SPPAS Automatic Other-Repetition Detection.
Detect automatically other-repetitions. Result must be re-filtered by an
expert. This annotation is performed on the basis of time-aligned tokens
or lemmas. The output is made of 2 tiers with sources and echos.
Create a new sppasOtherRepet instance.
Log is used for a better communication of the annotation process and its
results. If None, logs are redirected to the default logging system.
def __init__(self, log=None):
"""Create a new sppasOtherRepet instance.
Log is used for a better communication of the annotation process and its
results. If None, logs are redirected to the default logging system.
:param log: (sppasLog) Human-readable logs.
"""
super(sppasOtherRepet, self).__init__('otherrepet.json', log)
self.max_span = 12
Fix all options.
def fix_options(self, options):
"""Fix all options.
:param options: list of sppasOption instances
"""
for opt in options:
key = opt.get_key()
if 'stopwords' == key:
self.set_use_stopwords(opt.get_value())
elif 'span' == key:
self.set_span(opt.get_value())
elif 'alpha' == key:
self.set_alpha(opt.get_value())
elif 'allechos' == key:
self.set_all_echos_tier(opt.get_value())
elif 'pattern' in key:
self._options[key] = opt.get_value()
else:
raise AnnotationOptionError(key)
Create a tier with all tokens that are echo-candidates.
def set_all_echos_tier(self, all_echos):
"""Create a tier with all tokens that are echo-candidates.
:param all_echos: (bool)
"""
self._options['allechos'] = bool(all_echos)
Other-Repetition detection.
def other_detection(self, inputtier1, inputtier2):
"""Other-Repetition detection.
:param inputtier1: (Tier)
:param inputtier2: (Tier)
"""
if inputtier1.is_float():
inputtier1.set_radius(0.04)
if inputtier2.is_float():
inputtier2.set_radius(0.04)
stop_words = self._stop_words.copy()
stop_words.evaluate(inputtier2, merge=True)
repetition = OtherRepetition(stop_words)
trs_output = sppasTranscription(self.name)
trs_output.create_tier('OR-Source')
trs_output.create_tier('OR-SrcStrain')
trs_output.create_tier('OR-SrcLen')
trs_output.create_tier('OR-SrcType')
trs_output.create_tier('OR-Echo')
if self._options['allechos'] is True:
trs_output.create_tier('OR-AllEchos')
tok_start_src = 0
tok_end_src = min(20, len(inputtier1) - 1)
tok_start_echo = 0
tokens2 = list()
speaker2 = DataSpeaker(tokens2)
while tok_start_src < tok_end_src:
tokens1 = [serialize_labels(inputtier1[i].get_labels()) for i in range(tok_start_src, tok_end_src + 1)]
speaker1 = DataSpeaker(tokens1)
src_begin = inputtier1[tok_start_src].get_lowest_localization().get_midpoint()
echo_begin = inputtier2[tok_start_echo].get_lowest_localization().get_midpoint()
if len(tokens2) == 0 or echo_begin < src_begin:
tokens2 = list()
nb_breaks = 0
init_tok_start_echo = tok_start_echo
tok_start_echo = -1
for i in range(init_tok_start_echo, len(inputtier2)):
ann = inputtier2[i]
if ann.get_lowest_localization().get_midpoint() >= src_begin:
label = serialize_labels(ann.get_labels())
if tok_start_echo == -1:
tok_start_echo = i
if label == SIL_ORTHO:
nb_breaks += 1
if nb_breaks == self._options['span']:
break
tokens2.append(label)
speaker2 = DataSpeaker(tokens2)
repetition.detect(speaker1, speaker2, 1)
shift = 1
if repetition.get_source() is not None:
s, e = repetition.get_source()
saved = sppasOtherRepet.__add_repetition(repetition, inputtier1, inputtier2, tok_start_src, tok_start_echo, trs_output)
if saved is True:
shift = e + 1
tok_start_src = min(tok_start_src + shift, len(inputtier1) - 1)
tok_end_src = min(tok_start_src + 20, len(inputtier1) - 1)
return trs_output
Return 2 tiers with aligned tokens.
