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_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])
parser.add_argument('--language', action='store', dest='language', help="Language of the data", default="ESTONIAN") parser.add_argument('--stress', action='store', dest='stress_dict_filename', help="Stress dictionary") parser.add_argument('--left-context', action='store', dest='left_context', help="Left context length", default=2, type=int) parser.add_argument('--right-context', action='store', dest='right_context', help="Left context length", default=2, type=int) parser.add_argument('--no-duration-feature', action='store_true', dest='no_use_duration', help="Don't Use duration features") parser.add_argument('transitions', metavar='transitions.txt', help='Transition model, produced with show-transitions') parser.add_argument('nonsilence', metavar='nonsilence.txt', help='Nonsilence phonemes') parser.add_argument('words', metavar='words.txt', help='words.txt file') parser.add_argument('train_lattice', metavar='ali-lat.txt', help='Aligned phone lattice') args = parser.parse_args() durmodel_utils.LEFT_CONTEXT = args.left_context durmodel_utils.RIGHT_CONTEXT = args.right_context transitions = read_transitions(args.transitions) print >> sys.stderr, "DEBUG: transitions[%d] = %s" % (len(transitions) -2, transitions[-2]) 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
parser.add_argument( 'transitions', metavar='transitions.txt', help='Transition model, produced with show-transitions') parser.add_argument('nonsilence', metavar='nonsilence.txt', help='Nonsilence phonemes') parser.add_argument('words', metavar='words.txt', help='words.txt file') parser.add_argument('model', metavar='durmodel.pkl', help='duration model') args = parser.parse_args() durmodel_utils.LEFT_CONTEXT = args.left_context durmodel_utils.RIGHT_CONTEXT = args.right_context transitions = durmodel_utils.read_transitions(args.transitions) 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]) filler_words = ["<eps>"] + args.fillers.split(",") print >> sys.stderr, "Fillers: ", filler_words stress_dict = None
parser.add_argument('--left-context', action='store', dest='left_context', help="Left context length", default=2, type=int) parser.add_argument('--right-context', action='store', dest='right_context', help="Left context length", default=2, type=int) parser.add_argument('--skip-fillers', action='store_true', dest='skip_fillers', help="Don't calculate posterior probabilities for fillers", default=True) parser.add_argument('--utt2spk', action='store', dest='utt2spk', help="Use the mapping in the given file to add speaker ID to each sample") parser.add_argument('--speakers', action='store', dest='speakers', help="List of speakers in training data, in sorted order") parser.add_argument('transitions', metavar='transitions.txt', help='Transition model, produced with show-transitions') parser.add_argument('nonsilence', metavar='nonsilence.txt', help='Nonsilence phonemes') parser.add_argument('words', metavar='words.txt', help='words.txt file') parser.add_argument('model', metavar='durmodel.pkl', help='duration model') args = parser.parse_args() durmodel_utils.LEFT_CONTEXT = args.left_context durmodel_utils.RIGHT_CONTEXT = args.right_context transitions = durmodel_utils.read_transitions(args.transitions) 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]) filler_words = ["<eps>"] + args.fillers.split(",") print >> sys.stderr, "Fillers: ", filler_words stress_dict = None
'transitions', metavar='transitions.txt', help='Transition model, produced with show-transitions') parser.add_argument('nonsilence', metavar='nonsilence.txt', help='Nonsilence phonemes') parser.add_argument('words', metavar='words.txt', help='words.txt file') parser.add_argument('train_lattice', metavar='ali-lat.txt', help='Aligned phone lattice') args = parser.parse_args() durmodel_utils.LEFT_CONTEXT = args.left_context durmodel_utils.RIGHT_CONTEXT = args.right_context transitions = read_transitions(args.transitions) print >> sys.stderr, "DEBUG: transitions[%d] = %s" % (len(transitions) - 2, transitions[-2]) 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])