def test_basic_noposition(basic_dict_path, generated_dir):
d = Dictionary(basic_dict_path,
os.path.join(generated_dir, 'basic'),
position_dependent_phones=False)
x = d.write()
assert (set(d.phones) == set(
['sil', 'sp', 'spn', 'phonea', 'phoneb', 'phonec']))
def test_basic(basic_dict_path, generated_dir):
d = Dictionary(basic_dict_path, os.path.join(generated_dir, 'basic'))
d.write()
assert(set(d.phones) == set(['sil', 'sp','spn', 'phonea','phoneb','phonec']))
assert(set(d.positional_nonsil_phones) == set(['phonea_B','phonea_I','phonea_E', 'phonea_S',
'phoneb_B','phoneb_I','phoneb_E','phoneb_S',
'phonec_B','phonec_I','phonec_E','phonec_S']))
def test_stereo(basic_dict_path, stereo_corpus_dir, temp_dir):
temp = os.path.join(temp_dir, 'stereo')
dictionary = Dictionary(basic_dict_path, os.path.join(temp, 'basic'))
dictionary.write()
d = Corpus(stereo_corpus_dir, temp)
d.initialize_corpus(dictionary)
assert (d.get_feat_dim() == '39')
def test_basic(basic_dict_path, basic_corpus_dir, generated_dir):
dictionary = Dictionary(basic_dict_path, os.path.join(generated_dir, 'basic'))
dictionary.write()
output_directory = os.path.join(generated_dir, 'basic')
d = Corpus(basic_corpus_dir, output_directory)
d.initialize_corpus(dictionary)
assert (d.get_feat_dim() == '39')
def test_basic(basic_dict_path, generated_dir):
d = Dictionary(basic_dict_path, os.path.join(generated_dir, 'basic'))
x = d.write()
assert (set(d.phones) == set(['sil', 'spn', 'phonea', 'phoneb', 'phonec']))
assert (set(d.positional_nonsil_phones) == set([
'phonea_B', 'phonea_I', 'phonea_E', 'phonea_S', 'phoneb_B', 'phoneb_I',
'phoneb_E', 'phoneb_S', 'phonec_B', 'phonec_I', 'phonec_E', 'phonec_S'
]))
def test_basic(basic_dict_path, basic_dir, generated_dir):
dictionary = Dictionary(basic_dict_path, os.path.join(generated_dir, "basic"))
dictionary.write()
output_directory = os.path.join(generated_dir, "basic")
d = Corpus(basic_dir, output_directory)
d.write()
d.create_mfccs()
d.setup_splits(dictionary)
assert d.get_feat_dim() == "39"
def test_stereo(basic_dict_path, textgrid_directory, generated_dir):
temp = os.path.join(generated_dir, 'stereo')
dictionary = Dictionary(basic_dict_path, os.path.join(temp, 'basic'))
dictionary.write()
d = Corpus(os.path.join(textgrid_directory, 'stereo'), temp)
d.write()
d.create_mfccs()
d.setup_splits(dictionary)
assert (d.get_feat_dim() == '39')
def test_stereo(basic_dict_path, textgrid_directory, generated_dir):
temp = os.path.join(generated_dir, "stereo")
dictionary = Dictionary(basic_dict_path, os.path.join(temp, "basic"))
dictionary.write()
d = Corpus(os.path.join(textgrid_directory, "stereo"), temp)
d.write()
d.create_mfccs()
d.setup_splits(dictionary)
assert d.get_feat_dim() == "39"
def test_stereo(basic_dict_path, stereo_corpus_dir, temp_dir):
temp = os.path.join(temp_dir, 'stereo')
dictionary = Dictionary(basic_dict_path, os.path.join(temp, 'basic'))
dictionary.write()
d = Corpus(stereo_corpus_dir, temp)
d.initialize_corpus(dictionary)
fc = FeatureConfig()
fc.generate_features(d)
assert d.get_feat_dim(fc) == 39
def test_basic(basic_dict_path, basic_corpus_dir, generated_dir):
dictionary = Dictionary(basic_dict_path, os.path.join(generated_dir, 'basic'))
dictionary.write()
output_directory = os.path.join(generated_dir, 'basic')
c = Corpus(basic_corpus_dir, output_directory)
c.initialize_corpus(dictionary)
fc = FeatureConfig()
fc.generate_features(c)
assert c.get_feat_dim(fc) == 39
def test_basic(basic_dict_path, basic_corpus_dir, generated_dir):
dictionary = Dictionary(basic_dict_path, os.path.join(generated_dir, 'basic'))
dictionary.write()
output_directory = os.path.join(generated_dir, 'basic')
d = Corpus(basic_corpus_dir, output_directory)
d.initialize_corpus(dictionary)
fc = FeatureConfig()
fc.generate_features(d)
assert d.get_feat_dim(fc) == 39
def test_stereo(basic_dict_path, stereo_corpus_dir, temp_dir):
temp = os.path.join(temp_dir, 'stereo')
dictionary = Dictionary(basic_dict_path, os.path.join(temp, 'basic'))
dictionary.write()
d = Corpus(stereo_corpus_dir, temp)
d.initialize_corpus(dictionary)
fc = FeatureConfig()
fc.generate_features(d)
assert d.get_feat_dim(fc) == 39
def test_basic_txt(basic_corpus_txt_dir, basic_dict_path, generated_dir):
dictionary = Dictionary(basic_dict_path, os.path.join(generated_dir, 'basic'))
dictionary.write()
output_directory = os.path.join(generated_dir, 'basic')
c = Corpus(basic_corpus_txt_dir, output_directory)
assert len(c.no_transcription_files) == 0
c.initialize_corpus(dictionary)
fc = FeatureConfig()
fc.generate_features(c)
assert c.get_feat_dim(fc) == 39
def test_basic(basic_dict_path, basic_dir, generated_dir):
dictionary = Dictionary(basic_dict_path,
os.path.join(generated_dir, 'basic'))
dictionary.write()
output_directory = os.path.join(generated_dir, 'basic')
d = Corpus(basic_dir, output_directory)
