def test_wfst_decoder_normal_transition():
phoneme_table = PhonemeTable()
phoneme_table.add_labels(phonemes)
fst_compiler = _FstCompiler()
eps = phoneme_table.get_epsilon_id()
blank = phoneme_table.get_blank_id()
a = phoneme_table.get_label_id('a')
i = phoneme_table.get_label_id('i')
fst_compiler.add_arc(0, 1, blank, eps, 0.2)
fst_compiler.add_arc(1, 2, a, eps, 0.1)
fst_compiler.add_arc(1, 3, i, eps, 0.2)
fst = fst_compiler.compile()
wfst_decoder = WFSTDecoder(fst)
prev_paths = {
0: wfst_decoder.Path(score=0,
prev_path=None,
frame_index=0,
olabel=None)
}
curr_paths = {}
wfst_decoder.normal_transition(prev_paths, curr_paths, 0, blank)
assert 1 in curr_paths
assert round(curr_paths[1].score, 6) == 0.2
assert round(curr_paths[1].prev_path.score, 6) == 0
prev_paths = curr_paths
curr_paths = {}
wfst_decoder.normal_transition(prev_paths, curr_paths, 1, a)
assert 2 in curr_paths
assert round(curr_paths[2].score, 6) == 0.3
assert curr_paths[2].frame_index == 1
assert round(curr_paths[2].prev_path.score, 6) == 0.2
def test_phoneme_table_add_labels():
phoneme_table = PhonemeTable()
phoneme_table.add_labels(['a', 'i'])
assert phoneme_table.num_labels() == 4
assert phoneme_table.get_label_id('a') == 2
assert phoneme_table.get_label(2) == 'a'
assert phoneme_table.get_label_id('i') == 3
assert phoneme_table.get_label(3) == 'i'
def test_lexicon_create_fst_without_homophones(workdir,
words_without_homophones):
vocab = get_vocabulary_table(workdir, words_without_homophones)
lexicon = get_lexicon(words_without_homophones)
phoneme_table = PhonemeTable()
phoneme_table.add_labels(phonemes)
epsilon_id = phoneme_table.get_epsilon_id()
a = phoneme_table.get_label_id('a')
i = phoneme_table.get_label_id('i')
o = phoneme_table.get_label_id('o')
fst = lexicon.create_fst(phoneme_table, vocab, min_freq=0)
assert (fst.num_states() == 7)
aux0 = phoneme_table.get_auxiliary_label_id('#0')
state = 0
assert (fst.num_arcs(0) == 2)
gen = fst.arcs(state)
arc = next(gen)
is_expected_arc(arc, a, vocab.get_label_id('愛'), 1)
arc = gen.__next__()
is_expected_arc(arc, a, vocab.get_label_id('青'), 4)
state = 1
assert (fst.num_arcs(state) == 1)
arc = next(fst.arcs(state))
is_expected_arc(arc, i, epsilon_id, 2)
state = 2
assert (fst.num_arcs(state) == 1)
arc = next(fst.arcs(state))
is_expected_arc(arc, aux0, epsilon_id, 3)
state = 3
assert (fst.num_arcs(state) == 1)
arc = next(fst.arcs(state))
is_expected_arc(arc, epsilon_id, epsilon_id, 0)
state = 4
assert (fst.num_arcs(state) == 1)
arc = next(fst.arcs(state))
is_expected_arc(arc, o, epsilon_id, 5)
state = 5
assert (fst.num_arcs(state) == 1)
arc = next(fst.arcs(state))
is_expected_arc(arc, aux0, epsilon_id, 6)
state = 6
assert (fst.num_arcs(state) == 1)
arc = next(fst.arcs(state))
is_expected_arc(arc, epsilon_id, epsilon_id, 0)
def test_compose_token_and_lexicon_fst(workdir, words_without_homophones):
vocab = get_vocabulary_table(workdir, words_without_homophones)
lexicon = get_lexicon(words_without_homophones)
phoneme_table = PhonemeTable()
phoneme_table.add_labels(phonemes)
lexicon_fst = lexicon.create_fst(phoneme_table, vocab)
token = Token()
token_fst = token.create_fst(phoneme_table)
fst = pywrapfst.compose(token_fst.arcsort('olabel'), lexicon_fst)
fst = pywrapfst.determinize(fst)
def test_compose_token_and_lexicon_fst_with_homophones(workdir,
words_with_homophones):
