_file = wav_filename.replace('.wav', label_suffix)
# fid = open(label_file, 'w')
# fid.write('1 2\n')
# fid.write('%s %s %s\n' % (str(1), str(1), str(1)))
# fid.close()
# creating the files
input_file = open(temp_features_filename, 'wb') # open the input file for the feature extraction
features_file = open(temp_input_filename, 'wb') # open file for the feature list path
labels_file = open(temp_label_filename, 'wb') # open file for the labels
# write the data
input_file.write(
'"' + temp_wav16_filename + '" ' + str('%.8f' % float(start_extract)) + ' ' + str(
float(end_extract)) + ' ' + str(
'%.8f' % float(onset)) + ' ' + str('%.8f' % float(offset)))
features_file.write(output_path.replace('.wav', '.txt'))
input_file.close()
features_file.close()
labels_file.close()
command = "./sbin/VotFrontEnd2 %s %s %s" % (input_file.name, features_file.name, labels_file.name)
utils.easy_call(command)
# remove leftovers
os.remove(temp_input_filename)
os.remove(temp_label_filename)
os.remove(temp_features_filename)
os.remove(temp_wav16_filename)
except:
print wav_filename
def predict(input_path, output_path, model, csv_filename):
if not os.path.exists(input_path):
print >> sys.stderr, "%s file does not exits" % input_path
return
t_model = model.upper()
if t_model == 'RNN':
model_path = 'results/1_layer_model.net'
print '==> using single layer RNN'
elif t_model == '2RNN':
model_path = 'results/2_layer_model.net'
print '==> using 2 stacked layers RNN'
elif t_model == 'BIRNN':
model_path = 'results/bi_model.net'
print '==> using bi-directional RNN'
else:
model_path = 'results/1_layer_model.net'
print '==> unknown model, using default model: single RNN'
try:
length = utils.get_wav_file_length(input_path)
except:
print "The input file ", input_path, " is probably not a valid WAV file."
exit(-1)
feature_file = generate_tmp_filename('features')
prob_file = generate_tmp_filename('prob')
predict_file = generate_tmp_filename('prediction')
dur_file = generate_tmp_filename('dur')
print '\n1) Extracting features and classifying ...'
abs_path = os.path.abspath(input_path)
os.chdir("front_end/")
fe.main(abs_path, feature_file)
os.chdir("..")
print '\n2) Model predictions ...'
cmd = 'th classify.lua -x_filename %s -class_path %s -prob_path %s -model_path %s' % (
feature_file, predict_file, prob_file, model_path)
os.chdir("back_end/")
utils.easy_call(cmd)
os.chdir("..")
print '\n3) Extracting duration'
post_process(os.path.abspath(predict_file), dur_file)
print '\n4) Writing TextGrid file to %s ...' % output_path
create_text_grid(dur_file, input_path, output_path, length, float(0.0), csv_filename)
# remove leftovers
os.remove(feature_file)
os.remove(prob_file)
os.remove(predict_file)
os.remove(dur_file)
def extract_single_mfcc(in_path, out_path):
"""
Extract mfcc features from one audio file
:param in_path: the path to the audio file
:param out_path: the path to save the mfcc features
"""
import platform
plat = platform.system().lower()
if plat is 'darwin':
sox_path = 'sbin/osx/sox'
htk_path = 'sbin/osx'
elif 'linux' in plat:
sox_path = 'sox'
htk_path = 'sbin/linux'
else:
sox_path = 'sbin/osx/sox'
htk_path = 'sbin/osx'
tmp_file = utils.generate_tmp_filename('wav')
cmd = "%s %s -r 16000 -b 16 %s" % (sox_path, in_path, tmp_file)
utils.easy_call(cmd)
cmd = "%s/HCopy -C config/htk.config %s %s" % (htk_path, tmp_file, out_path)
utils.easy_call(cmd)
os.remove(tmp_file)
def predict(input_path, output_path, model):
tmp_dir = 'tmp/'
tmp_features = 'tmp.features'
tmp_prob = 'tmp.prob'
tmp_prediction = 'tmp.prediction'
tmp_duration = 'tmp.dur'
if not os.path.exists(input_path):
print >> sys.stderr, "wav file does not exits"
return
t_model = model.upper()
if t_model == 'RNN':
model_path = 'results/1_layer_model.net'
print '==> using single RNN layer'
elif t_model == '2RNN':
model_path = 'results/2_layer_model.net'
print '==> using 2 stacked layers of RNN'
elif t_model == 'BIRNN':
model_path = 'results/1_bi_model.net'
print '==> using single bi-directional RNN layer'
elif t_model == '2BIRNN':
model_path = 'results/2_bi_model.net'
print '==> using two stacked layers of bi-directional RNN'
else:
model_path = 'results/1_layer_model.net'
print '==> unknown model, using default model: single layer of RNN'
length = utils.get_wav_file_length(input_path)
prob_file = tmp_dir + tmp_prob
predict_file = tmp_dir + tmp_prediction
dur_file = tmp_dir+tmp_duration
# remove tmo dir if exists
if os.path.exists(tmp_dir):
shutil.rmtree(tmp_dir)
os.mkdir(tmp_dir)
print '\n1) Extracting features and classifying ...'
