def test_tiered_composition(self):
transducer = make_g2p("dan", "eng-arpabet")
tg = transducer("hej")
self.assertEqual(tg.output_string, "HH EH Y")
self.assertEqual(
tg.edges,
[
[(0, 0), (1, 1), (2, 2)],
[(0, 0), (1, 1), (2, 2)],
[(0, 0), (0, 1), (0, 2), (1, 3), (1, 4), (1, 5), (2, 6)],
],
)
self.assertEqual(
tg.pretty_edges(),
[
[["h", "h"], ["e", "ɛ"], ["j", "j"]],
[["h", "h"], ["ɛ", "ɛ"], ["j", "j"]],
[
["h", "H"],
["h", "H"],
["h", " "],
["ɛ", "E"],
["ɛ", "H"],
["ɛ", " "],
["j", "Y"],
],
],
)
self.assertEqual(compose_tiers(tg.edges), [(0, 2), (1, 5), (2, 6)])
def test_composition_with_none(self):
transducer = make_g2p("ctp", "eng-arpabet")
tg = transducer("Qne\u1D2C")
self.assertEqual(tg.output_string, "HH N EY")
self.assertEqual(
tg.edges,
[
[(0, 0), (1, 1), (2, 2), (3, None)],
[(0, 0), (1, 1), (2, 2), (2, 3)],
[(0, 0), (0, 1), (0, 2), (1, 3), (1, 4), (2, 5), (3, 6)],
],
)
self.assertEqual(
tg.pretty_edges(),
[
[["q", "ʔ"], ["n", "n"], ["e", "e"], ["ᴬ", None]],
[["ʔ", "ʔ"], ["n", "n"], ["e", "e"], ["e", "ː"]],
[
["ʔ", "H"],
["ʔ", "H"],
["ʔ", " "],
["n", "N"],
["n", " "],
["e", "E"],
["ː", "Y"],
],
],
)
self.assertEqual(compose_tiers(tg.edges), [(0, 2), (1, 4), (2, 6),
(3, 6)])
def test_fra(self):
transducer = make_g2p("fra", "eng-arpabet")
tg = transducer("mais")
self.assertEqual(tg.output_string, "M EH")
self.assertEqual(compose_tiers(increment_tiers(tg.edges)), [(1, 2),
(2, 4),
(3, 4),
(4, 4)])
def convert_words(xml, word_unit="w", output_orthography="eng-arpabet"):
for word in xml.xpath(".//" + word_unit):
# only convert text within words
same_language_units = get_same_language_units(word)
if not same_language_units:
return
all_text = ""
all_indices = []
for unit in same_language_units:
# Hack to use old English LexiconG2P
if unit["lang"] != "eng":
converter = make_g2p(unit["lang"], output_orthography)
tg = converter(unit["text"])
text = tg.output_string
indices = tg.edges
else:
tg = False
converter = LexiconG2P(
os.path.join(
os.path.dirname(LEXICON_PATH), "cmu_sphinx.metadata.json"
)
)
text, indices = converter.convert(unit["text"])
all_text += text
all_indices += indices
if tg and isinstance(tg, CompositeTransductionGraph):
norm_form = converter._transducers[0].norm_form
indices = increment_tiers(indices)
all_indices = compose_tiers(indices)
elif tg and isinstance(tg, TransductionGraph):
norm_form = converter.norm_form
indices = increment_indices(indices)
all_indices = compose_indices([], indices)
else:
norm_form = None
all_indices = indices
if norm_form:
word.text = ud.normalize(norm_form, word.text)
replace_text_in_node(word, all_text, all_indices)
return xml