d.write()
d.create_mfccs()
d.setup_splits(dictionary)
assert (d.get_feat_dim() == '39')
def align_corpus(corpus_dir, dict_path, output_directory, temp_dir,
output_model_path, args):
if temp_dir == '':
temp_dir = TEMP_DIR
else:
temp_dir = os.path.expanduser(temp_dir)
corpus_name = os.path.basename(corpus_dir)
if corpus_name == '':
corpus_dir = os.path.dirname(corpus_dir)
corpus_name = os.path.basename(corpus_dir)
data_directory = os.path.join(temp_dir, corpus_name)
if args.clean:
shutil.rmtree(data_directory, ignore_errors=True)
shutil.rmtree(output_directory, ignore_errors=True)
os.makedirs(data_directory, exist_ok=True)
os.makedirs(output_directory, exist_ok=True)
dictionary = Dictionary(dict_path, data_directory)
dictionary.write()
corpus = Corpus(corpus_dir,
data_directory,
args.speaker_characters,
num_jobs=args.num_jobs)
print(corpus.speaker_utterance_info())
corpus.write()
corpus.create_mfccs()
corpus.setup_splits(dictionary)
utt_oov_path = os.path.join(corpus.split_directory, 'utterance_oovs.txt')
if os.path.exists(utt_oov_path):
shutil.copy(utt_oov_path, output_directory)
oov_path = os.path.join(corpus.split_directory, 'oovs_found.txt')
if os.path.exists(oov_path):
shutil.copy(oov_path, output_directory)
mono_params = {'align_often': not args.fast}
tri_params = {'align_often': not args.fast}
tri_fmllr_params = {'align_often': not args.fast}
a = TrainableAligner(corpus,
dictionary,
output_directory,
temp_directory=data_directory,
mono_params=mono_params,
tri_params=tri_params,
tri_fmllr_params=tri_fmllr_params,
num_jobs=args.num_jobs)
a.verbose = args.verbose
a.train_mono()
a.export_textgrids()
a.train_tri()
a.export_textgrids()
a.train_tri_fmllr()
a.export_textgrids()
if output_model_path is not None:
a.save(output_model_path)
def test_short_segments(basic_dict_path, shortsegments_corpus_dir, temp_dir):
temp = os.path.join(temp_dir, 'short_segments')
dictionary = Dictionary(basic_dict_path, temp)
dictionary.write()
corpus = Corpus(shortsegments_corpus_dir, temp)
corpus.initialize_corpus(dictionary)
assert (len(corpus.feat_mapping.keys()) == 2)
assert (len(corpus.utt_speak_mapping.keys()) == 2)
assert (len(corpus.speak_utt_mapping.keys()) == 1)
assert (len(corpus.text_mapping.keys()) == 2)
assert (len(corpus.utt_wav_mapping.keys()) == 1)
assert (len(corpus.segments.keys()) == 2)
assert (len(corpus.ignored_utterances) == 1)
def test_subset(large_prosodylab_format_directory, temp_dir, large_dataset_dictionary):
output_directory = os.path.join(temp_dir, 'large_subset')
shutil.rmtree(output_directory, ignore_errors=True)
d = Dictionary(large_dataset_dictionary, output_directory)
d.write()
c = Corpus(large_prosodylab_format_directory, output_directory)
c.initialize_corpus(d)
sd = c.split_directory()
fc = FeatureConfig()
fc.generate_features(c)
s = c.subset_directory(10, fc)
assert os.path.exists(sd)
assert os.path.exists(s)
def test_subset(large_prosodylab_format_directory, temp_dir, large_dataset_dictionary):
output_directory = os.path.join(temp_dir, 'large_subset')
shutil.rmtree(output_directory, ignore_errors=True)
d = Dictionary(large_dataset_dictionary, output_directory)
d.write()
c = Corpus(large_prosodylab_format_directory, output_directory)
c.initialize_corpus(d)
sd = c.split_directory()
fc = FeatureConfig()
fc.generate_features(c)
s = c.subset_directory(10, fc)
assert os.path.exists(sd)
assert os.path.exists(s)
def test_short_segments(basic_dict_path, shortsegments_corpus_dir, temp_dir):
temp = os.path.join(temp_dir, 'short_segments')
dictionary = Dictionary(basic_dict_path, temp)
dictionary.write()
corpus = Corpus(shortsegments_corpus_dir, temp)
corpus.initialize_corpus(dictionary)
fc = FeatureConfig()
fc.generate_features(corpus)
assert len(corpus.feat_mapping.keys()) == 2
assert len(corpus.utt_speak_mapping.keys()) == 3
assert len(corpus.speak_utt_mapping.keys()) == 1
assert len(corpus.text_mapping.keys()) == 3
assert len(corpus.utt_wav_mapping.keys()) == 1
assert len(corpus.segments.keys()) == 3
assert len(corpus.ignored_utterances) == 1
def validate_corpus(args):
if not args.temp_directory:
temp_dir = TEMP_DIR
else:
temp_dir = os.path.expanduser(args.temp_directory)
corpus_name = os.path.basename(args.corpus_directory)
if corpus_name == '':
args.corpus_directory = os.path.dirname(args.corpus_directory)
corpus_name = os.path.basename(args.corpus_directory)
data_directory = os.path.join(temp_dir, corpus_name)
shutil.rmtree(data_directory, ignore_errors=True)
os.makedirs(data_directory, exist_ok=True)
corpus = Corpus(args.corpus_directory,
data_directory,
speaker_characters=args.speaker_characters,
num_jobs=getattr(args, 'num_jobs', 3))
dictionary = Dictionary(args.dictionary_path,
data_directory,
word_set=corpus.word_set)
a = CorpusValidator(corpus,
dictionary,
temp_directory=data_directory,
