def test_tedlium(self):
transitions = durmodel_utils.read_transitions(os.path.dirname(__file__) + "/test_data/tedlium/transitions.txt")
nonsilence_phonemes = set()
for l in open(os.path.dirname(__file__) + "/test_data/tedlium/nonsilence.txt"):
nonsilence_phonemes.add(l.partition("_")[0])
word_list = []
for l in codecs.open(os.path.dirname(__file__) + "/test_data/tedlium/words.txt", encoding="UTF-8"):
word_list.append(l.split()[0])
stress_dict = durmodel_utils.load_stress_dict(os.path.dirname(__file__) + "/data/en/cmudict.0.7a.lc")
# about
word_id = 437
frames = [int(s) for s in "1794_2062_2061_2061_2300_5602_5601_5650_5662_5661_5661_4808_4807_4807_4807_4807_4832_4831_4831_4831_4860_4859_4859_22924_22923_22923_23018_23118".split("_")]
features_and_dur_seq = durmodel_utils.make_local(0, word_id, frames, transitions, word_list, nonsilence_phonemes, language="ENGLISH", stress_dict=stress_dict)
print features_and_dur_seq
self.assert_(('syllable', 1) in features_and_dur_seq[0][0])
self.assert_(('AH', 1) in features_and_dur_seq[0][0])
self.assert_(('vowel', 1) in features_and_dur_seq[0][0])
self.assert_(5, features_and_dur_seq[0][1])
self.assert_(('stress1', 1) in features_and_dur_seq[2][0])
self.assert_(('syllable', 2) in features_and_dur_seq[1][0])
self.assert_(('stop', 1) in features_and_dur_seq[-1][0])
self.assert_(5, features_and_dur_seq[0][-1])
def test_stress(self):
word = "abstain"
phones = "AH B S T EY N".split()
stress_dict = durmodel_utils.load_stress_dict(os.path.dirname(__file__) + "/data/en/cmudict.0.7a.lc")
self.assertEqual(['AH', 'B', 'S', 'T', 'EY', 'N'], stress_dict["abstain"][0][0])
self.assertEqual([0, 0, 0, 0, 1, 0], stress_dict["abstain"][0][1])
stress = durmodel_utils.get_stress(word, phones, stress_dict)
self.assertEqual([0, 0, 0, 0, 1, 0], stress)
# test unknown word
stress = durmodel_utils.get_stress("tanel", "T AH N EH L".split(), stress_dict)
self.assertEqual([0, 0, 0, 0, 0], stress)
def test_stress(self):
word = 'abstain'
phones = 'AH B S T EY N'.split()
stress_dict = durmodel_utils.load_stress_dict(
dirname(__file__) + '/data/en/cmudict.0.7a.lc')
self.assertEqual(['AH', 'B', 'S', 'T', 'EY', 'N'],
stress_dict['abstain'][0][0])
self.assertEqual([0, 0, 0, 0, 1, 0], stress_dict['abstain'][0][1])
stress = durmodel_utils.get_stress(word, phones, stress_dict)
self.assertEqual([0, 0, 0, 0, 1, 0], stress)
# test unknown word
stress = durmodel_utils.get_stress('tanel', 'T AH N EH L'.split(),
stress_dict)
self.assertEqual([0, 0, 0, 0, 0], stress)
def test_stress(self):
word = "abstain"
phones = "AH B S T EY N".split()
stress_dict = durmodel_utils.load_stress_dict(
os.path.dirname(__file__) + "/data/en/cmudict.0.7a.lc")
self.assertEqual(['AH', 'B', 'S', 'T', 'EY', 'N'],
stress_dict["abstain"][0][0])
self.assertEqual([0, 0, 0, 0, 1, 0], stress_dict["abstain"][0][1])
stress = durmodel_utils.get_stress(word, phones, stress_dict)
self.assertEqual([0, 0, 0, 0, 1, 0], stress)
# test unknown word
stress = durmodel_utils.get_stress("tanel", "T AH N EH L".split(),
stress_dict)
self.assertEqual([0, 0, 0, 0, 0], stress)
def test_tedlium(self):
transitions = durmodel_utils.read_transitions(
