def run_gp(mapping,crf,lm):
"""
Run the g2p server.
"""
#
# loading
#
click.echo(u"Loading...", err=True)
click.echo(u"...data alignment", err=True)
aligner = gp.Aligner(mapping=mapping)
click.echo(u"...transcription CRF model", err=True)
transcriber = gp.Transcriber()
transcriber.load(crf)
click.echo(u"...n-gram language model", err=True)
rater = gp.Rater.load(lm)
click.echo(u"...output formatter", err=True)
formatter = gp.Formatter()
#
# load app and run
#
app = apps.create_gp_app(aligner,transcriber,rater,formatter)
app.run()
def apply_gp(mapping, crf, lm, strings):
"""Convert strings"""
#
# loading
#
click.echo("Loading...", err=True)
click.echo("...data alignment", err=True)
aligner = gp.Aligner(mapping=mapping)
click.echo("...transcription CRF model", err=True)
transcriber = gp.Transcriber()
transcriber.load(crf)
click.echo("...n-gram language model", err=True)
rater = gp.Rater.load(lm)
#
# conversion
#
# read input
in_strings = []
if strings and strings[0] == "-":
for line in sys.stdin:
in_strings.append(line.strip())
elif strings:
for datum in strings:
in_strings.append(datum)
else:
pass
# convert
for string in in_strings:
segmentations = aligner.scan(string.lower())
best_transcription = []
best_prob = 0.0
for segmentation in segmentations:
transcriptions = transcriber.transcribe(segmentation)
for transcription in transcriptions:
prob = rater.rate([segmentation, transcription])
#click.echo("%s: %f" % (",".join(transcription),prob), err=True)
if prob >= best_prob:
best_prob = prob
best_transcription = transcription
click.echo(",".join(best_transcription))
def train_gp(mapping, model, data):
"""Train a model."""
#
# stage 1: alignment
#
click.echo("Stage 1a: creating data alignment", err=True)
# the aligner
aligner = gp.Aligner(mapping=mapping)
click.echo("Stage 1b: aligning training data", err=True)
# iterate over input and align training data
aligned_training_data = []
with open(str(data), "r") as f:
training_data = f.read()
for line in tqdm(training_data.split("\n")):
# skip comments
if line.startswith("#"):
continue
# assume tab-separated values
fields = line.split("\t")
if len(fields) < 2:
continue
# align
alignment = aligner.align(fields[0], fields[1])
if alignment:
aligned_training_data.append(alignment)
else:
click.echo("%s and %s could not be aligned." %
(fields[0], fields[1]))
#
# stage 2: crf training
#
click.echo("Stage 2: training transcription CRF model", err=True)
# the transcriber
transcriber = gp.Transcriber()
# train with previously aligned training data
transcriber.train(aligned_training_data)
# save
transcriber.save(model + ".gp.crf")
#
# stage 3: language model training
#
click.echo("Stage 3: training rating n-gram language model", err=True)
# the rater
rater = gp.Rater()
# train with previously aligned training data
rater.train(aligned_training_data)
# save
rater.save(model + ".gp.ngram")
def test_scan(datadir):
aligner = gp.Aligner(mapping=datadir.join('test_alignment.txt'))
segmentations = aligner.scan(u"aabb")
assert (segmentations == [['a', 'a', 'b', 'b'], ['aa', 'b', 'b']])
def test_align(datadir):
aligner = gp.Aligner(mapping=datadir.join('test_alignment.txt'))
alignment = aligner.align(u"aabb", u"abbbb")
assert (alignment[0] == ['aa', 'b', 'b'])
assert (alignment[1] == ['a', 'bb', 'bb'])
def test_expand(datadir):
aligner = gp.Aligner(mapping=datadir.join('test_alignment.txt'))
exp_fst = aligner.expand(u"aabb")
exp_fst.draw('/tmp/exp.dot')
assert (exp_fst.verify())
def test_scan(datadir):
aligner = gp.Aligner(mapping=datadir.join('test_alignment.txt'))
seg_fst = aligner.segment(u"aabb")
seg_fst.draw('/tmp/seg.dot')
assert (seg_fst.verify())
def test_chain(datadir):
aligner = gp.Aligner(mapping=datadir.join('test_alignment.txt'))
chain_fst = aligner.chain(u"aabb")
chain_fst.draw('/tmp/chain.dot')
assert (chain_fst.verify())
def test_loading(datadir):
aligner = gp.Aligner(mapping=datadir.join('test_alignment.txt'))
assert (aligner.status == 1)
def test_constructor():
aligner = gp.Aligner()
assert (aligner != None)