vocab = get_vocabulary_table(workdir, words_with_homophones)
lexicon = get_lexicon(words_with_homophones)
phoneme_table = PhonemeTable()
phoneme_table.add_labels(phonemes)
lexicon_fst = lexicon.create_fst(phoneme_table, vocab, min_freq=0)
token = Token()
token_fst = token.create_fst(phoneme_table)
fst = pywrapfst.compose(token_fst.arcsort('olabel'), lexicon_fst)
with pytest.raises(pywrapfst.FstOpError):
pywrapfst.determinize(fst)
def test_token_create_fst_with_auxiliary_labels():
phoneme_table = PhonemeTable()
phoneme_table.add_labels(['a', 'i'])
epsilon_id = phoneme_table.get_epsilon_id()
blank_id = phoneme_table.get_blank_id()
a = phoneme_table.get_label_id('a')
i = phoneme_table.get_label_id('i')
phoneme_table.set_auxiliary_label('#0')
phoneme_table.set_auxiliary_label('#1')
aux0 = phoneme_table.get_auxiliary_label_id('#0')
aux1 = phoneme_table.get_auxiliary_label_id('#1')
fst = Token().create_fst(phoneme_table)
assert (fst.num_states() == 5)
# start state
state = 0
assert (fst.num_arcs(state) == 3)
gen_arc = fst.arcs(state)
is_expected_arc(next(gen_arc), blank_id, epsilon_id, state)
is_expected_arc(next(gen_arc), a, a, 3)
is_expected_arc(next(gen_arc), i, i, 4)
# second state
state = 1
assert (fst.num_arcs(state) == 2)
gen_arc = fst.arcs(state)
is_expected_arc(next(gen_arc), blank_id, epsilon_id, state)
is_expected_arc(next(gen_arc), epsilon_id, epsilon_id, 2)
# final(auxiliary) state
state = 2
assert (fst.num_arcs(state) == 3)
gen_arc = fst.arcs(state)
is_expected_arc(next(gen_arc), epsilon_id, epsilon_id, 0)
is_expected_arc(next(gen_arc), epsilon_id, aux0, state)
is_expected_arc(next(gen_arc), epsilon_id, aux1, state)
# a
state = 3
assert (fst.num_arcs(state) == 2)
gen_arc = fst.arcs(state)
is_expected_arc(next(gen_arc), a, epsilon_id, state)
is_expected_arc(next(gen_arc), epsilon_id, epsilon_id, 1)
# b
state = 4
assert (fst.num_arcs(state) == 2)
gen_arc = fst.arcs(state)
is_expected_arc(next(gen_arc), i, epsilon_id, state)
is_expected_arc(next(gen_arc), epsilon_id, epsilon_id, 1)
def test_wfst_decoder_epsilon_transition():
phoneme_table = PhonemeTable()
phoneme_table.add_labels(phonemes)
fst_compiler = _FstCompiler()
eps = phoneme_table.get_epsilon_id()
a = phoneme_table.get_label_id('a')
fst_compiler.add_arc(0, 1, eps, eps, 0.1)
fst_compiler.add_arc(1, 2, eps, eps, 0.2)
fst_compiler.add_arc(1, 3, eps, eps, 0.3)
fst_compiler.add_arc(0, 2, eps, eps, 0.15)
fst_compiler.add_arc(0, 3, eps, eps, 0.5)
fst_compiler.add_arc(0, 4, a, eps, 0.15)
fst_compiler.set_final(4)
fst = fst_compiler.compile()
wfst_decoder = WFSTDecoder(fst)
paths = {
0: wfst_decoder.Path(score=0,
prev_path=None,
frame_index=0,
olabel=None)
}
frame_index = 0
wfst_decoder.epsilon_transition(paths, phoneme_table, frame_index)
# check new state 1 is added to paths
# TODO: state:1が残るのは果たしてよいのか？無限ループしそう
assert 1 in paths
assert round(paths[1].score, 6) == 0.1
assert paths[1].prev_path.score == 0
# check existing state 2 is updated via better path
assert 2 in paths
assert round(paths[2].score, 6) == 0.15
assert paths[2].prev_path.score == 0
# check existing state 3 is not updated
assert round(paths[3].score, 6) == 0.4
assert round(paths[3].prev_path.score, 6) == 0.1
assert paths[3].prev_path.prev_path.score == 0
# check new state 4 is not added to paths
# because it's not epsilon transition
assert 4 not in paths