cmd = 'python predict_single_file.py %s %s ' % (
os.path.abspath(os.path.abspath(input_path)), os.path.abspath(tmp_dir) + '/' + tmp_features)
os.chdir("front_end/")
utils.easy_call(cmd)
os.chdir("..")
print '\n2) Model predictions ...'
cmd = 'th classify.lua -folder_path %s -x_filename %s -class_path %s -prob_path %s -model_path %s' % (
os.path.abspath(tmp_dir), tmp_features, os.path.abspath(predict_file), os.path.abspath(prob_file), model_path)
os.chdir("back_end/")
utils.easy_call(cmd)
os.chdir("..")
print '\n3) Extracting duration'
post_process(os.path.abspath(predict_file), dur_file)
print '\n4) Writing TextGrid file to %s ...' % output_path
create_text_grid(dur_file, output_path, length, float(0.0))
# remove leftovers
if os.path.exists(tmp_dir):
shutil.rmtree(tmp_dir)
def measurement_features(audio_path, textgrid_path, output_path):
# defines
tmp_dir = 'tmp/'
tmp_input = tmp_dir + 'tmp.input'
tmp_label = tmp_dir + 'tmp.labels'
label_suffix = '.labels'
tmp_features = tmp_dir + 'tmp.features'
tmp_file = tmp_dir + 'tmp.wav'
epsilon = 0.001
# validation
if not os.path.exists(audio_path):
print >> sys.stderr, 'Error: input path does not exists.'
return
if not os.path.exists(output_path):
print 'output path does not exists, creating output directory.'
os.mkdir(output_path)
# create tmp dir
if os.path.exists(tmp_dir):
st.rmtree(tmp_dir)
os.mkdir(tmp_dir)
# loop over all the files in the input dir
for item in os.listdir(audio_path):
if item.endswith('.wav'):
try:
# convert to 16K 16bit
cmd = 'sbin/sox %s -r 16000 -b 16 %s' % (audio_path + item,
tmp_file)
utils.easy_call(cmd)
# parse the textgrid
textgrid = TextGrid()
textgrid.read(textgrid_path +
item.replace('.wav', '.TextGrid'))
length = textgrid._TextGrid__tiers[0]._IntervalTier__intervals[
2]._Interval__xmax
onset = textgrid._TextGrid__tiers[0]._IntervalTier__intervals[
1]._Interval__xmin
offset = textgrid._TextGrid__tiers[0]._IntervalTier__intervals[
1]._Interval__xmax
start_extract = 0
end_extract = min(offset + 0.08, length - epsilon)
# =================== ACOUSTIC FEATURES =================== #
# write labels
label_file = output_path + item.replace('.wav', label_suffix)
fid = open(label_file, 'w')
fid.write('1 2\n')
# fid.write('%s %s %s\n' % (
# int((voicing_start - start_extract) * 1000 + 1), int((voicing_end - start_extract) * 1000 + 1),
# int((release_end - start_extract) * 1000 + 1)))
fid.write('%s %s %s\n' %
(int(onset * 1000) + 1, int(offset * 1000) + 1,
int(offset * 1000) + 4))
fid.close()
# creating the files
input_file = open(
tmp_features,
'wb') # open the input file for the feature extraction
features_file = open(
tmp_input, 'wb') # open file for the feature list path
labels_file = open(tmp_label, 'wb') # open file for the labels
# write the data
input_file.write('"' + tmp_file + '" ' +
str('%.8f' % float(start_extract)) + ' ' +
str(float(end_extract)) + ' ' +
str('%.8f' % float(onset)) + ' ' +
str('%.8f' % float(offset)))
features_file.write(output_path + item.replace('.wav', '.txt'))
input_file.close()
features_file.close()
labels_file.close()
command = "./sbin/VotFrontEnd2 %s %s %s" % (
input_file.name, features_file.name, labels_file.name)
utils.easy_call(command)
# remove leftovers
os.remove(tmp_input)
os.remove(tmp_label)
os.remove(tmp_features)
except:
print item
st.rmtree(tmp_dir)
def extract_features(audio_path, output_path, start_extract, end_extract):
# defines
tmp_dir = 'tmp/'
tmp_input = tmp_dir + 'tmp.input'
tmp_label = tmp_dir + 'tmp.labels'
label_suffix = '.labels'
tmp_features = tmp_dir + 'tmp.features'
tmp_file = tmp_dir + 'tmp.wav'
# validation
if not os.path.exists(audio_path):
print >> sys.stderr, 'Error: input path does not exists.'
