def error_chars(transcription, ground_truth, missing_char="_"):
alignment = sequence_alignment.levenshtein_alignment(
list(transcription), list(ground_truth))
source_transcription = ""
target_transcription = ""
errors_source = []
errors_target = []
for (source_char, target_char) in alignment:
if source_char != target_char:
color_target = False
if source_char is None:
source_char = missing_char
if target_char is None:
target_char = missing_char
color_target = True
errors_source.append(True)
errors_target.append(color_target)
else:
errors_source.append(False)
errors_target.append(False)
source_transcription += source_char
target_transcription += target_char
return source_transcription, target_transcription, errors_source, errors_target
def console_transcription_errors(transcription,
ground_truth,
color=Color.RED,
missing_char="_"):
alignment = sequence_alignment.levenshtein_alignment(
list(transcription), list(ground_truth))
source_transcription = ""
target_transcription = ""
for pair in alignment:
source_char, target_char = pair
if source_char != target_char:
color_target = False
if source_char is None:
source_char = missing_char
if target_char is None:
target_char = missing_char
color_target = True
source_char = color.color_console_text(source_char)
if color_target:
target_char = color.color_console_text(target_char)
source_transcription += source_char
target_transcription += target_char
return source_transcription, target_transcription
def test_inner_replacement_rev(self):
a = ['a', 'x', 'y', 'c']
b = ['a', 'b', 'c']
self.assertTrue(
levenshtein_alignment(a, b) in [
[('a', 'a'), ('x', None), ('y', 'b'), ('c', 'c')],
[('a', 'a'), ('x', 'b'), ('y', None), ('c', 'c')],
])
def from_lists(cls, ref, hyp):
ref_len = len(ref)
nb_errors = levenshtein_distance(ref, hyp)
alignment = levenshtein_alignment(hyp, ref)
_, _, nb_inss, nb_dels, nb_subs = edit_stats_for_alignment(alignment)
confusions = defaultdict(Counter)
for hyp_sym, ref_sym in alignment:
confusions[ref_sym][hyp_sym] += 1
match_types = [get_match_type(a[1], a[0]) for a in alignment]
ending_mistakes = get_non_matching_suffix(match_types)
end_errors = BoundaryErrorsSummary(ending_mistakes)
return cls(1, ref_len, nb_errors, nb_subs, nb_inss, nb_dels,
confusions, end_errors)
def test_insertion_only(self):
a = []
b = ['a', 'b', 'c']
self.assertEqual(levenshtein_alignment(a, b),
[(None, 'a'), (None, 'b'), (None, 'c')])
def test_deletion_only(self):
a = ['a', 'b', 'c']
b = []
self.assertEqual(levenshtein_alignment(a, b),
[('a', None), ('b', None), ('c', None)])
def test_trivial_deletion(self):
a = ['a', 'b']
b = ['a']
self.assertEqual(levenshtein_alignment(a, b), [('a', 'a'),
('b', None)])
def test_trivial_insertion(self):
a = ['a']
b = ['b', 'a']
self.assertEqual(levenshtein_alignment(a, b), [(None, 'b'),
('a', 'a')])
def test_trivial_substitution(self):
a = ['a']
b = ['b']
self.assertEqual(levenshtein_alignment(a, b), [('a', 'b')])
def test_trivial_match(self):
a = ['a']
b = ['a']
self.assertEqual(levenshtein_alignment(a, b), [('a', 'a')])
def test_alignment_to_eps_rev(self):
a = ['a', 'b', 'c']
b = ['a', None, 'c']
self.assertEqual(levenshtein_alignment(a, b), [('a', 'a'), ('b', None),
('c', 'c')])