def __init__(self, path):
'''
Create the sentence object
path: (string)
the path to the particular sample
'''
if not have_sk_audiolab:
raise ValueError(
'scikits.audiolab module is required to read the TIMIT database.'
)
path, ext = os.path.splitext(path)
t = path.split('/')
# extract the attributes
self.dialect = t[-3]
self.sex = t[-2][0]
self.speaker = t[-2][1:5]
self.id = t[-1]
# Read in the wav file
f = Sndfile(path + '.WAV', 'r')
self.data = f.read_frames(f.nframes)
self.fs = f.samplerate
# Read the sentence text
f = open(path + '.TXT', 'r')
lines = f.readlines()
self.text = ' '.join(lines[0].split()[2:])
f.close()
# Read the word list
self.words = []
self.phonems = []
# open the word file
f = open(path + '.WRD', 'r')
w_lines = f.readlines()
f.close()
# get all lines from the phonem file
f_ph = open(path + '.PHN', 'r')
ph_lines = f_ph.readlines()
ph_l_index = 0
f_ph.close()
for line in w_lines:
t = line.split()
# just a sanity check
if len(t) == 3:
# the word boundary
w_bnd = np.array([int(t[0]), int(t[1])])
# recover the phonems making up the word
w_ph_list = []
while ph_l_index < len(ph_lines):
ph_line = ph_lines[ph_l_index]
u = ph_line.split()
# phonem boundary
ph_bnd = np.array([int(u[0]), int(u[1])])
# Check phonem boundary does not exceeds word boundary
if ph_bnd[1] > w_bnd[1]:
break
# add to sentence object phonems list
self.phonems.append({'name': u[2], 'bnd': ph_bnd})
# increase index
ph_l_index += 1
# now skip until beginning of word
if ph_bnd[0] < w_bnd[0]:
continue
# add phonem to word if (with adjusted boundaries wrt to start of word)
w_ph_list.append({'name': u[2], 'bnd': ph_bnd - w_bnd[0]})
# Finally create word object
self.words.append(
Word(t[2], w_bnd, self.data, self.fs, phonems=w_ph_list))
# Read the remaining phonem(s)
while ph_l_index < len(ph_lines):
ph_line = ph_lines[ph_l_index]
u = ph_line.split()
if len(u) == 3:
# phonem boundary
ph_bnd = (int(u[0]), int(u[1]))
# add to sentence object phonems list
self.phonems.append({'name': u[2], 'bnd': ph_bnd})
ph_l_index += 1