ignore_acoustics=getattr(args, 'ignore_acoustics',
False),
test_transcriptions=getattr(args,
'test_transcriptions',
False))
a.validate()
def align_corpus(corpus_dir, dict_path, output_directory, temp_dir,
output_model_path, args):
if temp_dir == '':
temp_dir = TEMP_DIR
else:
temp_dir = os.path.expanduser(temp_dir)
corpus_name = os.path.basename(corpus_dir)
if corpus_name == '':
corpus_dir = os.path.dirname(corpus_dir)
corpus_name = os.path.basename(corpus_dir)
data_directory = os.path.join(temp_dir, corpus_name)
if args.clean:
shutil.rmtree(data_directory, ignore_errors = True)
shutil.rmtree(output_directory, ignore_errors = True)
os.makedirs(data_directory, exist_ok = True)
os.makedirs(output_directory, exist_ok = True)
corpus = Corpus(corpus_dir, data_directory, args.speaker_characters, num_jobs = args.num_jobs)
print(corpus.speaker_utterance_info())
corpus.write()
corpus.create_mfccs()
dictionary = Dictionary(dict_path, data_directory, word_set=corpus.word_set)
dictionary.write()
corpus.setup_splits(dictionary)
utt_oov_path = os.path.join(corpus.split_directory, 'utterance_oovs.txt')
if os.path.exists(utt_oov_path):
shutil.copy(utt_oov_path, output_directory)
oov_path = os.path.join(corpus.split_directory, 'oovs_found.txt')
if os.path.exists(oov_path):
shutil.copy(oov_path, output_directory)
mono_params = {'align_often': not args.fast}
tri_params = {'align_often': not args.fast}
tri_fmllr_params = {'align_often': not args.fast}
a = TrainableAligner(corpus, dictionary, output_directory,
temp_directory = data_directory,
mono_params = mono_params, tri_params = tri_params,
tri_fmllr_params = tri_fmllr_params, num_jobs = args.num_jobs)
a.verbose = args.verbose
a.train_mono()
a.export_textgrids()
a.train_tri()
a.export_textgrids()
a.train_tri_fmllr()
a.export_textgrids()
if output_model_path is not None:
a.save(output_model_path)
def test_generate_orthography_dict(basic_corpus_dir, orth_sick_output,
temp_dir):
args = G2PDummyArgs()
args.g2p_model_path = None
args.input_path = basic_corpus_dir
args.output_path = orth_sick_output
args.temp_directory = temp_dir
run_g2p(args)
assert os.path.exists(orth_sick_output)
d = Dictionary(orth_sick_output, temp_dir)
assert len(d.words) > 0
def test_generate_dict(basic_corpus_dir, sick_g2p_model_path, g2p_sick_output,
temp_dir):
args = G2PDummyArgs()
args.g2p_model_path = sick_g2p_model_path
args.input_path = basic_corpus_dir
args.output_path = g2p_sick_output
args.temp_directory = temp_dir
run_g2p(args)
assert os.path.exists(g2p_sick_output)
d = Dictionary(g2p_sick_output, temp_dir)
assert len(d.words) > 0
def train_g2p(args):
if not args.temp_directory:
temp_dir = TEMP_DIR
else:
temp_dir = os.path.expanduser(args.temp_directory)
dictionary = Dictionary(args.dictionary_path, '')
t = PhonetisaurusTrainer(dictionary,
args.output_model_path,
temp_directory=temp_dir,
korean=args.korean)
t.train()
def test_speaker_groupings(large_prosodylab_format_directory, temp_dir, large_dataset_dictionary):
output_directory = os.path.join(temp_dir, 'large')
shutil.rmtree(output_directory, ignore_errors=True)
d = Dictionary(large_dataset_dictionary, output_directory)
d.write()
c = Corpus(large_prosodylab_format_directory, output_directory)
c.initialize_corpus(d)
fc = FeatureConfig()
fc.generate_features(c)
speakers = os.listdir(large_prosodylab_format_directory)
for s in speakers:
assert any(s in x for x in c.speaker_groups)
for root, dirs, files in os.walk(large_prosodylab_format_directory):
for f in files:
name, ext = os.path.splitext(f)
assert any(name in x for x in c.groups)
for root, dirs, files in os.walk(large_prosodylab_format_directory):
for f in files:
name, ext = os.path.splitext(f)
assert any(name in x for x in c.feat_mapping)
shutil.rmtree(output_directory, ignore_errors=True)
d.write()
c = Corpus(large_prosodylab_format_directory, output_directory, num_jobs=2)
c.initialize_corpus(d)
fc.generate_features(c)
for s in speakers:
assert any(s in x for x in c.speaker_groups)
for root, dirs, files in os.walk(large_prosodylab_format_directory):
for f in files:
name, ext = os.path.splitext(f)
assert any(name in x for x in c.groups)
for root, dirs, files in os.walk(large_prosodylab_format_directory):
for f in files:
name, ext = os.path.splitext(f)
assert any(name in x for x in c.feat_mapping)
def train_g2p(args):
if not args.temp_directory:
temp_dir = TEMP_DIR
else:
temp_dir = os.path.expanduser(args.temp_directory)
dictionary = Dictionary(args.dictionary_path, '')
t = PhonetisaurusTrainer(dictionary,
args.output_model_path,
temp_directory=temp_dir,
window_size=args.window_size)
if args.validate:
t.validate()
t.train()
def test_speaker_groupings(large_prosodylab_format_directory, temp_dir,
large_dataset_dictionary):
output_directory = os.path.join(temp_dir, 'large')
shutil.rmtree(output_directory, ignore_errors=True)