return
# create tmp dir
if os.path.exists(tmp_dir):
st.rmtree(tmp_dir)
os.mkdir(tmp_dir)
# loop over all the files in the input dir
if audio_path.endswith('.wav'):
try:
# convert to 16K 16bit
cmd = 'sbin/sox %s -r 16000 -b 16 %s' % (audio_path, tmp_file)
utils.easy_call(cmd)
onset = (float(start_extract) + float(end_extract)) / 2
offset = (float(start_extract) + float(end_extract)) / 2
# =================== ACOUSTIC FEATURES =================== #
# # write labels
# label_file = audio_path.replace('.wav', label_suffix)
# fid = open(label_file, 'w')
# fid.write('1 2\n')
# fid.write('%s %s %s\n' % (str(1), str(1), str(1)))
# fid.close()
# creating the files
input_file = open(tmp_features, 'wb') # open the input file for the feature extraction
features_file = open(tmp_input, 'wb') # open file for the feature list path
labels_file = open(tmp_label, 'wb') # open file for the labels
# write the data
input_file.write(
'"' + tmp_file + '" ' + str('%.8f' % float(start_extract)) + ' ' + str(
float(end_extract)) + ' ' + str(
'%.8f' % float(onset)) + ' ' + str('%.8f' % float(offset)))
features_file.write(output_path.replace('.wav', '.txt'))
input_file.close()
features_file.close()
labels_file.close()
command = "./sbin/VotFrontEnd2 %s %s %s" % (input_file.name, features_file.name, labels_file.name)
utils.easy_call(command)
# remove leftovers
os.remove(tmp_input)
os.remove(tmp_label)
os.remove(tmp_features)
except:
print audio_path
st.rmtree(tmp_dir)
def extract_features(wav_filename, output_path, start_extract, end_extract):
# defines
temp_input_filename = utils.generate_tmp_filename('input')
temp_label_filename = utils.generate_tmp_filename('labels')
temp_features_filename = utils.generate_tmp_filename('features')
temp_wav16_filename = utils.generate_tmp_filename('wav')
# validation
if not os.path.exists(wav_filename):
print >> sys.stderr, 'Error: input path %s does not exists.' % wav_filename
return
# loop over all the files in the input dir
if wav_filename.endswith('.wav'):
try:
# convert to 16K 16bit
cmd = 'sox %s -r 16000 -b 16 %s' % (wav_filename,
temp_wav16_filename)
utils.easy_call(cmd)
onset = (float(start_extract) + float(end_extract)) / 2
offset = (float(start_extract) + float(end_extract)) / 2
# =================== ACOUSTIC FEATURES =================== #
# # write labels
# label_file = wav_filename.replace('.wav', label_suffix)
# fid = open(label_file, 'w')
# fid.write('1 2\n')
# fid.write('%s %s %s\n' % (str(1), str(1), str(1)))
# fid.close()
# creating the files
input_file = open(
temp_features_filename,
'wb') # open the input file for the feature extraction
features_file = open(temp_input_filename,
'wb') # open file for the feature list path
labels_file = open(temp_label_filename,
'wb') # open file for the labels
# write the data
input_file.write('"' + temp_wav16_filename + '" ' +
str('%.8f' % float(start_extract)) + ' ' +
str(float(end_extract)) + ' ' +
str('%.8f' % float(onset)) + ' ' +
str('%.8f' % float(offset)))
features_file.write(output_path.replace('.wav', '.txt'))
input_file.close()
features_file.close()
labels_file.close()
command = "./sbin/VotFrontEnd2 %s %s %s" % (
input_file.name, features_file.name, labels_file.name)
utils.easy_call(command)
# remove leftovers
os.remove(temp_input_filename)
os.remove(temp_label_filename)
os.remove(temp_features_filename)
os.remove(temp_wav16_filename)
except:
print wav_filename
def measurement_features(audio_path, textgrid_path, output_path):
# defines
tmp_dir = 'tmp/'
tmp_input = tmp_dir + 'tmp.input'
tmp_label = tmp_dir + 'tmp.labels'
label_suffix = '.labels'
tmp_features = tmp_dir + 'tmp.features'
tmp_file = tmp_dir + 'tmp.wav'
gap_start = 0.05
# gap_start = 0.1
gap_end = 0.05
# validation
if not os.path.exists(audio_path):
print >> sys.stderr, 'Error: input path does not exists.'