def test_wfst_decoder_decode(workdir, words_for_corpus_with_homophones):
corpus_path = os.path.join(workdir, 'corpus.txt')
create_corpus(corpus_path, words_for_corpus_with_homophones)
vocab_path = os.path.join(workdir, 'vocab.syms')
vocab = create_vocabulary_symbol_table(vocab_path, corpus_path)
phoneme_table = PhonemeTable()
phoneme_table.add_labels(phonemes)
lexicon = get_lexicon(words_for_corpus_with_homophones)
lexicon_fst = lexicon.create_fst(phoneme_table, vocab, min_freq=0)
token = Token()
token_fst = token.create_fst(phoneme_table)
grammar_path = os.path.join(workdir, 'grammar.fst')
grammar = Grammar()
grammar_fst = grammar.create_fst(grammar_path, vocab_path, corpus_path)
wfst_decoder = WFSTDecoder()
wfst_decoder.create_fst(token_fst, lexicon_fst, grammar_fst)
blank_id = phoneme_table.get_blank_id()
a = phoneme_table.get_label_id('a')
i = phoneme_table.get_label_id('i')
d = phoneme_table.get_label_id('d')
e = phoneme_table.get_label_id('e')
s = phoneme_table.get_label_id('s')
o = phoneme_table.get_label_id('o')
m = phoneme_table.get_label_id('m')
r = phoneme_table.get_label_id('r')
u = phoneme_table.get_label_id('u')
frame_labels = [
blank_id, blank_id, a, a, i, i, i, d, e, blank_id, s, s, o, o, o, m, e,
r, r, u
]
got = wfst_decoder.decode(frame_labels, vocab)
assert got == '藍で染める'
def test_wfst_decoder_create_fst(workdir, words_for_corpus_without_homophones):
corpus_path = os.path.join(workdir, 'corpus.txt')
create_corpus(corpus_path, words_for_corpus_without_homophones)
vocab_path = os.path.join(workdir, 'vocab.syms')
vocab = create_vocabulary_symbol_table(vocab_path, corpus_path)
phoneme_table = PhonemeTable()
phoneme_table.add_labels(phonemes)
lexicon = get_lexicon(words_for_corpus_without_homophones)
lexicon_fst = lexicon.create_fst(phoneme_table, vocab, min_freq=0)
token = Token()
token_fst = token.create_fst(phoneme_table)
grammar_path = os.path.join(workdir, 'grammar.fst')
grammar = Grammar()
grammar_fst = grammar.create_fst(grammar_path, vocab_path, corpus_path)
wfst_decoder = WFSTDecoder()
wfst_decoder.create_fst(token_fst, lexicon_fst, grammar_fst)
parser.add_argument('--lr', type=float, default=0.001, help='Learning rate')
parser.add_argument('--epochs',
type=int,
default=5,
help='number of epochs for training')
parser.add_argument('--batch-size', type=int, default=32, help='batch size')
parser.add_argument('--device', type=str, default='cpu', help='Device string')
parser.add_argument('--model-file',
type=str,
default="model.bin",
help='Model file to save')
parser.add_argument('--resume', action='store_true')
args = parser.parse_args()
phoneme_table = PhonemeTable()
phoneme_table.add_labels(phonemes)
print('Loading training data ...')
training_data_dirpath = os.path.join(args.workdir, args.training_data_dirname)
repository_tr = TrainingDatasetRepository(training_data_dirpath)
dataset_tr = IterableAudioDataset(repository_tr, phoneme_table)
dataloader_tr = DataLoader(dataset_tr,
batch_size=args.batch_size,
collate_fn=collate_for_ctc)
print('Loading development data ...')
development_data_dirpath = os.path.join(args.workdir,
args.development_data_dirname)
repository_dev = DevelopmentDatasetRepository(development_data_dirpath)
dataloaders_dev = []
for dataset_dev in AudioDataset.load_all(repository_dev, phoneme_table):
dataloader_dev = DataLoader(dataset_dev,
batch_size=args.batch_size,