d = Dictionary(large_dataset_dictionary, output_directory)
d.write()
c = Corpus(large_prosodylab_format_directory, output_directory)
c.initialize_corpus(d)
fc = FeatureConfig()
fc.generate_features(c)
speakers = os.listdir(large_prosodylab_format_directory)
for s in speakers:
assert any(s in x for x in c.speaker_groups)
for root, dirs, files in os.walk(large_prosodylab_format_directory):
for f in files:
name, ext = os.path.splitext(f)
assert any(name in x for x in c.groups)
for root, dirs, files in os.walk(large_prosodylab_format_directory):
for f in files:
name, ext = os.path.splitext(f)
assert any(name in x for x in c.feat_mapping)
shutil.rmtree(output_directory, ignore_errors=True)
d.write()
c = Corpus(large_prosodylab_format_directory, output_directory, num_jobs=2)
c.initialize_corpus(d)
fc.generate_features(c)
for s in speakers:
assert any(s in x for x in c.speaker_groups)
for root, dirs, files in os.walk(large_prosodylab_format_directory):
for f in files:
name, ext = os.path.splitext(f)
assert any(name in x for x in c.groups)
for root, dirs, files in os.walk(large_prosodylab_format_directory):
for f in files:
name, ext = os.path.splitext(f)
assert any(name in x for x in c.feat_mapping)
def test_extra_annotations(extra_annotations_path, generated_dir):
d = Dictionary(extra_annotations_path,
os.path.join(generated_dir, 'extra'))
assert ('{' in d.graphemes)
d.write()
def align_corpus(args):
all_begin = time.time()
if not args.temp_directory:
temp_dir = TEMP_DIR
else:
temp_dir = os.path.expanduser(args.temp_directory)
corpus_name = os.path.basename(args.corpus_directory)
if corpus_name == "":
args.corpus_directory = os.path.dirname(args.corpus_directory)
corpus_name = os.path.basename(args.corpus_directory)
data_directory = os.path.join(temp_dir, corpus_name)
conf_path = os.path.join(data_directory, "config.yml")
if os.path.exists(conf_path):
with open(conf_path, "r") as f:
conf = yaml.load(f)
else:
conf = {
"dirty": False,
"begin": time.time(),
"version": __version__,
"type": "align",
"corpus_directory": args.corpus_directory,
"dictionary_path": args.dictionary_path,
}
if (
getattr(args, "clean", False)
or conf["dirty"]
or conf["type"] != "align"
or conf["corpus_directory"] != args.corpus_directory
or conf["version"] != __version__
or conf["dictionary_path"] != args.dictionary_path
):
shutil.rmtree(data_directory, ignore_errors=True)
os.makedirs(data_directory, exist_ok=True)
os.makedirs(args.output_directory, exist_ok=True)
try:
corpus = Corpus(
args.corpus_directory,
data_directory,
speaker_characters=args.speaker_characters,
num_jobs=args.num_jobs,
ignore_exceptions=getattr(args, "ignore_exceptions", False),
)
if corpus.issues_check:
print(
"WARNING: Some issues parsing the corpus were detected. "
"Please run the validator to get more information."
)
print(corpus.speaker_utterance_info())
acoustic_model = AcousticModel(args.acoustic_model_path)
dictionary = Dictionary(
args.dictionary_path, data_directory, word_set=corpus.word_set
)
acoustic_model.validate(dictionary)
begin = time.time()
if args.config_path:
align_config = align_yaml_to_config(args.config_path)
else:
align_config = load_basic_align()
a = PretrainedAligner(
corpus,
dictionary,
acoustic_model,
align_config,
args.output_directory,
temp_directory=data_directory,
debug=getattr(args, "debug", False),
)
if getattr(args, "errors", False):
check = a.test_utterance_transcriptions()
if not getattr(args, "quiet", False) and not check:
user_input = input(
"Would you like to abort to fix transcription issues? (Y/N)"
)
if user_input.lower() == "y":
return
if args.debug:
print("Setup pretrained aligner in {} seconds".format(time.time() - begin))
a.verbose = args.verbose
begin = time.time()
a.align()
if args.debug:
print("Performed alignment in {} seconds".format(time.time() - begin))
begin = time.time()
a.export_textgrids()
if args.debug:
print("Exported TextGrids in {} seconds".format(time.time() - begin))
print("Done! Everything took {} seconds".format(time.time() - all_begin))
except:
conf["dirty"] = True
raise
finally:
with open(conf_path, "w") as f:
yaml.dump(conf, f)
def test_basic_noposition(basic_dict_path, generated_dir):
d = Dictionary(basic_dict_path, os.path.join(generated_dir, 'basic'), position_dependent_phones = False)
x = d.write()
assert(set(d.phones) == set(['sil', 'sp','spn', 'phonea','phoneb','phonec']))
def align_corpus(args):
if not args.temp_directory:
temp_dir = TEMP_DIR
else:
temp_dir = os.path.expanduser(args.temp_directory)
corpus_name = os.path.basename(args.corpus_directory)
if corpus_name == '':
args.corpus_directory = os.path.dirname(args.corpus_directory)
corpus_name = os.path.basename(args.corpus_directory)
data_directory = os.path.join(temp_dir, corpus_name)
conf_path = os.path.join(data_directory, 'config.yml')
if os.path.exists(conf_path):
