mel_min = np.amin(mel)

mel_max = np.amax(mel)

text = ""

for line in turn:

if line == '':

continue

features = {}

words = line.split()

start = float(words[0]) / 1000

end = float(words[1]) / 1000

time_scale = [t*0.02 for t in range(mel.shape[0])]

times = [t/0.02 for t in time_scale if (t>start and t<end)]

for i in range(1, mel.shape[1]):

j = "mfcc_%s" %i

# features["moy_loc_%s" %j] =\

# np.mean(map(lambda x: (mel[x,i] - moy[j]) / np.sqrt(var[j]), times))

# features["var_loc_%s" %j] =\

# np.var(map(lambda x: (mel[x,i] - moy[j]) / np.sqrt(var[j]), times))

features["moy_loc_%s" %j] =\

np.mean(map(lambda x: (mel[x,i] - mel_min) / (mel_max - mel_min), times))

features["var_loc_%s" %j] =\

np.var(map(lambda x: (mel[x,i] - mel_min) / (mel_max - mel_min), times))

try:

text += words[2]+'\t'+str(features)+'\n'

except BaseException, e:

print e

"""attention, retourne directement le tableau du signal."""

f = Sndfile(path1)

n = int(f.nframes)

fs = f.samplerate

nc = f.channels # checker que c'est bien 1

if nc != 1:

raise Exception('Fichier wav possédant plusieurs canaux.')

enc = f.encoding # checker que c'est bien 'pcm24'

if enc != 'pcm24':

raise Exception("Encodage différent de 'pcm24'.")

data = f.read_frames(n, np.float64)

m = np.mean(data, dtype=np.float64)

v = np.var(data, dtype=np.float64)

data = (data - m) / np.sqrt(v)

name = session_dir

if len(name) == 1:

name2 = '00'+name

elif len(name) == 2:

name2 = '0'+name

else:

name2 = name

print "Chargement des différents fichiers... session %s" % session_dir

# trouver les fichiers audio

l_op = glob.glob(S_PATH+name+'/*Operator HeadMounted*.wav')

l_us = glob.glob(S_PATH+name+'/*User HeadMounted*.wav')

if len(l_op) != 1 or len(l_us) != 1:

raise Exception('Zero ou multiples matchs pour les fichiers audio')

else:

wav_op = l_op[0]

wav_us = l_us[0]

# trouver les fichiers txt:

l_op = glob.glob(S_PATH+name+'/word*operator*')

l_us = glob.glob(S_PATH+name+'/word*user*')

if len(l_op) != 1 or len(l_us) != 1:

raise Exception('Zero ou multiples matchs pour les fichiers txt')

else:

txt_op = l_op[0]

txt_us = l_us[0]

# normaliser le signal

try:

data_op, fs_op = normalize_signal(wav_op, S_PATH+"normalized/op_"+name)

data_us, fs_us = normalize_signal(wav_us, S_PATH+"normalized/us_"+name)

except BaseException, e:

print e

# scripts mfcc operator

print "Calcul des mfcc operator..."

mel_op = mfcc(data_op, samplerate=fs_op, winlen=0.04, winstep=0.02)

moy_op = {}

var_op = {}

for i in range(1, mel_op.shape[1]): # remove first coefficient

j = "mfcc_%s" %i

moy_op[j] = np.mean(map(lambda x: mel_op[x,i], mel_op[:,i]))

var_op[j] = np.var(map(lambda x: mel_op[x,i], mel_op[:,i]))

# scripts mfcc user

print "Calcul des mfcc user..."

mel_us = mfcc(data_us, samplerate=fs_us, winlen=0.04, winstep=0.02)

moy_us = {}

var_us = {}

for i in range(1, mel_us.shape[1]): # remove first coefficient

j = "mfcc_%s" %i

moy_us[j] = np.mean(map(lambda x: mel_us[x,i], mel_us[:,i]))

var_us[j] = np.var(map(lambda x: mel_us[x,i], mel_us[:,i]))

# chargement des fichiers

try:

f_op = open(txt_op,'Ur')

turns_op = f_op.read().split('\n.\n')

f_op.close()

f_us = open(txt_us,'Ur')

turns_us = f_us.read().split('\n.\n')

f_us.close()

except NameError:

print('Text file %s not found' % name)

print "Lecture des tours de parole..."

cpt_op = 0

cpt_us = 0

complete_dump = ""

while True:

#print cpt_op

#print cpt_us

turn_op = turns_op[cpt_op].split('\n')[1:]

turn_us = turns_us[cpt_us].split('\n')[1:]

try:

start_op = float(turn_op[0].split()[0])

except IndexError:

start_op = float('Inf')

try:

start_us = float(turn_us[0].split()[0])

except IndexError:

start_us = float('Inf')

try:

if start_op < start_us:

dump = read_turn(mel_op, moy_op, var_op, turn_op)

cpt_op += 1

else:

dump = read_turn(mel_us, moy_us, var_us, turn_us)

cpt_us += 1

except BaseException, e:

print e

print turn_op

print turn_us

complete_dump += "\n\n"+dump

if cpt_op >= len(turns_op)-1 and cpt_us >= len(turns_us)-1:

break

f2 = open(D_PATH+'/dump_'+name2, 'w')

f2.write(complete_dump)