os.path.dirname(__file__) + "/test_data/tedlium/transitions.txt")
nonsilence_phonemes = set()
for l in open(
os.path.dirname(__file__) +
"/test_data/tedlium/nonsilence.txt"):
nonsilence_phonemes.add(l.partition("_")[0])
word_list = []
for l in codecs.open(os.path.dirname(__file__) +
"/test_data/tedlium/words.txt",
encoding="UTF-8"):
word_list.append(l.split()[0])
stress_dict = durmodel_utils.load_stress_dict(
os.path.dirname(__file__) + "/data/en/cmudict.0.7a.lc")
# about
word_id = 437
frames = [
int(s) for s in
"1794_2062_2061_2061_2300_5602_5601_5650_5662_5661_5661_4808_4807_4807_4807_4807_4832_4831_4831_4831_4860_4859_4859_22924_22923_22923_23018_23118"
.split("_")
]
features_and_dur_seq = durmodel_utils.make_local(
0,
word_id,
frames,
transitions,
word_list,
nonsilence_phonemes,
language="ENGLISH",
stress_dict=stress_dict)
print features_and_dur_seq
self.assert_(('syllable', 1) in features_and_dur_seq[0][0])
self.assert_(('AH', 1) in features_and_dur_seq[0][0])
self.assert_(('vowel', 1) in features_and_dur_seq[0][0])
self.assert_(5, features_and_dur_seq[0][1])
self.assert_(('stress1', 1) in features_and_dur_seq[2][0])
self.assert_(('syllable', 2) in features_and_dur_seq[1][0])
self.assert_(('stop', 1) in features_and_dur_seq[-1][0])
self.assert_(5, features_and_dur_seq[0][-1])
print >> sys.stderr, "DEBUG: transitions[%d] = %s" % (len(transitions) -1, transitions[-1])
print >> sys.stderr, "Reading non-silence phonemes"
nonsilence_phonemes = set()
for l in open(args.nonsilence):
nonsilence_phonemes.add(l.partition("_")[0])
print >> sys.stderr, "Reading words.txt"
word_list = []
for l in codecs.open(args.words, encoding="UTF-8"):
word_list.append(l.split()[0])
stress_dict = None
if args.stress_dict_filename:
print >> sys.stderr, "Reading stress dictionary"
stress_dict = durmodel_utils.load_stress_dict(args.stress_dict_filename)
num_sentences_read = 0
full_features_and_durs = []
sentence_lines = []
print >> sys.stderr, "Processing alignements..."
for l in gzip.open(args.train_lattice):
if len(l.strip()) > 0:
sentence_lines.append(l)
elif len(sentence_lines) > 0:
try:
lat = lattice.parse_aligned_lattice(sentence_lines)
#print >> sys.stderr, "Processing lattice %s" % lat.name
features_and_durs = []
nonsilence_phonemes = set()
for l in open(args.nonsilence):
nonsilence_phonemes.add(l.partition("_")[0])
print >> sys.stderr, "Reading words.txt"
word_list = []
for l in codecs.open(args.words, encoding="UTF-8"):
word_list.append(l.split()[0])
filler_words = ["<eps>"] + args.fillers.split(",")
print >> sys.stderr, "Fillers: ", filler_words
stress_dict = None
if args.stress_dict_filename:
print >> sys.stderr, "Reading stress dictionary"
stress_dict = durmodel_utils.load_stress_dict(
args.stress_dict_filename)
feature_dict = OrderedDict()
if args.read_features_filename:
print >> sys.stderr, ".. Reading features from %s" % args.read_features_filename
feature_dict = OrderedDict()
for (i, feature) in enumerate(
codecs.open(args.read_features_filename, encoding="UTF-8")):
feature_dict[feature.strip()] = i
speaker_ids = None
if args.speakers:
print >> sys.stderr, ".. Reading speakers from %s" % args.speakers
speaker_ids = {}
for l in open(args.speakers):
ss = l.split()