with open(conf_path, 'r') as f:
conf = yaml.load(f)
else:
conf = {
'dirty': False,
'begin': time.time(),
'version': __version__,
'type': 'train_and_align',
'corpus_directory': args.corpus_directory,
'dictionary_path': args.dictionary_path
}
if getattr(args, 'clean', False) \
or conf['dirty'] or conf['type'] != 'train_and_align' \
or conf['corpus_directory'] != args.corpus_directory \
or conf['version'] != __version__ \
or conf['dictionary_path'] != args.dictionary_path:
shutil.rmtree(data_directory, ignore_errors=True)
shutil.rmtree(args.output_directory, ignore_errors=True)
os.makedirs(data_directory, exist_ok=True)
os.makedirs(args.output_directory, exist_ok=True)
try:
corpus = Corpus(args.corpus_directory,
data_directory,
speaker_characters=args.speaker_characters,
num_jobs=getattr(args, 'num_jobs', 3),
debug=getattr(args, 'debug', False),
ignore_exceptions=getattr(args, 'ignore_exceptions',
False))
dictionary = Dictionary(args.dictionary_path,
data_directory,
word_set=corpus.word_set)
utt_oov_path = os.path.join(corpus.split_directory,
'utterance_oovs.txt')
if os.path.exists(utt_oov_path):
shutil.copy(utt_oov_path, args.output_directory)
oov_path = os.path.join(corpus.split_directory, 'oovs_found.txt')
if os.path.exists(oov_path):
shutil.copy(oov_path, args.output_directory)
mono_params = {'align_often': not args.fast}
tri_params = {'align_often': not args.fast}
tri_fmllr_params = {'align_often': not args.fast}
a = TrainableAligner(corpus,
dictionary,
args.output_directory,
temp_directory=data_directory,
mono_params=mono_params,
tri_params=tri_params,
tri_fmllr_params=tri_fmllr_params,
num_jobs=args.num_jobs,
skip_input=getattr(args, 'quiet', False),
nnet=getattr(args, 'artificial_neural_net',
False))
a.verbose = args.verbose
# GMM training (looks like it needs to be done either way, as a starter for nnet)
a.train_mono()
a.export_textgrids()
a.train_tri()
a.export_textgrids()
a.train_tri_fmllr()
a.export_textgrids()
# nnet training
if args.artificial_neural_net:
# Do nnet training
a.train_lda_mllt()
#a.train_diag_ubm() # Uncomment to train i-vector extractor
#a.ivector_extractor() # Uncomment to train i-vector extractor (integrate with argument eventually)
a.train_nnet_basic()
a.export_textgrids()
if args.output_model_path is not None:
a.save(args.output_model_path)
except:
conf['dirty'] = True
raise
finally:
with open(conf_path, 'w') as f:
yaml.dump(conf, f)
# Copy text file
open(os.path.join(corpse_dir_in, '1.lab'), 'w').write(open(TEXT_PATH).read())
corpus = Corpus(corpse_dir_in, corpse_dir_out)
acoustic_model = AcousticModel('spanish.zip')
g2p_model = G2PModel('spanish_g2p.zip')
dict_dir = tempfile.mkdtemp()
with tempfile.NamedTemporaryFile() as g2pfh:
d_gen = PhonetisaurusDictionaryGenerator(g2p_model, WORDS, g2pfh.name)
d_gen.generate()
dictionary = Dictionary(g2pfh.name, dict_dir)
acoustic_model.validate(dictionary)
aligner = PretrainedAligner(corpus,
dictionary,
acoustic_model,
outdir,
temp_directory=corpse_dir_tmp)
check = aligner.test_utterance_transcriptions()
aligner.do_align()
aligner.export_textgrids()
grid = TextGrid.fromFile(os.path.join(outdir, 'in', '1.TextGrid'))
def sick_dict(sick_dict_path, generated_dir):
output_directory = os.path.join(generated_dir, 'sickcorpus')
dictionary = Dictionary(sick_dict_path, output_directory)
dictionary.write()
return dictionary
def align_corpus(args):
if not args.temp_directory:
temp_dir = TEMP_DIR
else:
temp_dir = os.path.expanduser(args.temp_directory)
corpus_name = os.path.basename(args.corpus_directory)
if corpus_name == '':
args.corpus_directory = os.path.dirname(args.corpus_directory)
corpus_name = os.path.basename(args.corpus_directory)
data_directory = os.path.join(temp_dir, corpus_name)
conf_path = os.path.join(data_directory, 'config.yml')
if os.path.exists(conf_path):
with open(conf_path, 'r') as f:
conf = yaml.load(f)
else:
conf = {
'dirty': False,
'begin': time.time(),
'version': __version__,
'type': 'train_and_align',
'corpus_directory': args.corpus_directory,
'dictionary_path': args.dictionary_path
}
if getattr(args, 'clean', False) \
or conf['dirty'] or conf['type'] != 'train_and_align' \
or conf['corpus_directory'] != args.corpus_directory \
or conf['version'] != __version__ \
or conf['dictionary_path'] != args.dictionary_path:
shutil.rmtree(data_directory, ignore_errors=True)
os.makedirs(data_directory, exist_ok=True)
os.makedirs(args.output_directory, exist_ok=True)
try:
corpus = Corpus(args.corpus_directory,
data_directory,
speaker_characters=args.speaker_characters,
num_jobs=getattr(args, 'num_jobs', 3),
debug=getattr(args, 'debug', False),
ignore_exceptions=getattr(args, 'ignore_exceptions',
False))
if corpus.issues_check:
print('WARNING: Some issues parsing the corpus were detected. '
'Please run the validator to get more information.')