return
if not os.path.exists(output_path):
print 'output path does not exists, creating output directory.'
os.mkdir(output_path)
# create tmp dir
if os.path.exists(tmp_dir):
st.rmtree(tmp_dir)
os.mkdir(tmp_dir)
# loop over all the files in the input dir
for item in os.listdir(audio_path):
if item.endswith('.wav'):
try:
# convert to 16K 16bit
cmd = 'sbin/sox %s -r 16000 -b 16 %s' % (audio_path + item, tmp_file)
utils.easy_call(cmd)
# parse the textgrid
textgrid = TextGrid()
textgrid.read(textgrid_path + item.replace('.wav', '.TextGrid'))
release_start = textgrid._TextGrid__tiers[2]._IntervalTier__intervals[1]._Interval__xmin
release_end = textgrid._TextGrid__tiers[2]._IntervalTier__intervals[1]._Interval__xmax
voicing_start = textgrid._TextGrid__tiers[5]._IntervalTier__intervals[1]._Interval__xmin
voicing_end = textgrid._TextGrid__tiers[5]._IntervalTier__intervals[1]._Interval__xmax
# onset = min(release_start, voicing_start)
# offset = max(release_end, voicing_end)
onset = release_start
offset = release_end
start_extract = onset - gap_start
end_extract = offset + gap_end
# =================== ACOUSTIC FEATURES =================== #
# write labels
label_file = output_path + item.replace('.wav', label_suffix)
fid = open(label_file, 'w')
fid.write('1 2\n')
# fid.write('%s %s %s\n' % (
# int((voicing_start - start_extract) * 1000 + 1), int((voicing_end - start_extract) * 1000 + 1),
# int((release_end - start_extract) * 1000 + 1)))
fid.write('%s %s %s %s\n' % (
int((release_start - start_extract) * 1000 + 1), int((release_end - start_extract) * 1000 + 1),
int((voicing_start - start_extract) * 1000 + 1), int((voicing_end - start_extract) * 1000 + 1)))
fid.close()
# creating the files
input_file = open(tmp_features, 'wb') # open the input file for the feature extraction
features_file = open(tmp_input, 'wb') # open file for the feature list path
labels_file = open(tmp_label, 'wb') # open file for the labels
# write the data
input_file.write(
'"' + tmp_file + '" ' + str('%.8f' % float(start_extract)) + ' ' + str(
float(end_extract)) + ' ' + str(
'%.8f' % float(onset)) + ' ' + str('%.8f' % float(offset)))
features_file.write(output_path + item.replace('.wav', '.txt'))
input_file.close()
features_file.close()
labels_file.close()
command = "./sbin/VotFrontEnd2 %s %s %s" % (input_file.name, features_file.name, labels_file.name)
utils.easy_call(command)
# remove leftovers
os.remove(tmp_input)
os.remove(tmp_label)
os.remove(tmp_features)
except:
print item
st.rmtree(tmp_dir)
def neg_vot_creator(audio_path, textgrid_path, output_path, l):
# defines
tmp_dir = 'tmp/'
tmp_input = tmp_dir + 'tmp.input'
tmp_label = tmp_dir + 'tmp.labels'
label_suffix = '.labels'
tmp_features = tmp_dir + 'tmp.features'
tmp_file = tmp_dir + 'tmp.wav'
# validation
if not os.path.exists(audio_path):
print >> sys.stderr, 'Error: input path does not exists.'
return
if not os.path.exists(output_path):
print 'output path does not exists, creating output directory.'
