def test_fast():
"""
A fast integration test that runs 1 training epoch over a tiny
dataset. Note that this does not run ffmpeg to normalize the WAVs since
Travis doesn't have that installed. So the normalized wavs are included in
the feat/ directory so that the normalization isn't run.
"""
# 4 utterance toy set
TINY_EXAMPLE_LINK = "https://cloudstor.aarnet.edu.au/plus/s/g2GreDNlDKUq9rz/download"
tiny_example_dir = join(DATA_BASE_DIR, "tiny_example/")
rm_dir(Path(tiny_example_dir))
download_example_data(TINY_EXAMPLE_LINK)
corp = corpus.ReadyCorpus(tiny_example_dir)
exp_dir = run.prep_exp_dir(directory=EXP_BASE_DIR)
model = run.get_simple_model(exp_dir, corp)
model.train(min_epochs=2, max_epochs=5)
# Assert the convergence of the model at the end by reading the test scores
ler = get_test_ler(exp_dir)
# Can't expect a decent test score but just check that there's something.
assert ler < 2.0
def test_full_na():
""" A full Na integration test. """
# Pulls Na wavs from cloudstor.
NA_WAVS_LINK = "https://cloudstor.aarnet.edu.au/plus/s/LnNyNa20GQ8qsPC/download"
download_example_data(NA_WAVS_LINK)
na_dir = join(DATA_BASE_DIR, "na/")
os.rm_dir(na_dir)
os.makedirs(na_dir)
org_wav_dir = join(na_dir, "org_wav/")
os.rename(join(DATA_BASE_DIR, "na_wav/"), org_wav_dir)
tgt_wav_dir = join(na_dir, "wav/")
NA_REPO_URL = "https://github.com/alexis-michaud/na-data.git"
with cd(DATA_BASE_DIR):
subprocess.run(["git", "clone", NA_REPO_URL, "na/xml/"], check=True)
# Note also that this subdirectory only containts TEXTs, so this integration
# test will include only Na narratives, not wordlists.
na_xml_dir = join(DATA_BASE_DIR, "na/xml/TEXT/F4")
label_dir = join(DATA_BASE_DIR, "na/label")
label_type = "phonemes_and_tones"
na.prepare_labels(label_type,
org_xml_dir=na_xml_dir, label_dir=label_dir)
tgt_feat_dir = join(DATA_BASE_DIR, "na/feat")
# TODO Make this fbank_and_pitch, but then I need to install kaldi on ray
# or run the tests on GPUs on slug or doe.
feat_type = "fbank"
na.prepare_feats(feat_type,
org_wav_dir=org_wav_dir,
tgt_wav_dir=tgt_wav_dir,
feat_dir=tgt_feat_dir,
org_xml_dir=na_xml_dir,
label_dir=label_dir)
from shutil import copyfile
copyfile("persephone/tests/test_sets/valid_prefixes.txt",
join(na_dir, "valid_prefixes.txt"))
copyfile("persephone/tests/test_sets/test_prefixes.txt",
join(na_dir, "test_prefixes.txt"))
na.make_data_splits(label_type, train_rec_type="text", tgt_dir=na_dir)
# Training with texts
exp_dir = run.prep_exp_dir(directory=EXP_BASE_DIR)
na_corpus = na.Corpus(feat_type, label_type,
train_rec_type="text",
tgt_dir=na_dir)
na_corpus_reader = corpus_reader.CorpusReader(na_corpus)
model = rnn_ctc.Model(exp_dir, na_corpus_reader,
num_layers=3, hidden_size=400)
model.train(min_epochs=30)
# Ensure LER < 0.20
ler = get_test_ler(exp_dir)
assert ler < 0.2
def train_bkw(num_layers: int) -> None:
exp_dir = prep_exp_dir(directory=config.TEST_EXP_PATH)
corp = bkw.create_corpus(tgt_dir=Path(config.TEST_DATA_PATH) / "bkw")
cr = CorpusReader(corp)
model = rnn_ctc.Model(exp_dir,
cr,
num_layers=num_layers,
hidden_size=250)
model.train(min_epochs=40)
def test_multispeaker(self, preprocessed_corpus):
""" Trains a multispeaker BKW system using default settings. """
exp_dir = prep_exp_dir(directory=config.TEST_EXP_PATH)
# TODO bkw.Corpus and elan.Corpus should take an org_dir argument.
corp = preprocessed_corpus
cr = CorpusReader(corp)
model = rnn_ctc.Model(exp_dir, cr, num_layers=2, hidden_size=250)
model.train(min_epochs=30)