dictionary = Dictionary(args.dictionary_path,
data_directory,
word_set=corpus.word_set)
utt_oov_path = os.path.join(corpus.split_directory(),
'utterance_oovs.txt')
if os.path.exists(utt_oov_path):
shutil.copy(utt_oov_path, args.output_directory)
oov_path = os.path.join(corpus.split_directory(), 'oovs_found.txt')
if os.path.exists(oov_path):
shutil.copy(oov_path, args.output_directory)
if args.config_path:
train_config, align_config = train_yaml_to_config(args.config_path)
else:
train_config, align_config = load_basic_train()
a = TrainableAligner(corpus,
dictionary,
train_config,
align_config,
args.output_directory,
temp_directory=data_directory)
a.verbose = args.verbose
a.train()
a.export_textgrids()
if args.output_model_path is not None:
a.save(args.output_model_path)
except:
conf['dirty'] = True
raise
finally:
with open(conf_path, 'w') as f:
yaml.dump(conf, f)
def g2p_gp(lang_code, full_name):
temp_directory = '/data/mmcauliffe/temp/MFA'
dictionary_path = '/media/share/corpora/GP_for_MFA/{0}/dict/{0}_dictionary.txt'.format(
lang_code)
if not os.path.exists(dictionary_path):
print('Skipping {}, no dictionary!'.format(lang_code))
return
output_model_path = '/data/mmcauliffe/aligner-models/g2p/{}_g2p.zip'.format(
full_name)
if os.path.exists(output_model_path):
print('Skipping {}, already a model!'.format(lang_code))
else:
dictionary = Dictionary(dictionary_path, '')
best_acc = 0
best_size = 0
for s in [2, 3, 4]:
begin = time.time()
t = PhonetisaurusTrainer(dictionary,
output_model_path,
temp_directory=temp_directory,
window_size=s)
acc = t.validate()
duration = time.time() - begin
line_dict = {
'Dictionary': dictionary_path,
'Language': lang_code,
'Total time': duration,
'Window size': s,
'Accuracy': acc
}
line_dict.update(dict_data)
with open(csv_path, 'a') as csv_file:
writer = csv.DictWriter(csv_file, fieldnames=csv_columns)
writer.writerow(line_dict)
if acc > best_acc:
best_acc = acc
best_size = s
print('The best window size for {} was {} with accuracy of {}.'.format(
lang_code, best_size, best_acc))
t = PhonetisaurusTrainer(dictionary,
output_model_path,
temp_directory=temp_directory,
window_size=best_size)
t.train()
if lang_code in ['FR', 'GE', 'CH']:
if lang_code == 'FR':
dictionary_path = '/media/share/corpora/GP_for_MFA/{0}/dict/fr.dict'.format(
lang_code)
output_model_path = '/data/mmcauliffe/aligner-models/g2p/{}_prosodylab_g2p.zip'.format(
full_name)
elif lang_code == 'GE':
dictionary_path = '/media/share/corpora/GP_for_MFA/{0}/dict/de.dict'.format(
lang_code)
output_model_path = '/data/mmcauliffe/aligner-models/g2p/{}_prosodylab_g2p.zip'.format(
full_name)
elif lang_code == 'CH':
dictionary_path = '/media/share/corpora/GP_for_MFA/{0}/dict/char_dict.txt'.format(
lang_code)
output_model_path = '/data/mmcauliffe/aligner-models/g2p/{}_character_g2p.zip'.format(
full_name)
if not os.path.exists(dictionary_path):
print('Skipping {}, no dictionary!'.format(lang_code))
return
if os.path.exists(output_model_path):
print('Skipping {}, already a model!'.format(lang_code))
return
temp_directory = '/data/mmcauliffe/temp/MFA'
dictionary = Dictionary(dictionary_path, '')
best_acc = 0
best_size = 0
for s in [2, 3, 4]:
begin = time.time()
t = PhonetisaurusTrainer(dictionary,
output_model_path,
temp_directory=temp_directory,
window_size=s)
acc = t.validate()
duration = time.time() - begin
line_dict = {
'Dictionary': dictionary_path,
'Language': lang_code,
'Total time': duration,
'Window size': s,
'Accuracy': acc
}
line_dict.update(dict_data)
with open(csv_path, 'a') as csv_file:
writer = csv.DictWriter(csv_file, fieldnames=csv_columns)
writer.writerow(line_dict)
if acc > best_acc:
best_acc = acc
best_size = s
print('The best window size for {} was {} with accuracy of {}.'.format(
lang_code, best_size, best_acc))
t = PhonetisaurusTrainer(dictionary,
output_model_path,
temp_directory=temp_directory,
window_size=best_size)
t.train()
def align_corpus(args):
if not args.temp_directory:
temp_dir = TEMP_DIR
else:
temp_dir = os.path.expanduser(args.temp_directory)
corpus_name = os.path.basename(args.corpus_directory)
if corpus_name == "":
args.corpus_directory = os.path.dirname(args.corpus_directory)
corpus_name = os.path.basename(args.corpus_directory)
data_directory = os.path.join(temp_dir, corpus_name)
conf_path = os.path.join(data_directory, "config.yml")
if os.path.exists(conf_path):
with open(conf_path, "r") as f:
conf = yaml.load(f)
else:
conf = {
"dirty": False,
"begin": time.time(),
"version": __version__,
"type": "train_and_align",
"corpus_directory": args.corpus_directory,
"dictionary_path": args.dictionary_path,
}
if (
getattr(args, "clean", False)
or conf["dirty"]
or conf["type"] != "train_and_align"
or conf["corpus_directory"] != args.corpus_directory
or conf["version"] != __version__
or conf["dictionary_path"] != args.dictionary_path
):
shutil.rmtree(data_directory, ignore_errors=True)
os.makedirs(data_directory, exist_ok=True)
os.makedirs(args.output_directory, exist_ok=True)
try:
corpus = Corpus(
args.corpus_directory,
data_directory,
speaker_characters=args.speaker_characters,
num_jobs=getattr(args, "num_jobs", 3),
debug=getattr(args, "debug", False),
ignore_exceptions=getattr(args, "ignore_exceptions", False),
)
if corpus.issues_check:
print(
"WARNING: Some issues parsing the corpus were detected. "
"Please run the validator to get more information."