os.mkdir(output_path)
# create tmp dir
if os.path.exists(tmp_dir):
st.rmtree(tmp_dir)
os.mkdir(tmp_dir)
count = 0
# loop over all the files in the input dir
for item in os.listdir(audio_path):
if item.endswith('.wav'):
try:
# convert to 16K 16bit
cmd = 'sbin/sox %s -r 16000 -b 16 %s' % (audio_path + item, tmp_file)
utils.easy_call(cmd)
# parse the textgrid
textgrid = TextGrid()
textgrid.read(textgrid_path + item.replace('.wav', '.TextGrid'))
release_start = textgrid._TextGrid__tiers[2]._IntervalTier__intervals[1]._Interval__xmin
end_time = release_start
if end_time - 0.1 < 0:
count += 1
start_time = max(0, end_time - 0.1)
# =================== ACOUSTIC FEATURES =================== #
# write labels
label_file = output_path + item.replace('.wav', label_suffix)
fid = open(label_file, 'w')
fid.write('%s\n' % str(l))
fid.close()
# creating the files
input_file = open(tmp_features, 'wb') # open the input file for the feature extraction
features_file = open(tmp_input, 'wb') # open file for the feature list path
labels_file = open(tmp_label, 'wb') # open file for the labels
# write the data
input_file.write(
'"' + tmp_file + '" ' + str('%.8f' % float(start_time)) + ' ' + str(
float(end_time)) + ' ' + str(
'%.8f' % float(start_time)) + ' ' + str('%.8f' % float(end_time)))
features_file.write(output_path + item.replace('.wav', '.txt'))
input_file.close()
features_file.close()
labels_file.close()
command = "./sbin/VowelDurationFrontEnd %s %s %s" % (input_file.name, features_file.name, labels_file.name)
utils.easy_call(command)
# remove leftovers
os.remove(tmp_input)
os.remove(tmp_label)
os.remove(tmp_features)
except:
print item
st.rmtree(tmp_dir)
def neg_vot_creator(audio_path, textgrid_path, output_path, l):
# defines
tmp_dir = 'tmp/'
tmp_input = tmp_dir + 'tmp.input'
tmp_label = tmp_dir + 'tmp.labels'
label_suffix = '.labels'
tmp_features = tmp_dir + 'tmp.features'
tmp_file = tmp_dir + 'tmp.wav'
# validation
if not os.path.exists(audio_path):
print >> sys.stderr, 'Error: input path does not exists.'
return
if not os.path.exists(output_path):
print 'output path does not exists, creating output directory.'
os.mkdir(output_path)
# create tmp dir
if os.path.exists(tmp_dir):
st.rmtree(tmp_dir)
os.mkdir(tmp_dir)
count = 0
# loop over all the files in the input dir
for item in os.listdir(audio_path):
if item.endswith('.wav'):
try:
# convert to 16K 16bit
cmd = 'sbin/sox %s -r 16000 -b 16 %s' % (audio_path + item,
tmp_file)
utils.easy_call(cmd)
# parse the textgrid
textgrid = TextGrid()
textgrid.read(textgrid_path +
item.replace('.wav', '.TextGrid'))
release_start = textgrid._TextGrid__tiers[
2]._IntervalTier__intervals[1]._Interval__xmin
end_time = release_start
if end_time - 0.1 < 0:
count += 1
start_time = max(0, end_time - 0.1)
# =================== ACOUSTIC FEATURES =================== #
# write labels
label_file = output_path + item.replace('.wav', label_suffix)
fid = open(label_file, 'w')
fid.write('%s\n' % str(l))
fid.close()
# creating the files
input_file = open(
tmp_features,
'wb') # open the input file for the feature extraction
features_file = open(
tmp_input, 'wb') # open file for the feature list path
labels_file = open(tmp_label, 'wb') # open file for the labels
# write the data
input_file.write('"' + tmp_file + '" ' +
str('%.8f' % float(start_time)) + ' ' +
str(float(end_time)) + ' ' +
str('%.8f' % float(start_time)) + ' ' +
str('%.8f' % float(end_time)))
features_file.write(output_path + item.replace('.wav', '.txt'))
input_file.close()
features_file.close()
labels_file.close()
command = "./sbin/VowelDurationFrontEnd %s %s %s" % (
input_file.name, features_file.name, labels_file.name)
utils.easy_call(command)
# remove leftovers
os.remove(tmp_input)
os.remove(tmp_label)
os.remove(tmp_features)
except:
print item
st.rmtree(tmp_dir)