)
dictionary = Dictionary(
args.dictionary_path, data_directory, word_set=corpus.word_set
)
utt_oov_path = os.path.join(corpus.split_directory(), "utterance_oovs.txt")
if os.path.exists(utt_oov_path):
shutil.copy(utt_oov_path, args.output_directory)
oov_path = os.path.join(corpus.split_directory(), "oovs_found.txt")
if os.path.exists(oov_path):
shutil.copy(oov_path, args.output_directory)
if args.config_path:
train_config, align_config = train_yaml_to_config(args.config_path)
else:
train_config, align_config = load_basic_train()
a = TrainableAligner(
corpus,
dictionary,
train_config,
align_config,
args.output_directory,
temp_directory=data_directory,
)
a.verbose = args.verbose
a.train()
a.export_textgrids()
if args.output_model_path is not None:
a.save(args.output_model_path)
except:
conf["dirty"] = True
raise
finally:
with open(conf_path, "w") as f:
yaml.dump(conf, f)
def align_corpus(args, skip_input=False):
all_begin = time.time()
if not args.temp_directory:
temp_dir = TEMP_DIR
else:
temp_dir = os.path.expanduser(args.temp_directory)
corpus_name = os.path.basename(args.corpus_directory)
if corpus_name == '':
args.corpus_directory = os.path.dirname(args.corpus_directory)
corpus_name = os.path.basename(args.corpus_directory)
data_directory = os.path.join(temp_dir, corpus_name)
conf_path = os.path.join(data_directory, 'config.yml')
if os.path.exists(conf_path):
with open(conf_path, 'r') as f:
conf = yaml.load(f)
else:
conf = {'dirty': False,
'begin': time.time(),
'version': __version__,
'type': 'align',
'corpus_directory': args.corpus_directory,
'dictionary_path': args.dictionary_path}
if getattr(args, 'clean', False) \
or conf['dirty'] or conf['type'] != 'align' \
or conf['corpus_directory'] != args.corpus_directory\
or conf['version'] != __version__\
or conf['dictionary_path'] != args.dictionary_path:
shutil.rmtree(data_directory, ignore_errors=True)
shutil.rmtree(args.output_directory, ignore_errors=True)
os.makedirs(data_directory, exist_ok=True)
os.makedirs(args.output_directory, exist_ok=True)
use_speaker_info = not args.no_speaker_adaptation
try:
corpus = Corpus(args.corpus_directory, data_directory,
speaker_characters=args.speaker_characters,
num_jobs=args.num_jobs,
use_speaker_information=use_speaker_info,
ignore_exceptions=getattr(args, 'ignore_exceptions', False))
print(corpus.speaker_utterance_info())
acoustic_model = AcousticModel(args.acoustic_model_path)
dictionary = Dictionary(args.dictionary_path, data_directory, word_set=corpus.word_set)
acoustic_model.validate(dictionary)
begin = time.time()
a = PretrainedAligner(corpus, dictionary, acoustic_model, args.output_directory, temp_directory=data_directory,
num_jobs=getattr(args, 'num_jobs', 3),
speaker_independent=getattr(args, 'no_speaker_adaptation', False),
debug=getattr(args, 'debug', False))
if getattr(args, 'errors', False):
check = a.test_utterance_transcriptions()
if not skip_input and not check:
user_input = input('Would you like to abort to fix transcription issues? (Y/N)')
if user_input.lower() == 'y':
return
if args.debug:
print('Setup pretrained aligner in {} seconds'.format(time.time() - begin))
a.verbose = args.verbose
utt_oov_path = os.path.join(corpus.split_directory, 'utterance_oovs.txt')
if os.path.exists(utt_oov_path):
shutil.copy(utt_oov_path, args.output_directory)
oov_path = os.path.join(corpus.split_directory, 'oovs_found.txt')
if os.path.exists(oov_path):
shutil.copy(oov_path, args.output_directory)
if not skip_input and a.dictionary.oovs_found:
user_input = input(
'There were words not found in the dictionary. Would you like to abort to fix them? (Y/N)')
if user_input.lower() == 'y':
return
begin = time.time()
a.do_align()
if args.debug:
print('Performed alignment in {} seconds'.format(time.time() - begin))
begin = time.time()
a.export_textgrids()
if args.debug:
print('Exported TextGrids in {} seconds'.format(time.time() - begin))
print('Done! Everything took {} seconds'.format(time.time() - all_begin))
except:
conf['dirty'] = True
raise
finally:
with open(conf_path, 'w') as f:
yaml.dump(conf, f)
def align_corpus(args, skip_input=False):
if not args.temp_directory:
temp_dir = TEMP_DIR
else:
temp_dir = os.path.expanduser(args.temp_directory)
corpus_name = os.path.basename(args.corpus_directory)
if corpus_name == '':
args.corpus_directory = os.path.dirname(args.corpus_directory)
corpus_name = os.path.basename(args.corpus_directory)
data_directory = os.path.join(temp_dir, corpus_name)
conf_path = os.path.join(data_directory, 'config.yml')
if os.path.exists(conf_path):
with open(conf_path, 'r') as f:
conf = yaml.load(f)
else:
conf = {
'dirty': False,
'begin': time.time(),
'version': __version__,
'type': 'train_and_align',
'corpus_directory': args.corpus_directory,
'dictionary_path': args.dictionary_path
}
if getattr(args, 'clean', False) \
or conf['dirty'] or conf['type'] != 'train_and_align' \
or conf['corpus_directory'] != args.corpus_directory \
or conf['version'] != __version__ \
or conf['dictionary_path'] != args.dictionary_path:
shutil.rmtree(data_directory, ignore_errors=True)
shutil.rmtree(args.output_directory, ignore_errors=True)
os.makedirs(data_directory, exist_ok=True)
os.makedirs(args.output_directory, exist_ok=True)
try:
corpus = Corpus(args.corpus_directory,
data_directory,
speaker_characters=args.speaker_characters,
num_jobs=getattr(args, 'num_jobs', 3),
debug=getattr(args, 'debug', False),
ignore_exceptions=getattr(args, 'ignore_exceptions',
False))
dictionary = Dictionary(args.dictionary_path,
data_directory,
word_set=corpus.word_set)
utt_oov_path = os.path.join(corpus.split_directory,
'utterance_oovs.txt')
if os.path.exists(utt_oov_path):
shutil.copy(utt_oov_path, args.output_directory)
oov_path = os.path.join(corpus.split_directory, 'oovs_found.txt')
if os.path.exists(oov_path):
shutil.copy(oov_path, args.output_directory)
mono_params = {'align_often': not args.fast}
tri_params = {'align_often': not args.fast}
tri_fmllr_params = {'align_often': not args.fast}
a = TrainableAligner(corpus,
dictionary,
args.output_directory,
temp_directory=data_directory,
mono_params=mono_params,
tri_params=tri_params,
tri_fmllr_params=tri_fmllr_params,
num_jobs=args.num_jobs)
a.verbose = args.verbose
a.train_mono()
a.export_textgrids()
a.train_tri()
a.export_textgrids()
a.train_tri_fmllr()
a.export_textgrids()
if args.output_model_path is not None:
a.save(args.output_model_path)
except:
conf['dirty'] = True
raise
finally:
with open(conf_path, 'w') as f:
yaml.dump(conf, f)
def align_corpus(args):
all_begin = time.time()
if not args.temp_directory:
temp_dir = TEMP_DIR
else:
temp_dir = os.path.expanduser(args.temp_directory)
corpus_name = os.path.basename(args.corpus_directory)
if corpus_name == '':
args.corpus_directory = os.path.dirname(args.corpus_directory)
corpus_name = os.path.basename(args.corpus_directory)
data_directory = os.path.join(temp_dir, corpus_name)
conf_path = os.path.join(data_directory, 'config.yml')
if os.path.exists(conf_path):
with open(conf_path, 'r') as f:
conf = yaml.load(f, Loader=yaml.SafeLoader)
else:
conf = {
'dirty': False,
'begin': time.time(),
'version': __version__,
'type': 'align',
'corpus_directory': args.corpus_directory,
'dictionary_path': args.dictionary_path
}
if getattr(args, 'clean', False) \
or conf['dirty'] or conf['type'] != 'align' \
or conf['corpus_directory'] != args.corpus_directory \
or conf['version'] != __version__ \
or conf['dictionary_path'] != args.dictionary_path:
shutil.rmtree(data_directory, ignore_errors=True)
os.makedirs(data_directory, exist_ok=True)
os.makedirs(args.output_directory, exist_ok=True)
try:
corpus = Corpus(args.corpus_directory,
data_directory,
speaker_characters=args.speaker_characters,
num_jobs=args.num_jobs,
ignore_exceptions=getattr(args, 'ignore_exceptions',
False))
if corpus.issues_check:
print('WARNING: Some issues parsing the corpus were detected. '
'Please run the validator to get more information.')
print(corpus.speaker_utterance_info())
acoustic_model = AcousticModel(args.acoustic_model_path)
dictionary = Dictionary(args.dictionary_path,
data_directory,
word_set=corpus.word_set)
acoustic_model.validate(dictionary)
begin = time.time()
if args.config_path:
align_config = align_yaml_to_config(args.config_path)
else:
align_config = load_basic_align()
a = PretrainedAligner(corpus,
dictionary,
acoustic_model,
align_config,
args.output_directory,
temp_directory=data_directory,
debug=getattr(args, 'debug', False))
if args.debug:
print('Setup pretrained aligner in {} seconds'.format(time.time() -
begin))
a.verbose = args.verbose
begin = time.time()
a.align()
if args.debug:
print('Performed alignment in {} seconds'.format(time.time() -
begin))
begin = time.time()
a.export_textgrids()
if args.debug:
print('Exported TextGrids in {} seconds'.format(time.time() -
begin))
print('Done! Everything took {} seconds'.format(time.time() -
all_begin))
except:
conf['dirty'] = True
raise
finally:
with open(conf_path, 'w') as f:
yaml.dump(conf, f)
def test_frclitics(frclitics_dict_path, generated_dir):
d = Dictionary(frclitics_dict_path, os.path.join(generated_dir, 'frclitics'))
x = d.write()
assert d.separate_clitics('aujourd') == ['aujourd']
assert d.separate_clitics('aujourd\'hui') == ['aujourd\'hui']
assert d.separate_clitics('vingt-six') == ['vingt', 'six']
assert d.separate_clitics('m\'appelle') == ['m\'', 'appelle']
assert d.separate_clitics('c\'est') == ['c\'est']
assert d.separate_clitics('purple-people-eater') == ['purple-people-eater']
assert d.separate_clitics('m\'appele') == ['m\'', 'appele']
assert d.separate_clitics('m\'ving-sic') == ["m'", 'ving', 'sic']
assert d.separate_clitics('flying\'purple-people-eater') == ['flying\'purple-people-eater']
def test_frclitics(frclitics_dict_path, generated_dir):
d = Dictionary(frclitics_dict_path, os.path.join(generated_dir,
'frclitics'))
x = d.write()
assert d.separate_clitics('aujourd') == ['aujourd']
assert d.separate_clitics('aujourd\'hui') == ['aujourd\'hui']
assert d.separate_clitics('vingt-six') == ['vingt', 'six']
assert d.separate_clitics('m\'appelle') == ['m\'', 'appelle']
assert d.separate_clitics('c\'est') == ['c\'est']
assert d.separate_clitics('purple-people-eater') == ['purple-people-eater']
assert d.separate_clitics('m\'appele') == ['m\'', 'appele']
assert d.separate_clitics('m\'ving-sic') == ["m'", 'ving', 'sic']
assert d.separate_clitics('flying\'purple-people-eater') == [
'flying\'purple-people-eater'
]
def test_extra_annotations(extra_annotations_path, generated_dir):
d = Dictionary(extra_annotations_path, os.path.join(generated_dir, 'extra'))
assert('{' in d.graphemes)
d.write()
def sick_dict(sick_dict_path, generated_dir):
output_directory = os.path.join(generated_dir, 'sickcorpus')
dictionary = Dictionary(sick_dict_path, output_directory)
dictionary.write()
return dictionary