NoTierInputError
def get_inputs(self, input_files):
"""Return 2 tiers with aligned tokens.
:param input_files: (list)
:raise: NoTierInputError
:return: (sppasTier)
"""
tier_src = None
for filename in input_files[0]:
parser = sppasTrsRW(filename)
trs_input = parser.read()
if tier_src is None:
tier_src = sppasFindTier.aligned_tokens(trs_input)
if tier_src is None:
logging.error('A source tier with time-aligned tokens was expected but not found.')
raise NoTierInputError
tier_echo = None
for filename in input_files[1]:
parser = sppasTrsRW(filename)
trs_input = parser.read()
if tier_echo is None:
tier_echo = sppasFindTier.aligned_tokens(trs_input)
if tier_echo is not None:
min_time_point = trs_input.get_min_loc()
max_time_point = trs_input.get_max_loc()
tier_echo = fill_gaps(tier_echo, min_time_point, max_time_point)
if tier_echo is None:
logging.error('An echo tier with time-aligned tokens was expected but not found.')
raise NoTierInputError
return (tier_src, tier_echo)
Run the automatic annotation process on an input.
Input file is a tuple with 2 files: the main speaker and the echoing
speaker.
def run(self, input_files, output=None):
"""Run the automatic annotation process on an input.
Input file is a tuple with 2 files: the main speaker and the echoing
speaker.
:param input_files: (list of str) File(s) with time-aligned token
:param output: (str) the output name
:returns: (sppasTranscription)
"""
tier_tokens_src, tier_tokens_echo = self.get_inputs(input_files)
s = 0
for ann in tier_tokens_src:
if ann.is_labelled():
if ann.get_best_tag().is_silence() is True:
s += 1
if float(s) / float(len(tier_tokens_src)) < 0.05:
logging.error("Error. The tier with tokens (source) should contain silences but it doesn't.To detect the repetitions, the Inter Pausal Units are required tofix the span. Without silences, there's no way to know where theIPUs are...")
raise ValueError('Invalid source input tier {:s}: no silences.'.format(tier_tokens_src.get_name()))
tier_input1 = self.make_word_strain(tier_tokens_src)
tier_input1.set_name(tier_input1.get_name() + '-source')
s = 0
for ann in tier_tokens_echo:
if ann.is_labelled():
if ann.get_best_tag().is_silence() is True:
s += 1
content = ann.get_best_tag().get_content()
if len(content) == 0:
ann.set_labels([sppasLabel(sppasTag('#'))])
else:
ann.set_labels([sppasLabel(sppasTag('#'))])
if float(s) / float(len(tier_tokens_echo)) < 0.05:
logging.error("Error. The tier with tokens (echo) should contain silences but it doesn't. To detect the repetitions, the Inter Pausal Units are required to fix the span. Without silences, there's no way to know where the IPUs are...")
raise ValueError('Invalid echo input tier {:s}: no silences.'.format(tier_tokens_echo.get_name()))
tier_input2 = self.make_word_strain(tier_tokens_echo)
tier_input2.set_name(tier_input2.get_name() + '-echo')
trs_output = self.other_detection(tier_input1, tier_input2)
trs_output.set_meta('annotation_result_of', input_files[0][0])
if len(self._word_strain) > 0:
trs_output.append(tier_input1)
if self._options['stopwords'] is True:
trs_output.append(self.make_stop_words(tier_input1))
if len(self._word_strain) > 0:
trs_output.append(tier_input2)
if output is not None:
if len(trs_output) > 0:
output_file = self.fix_out_file_ext(output)
parser = sppasTrsRW(output_file)
parser.write(trs_output)
return [output_file]
else:
raise EmptyOutputError
return trs_output
Pattern this annotation uses in an output filename.
def get_output_pattern(self):
"""Pattern this annotation uses in an output filename."""
return self._options.get('outputpattern', '-orepet')
Pattern this annotation expects for its input filename.
def get_input_patterns(self):
"""Pattern this annotation expects for its input filename."""
return [self._options.get('inputpattern', '-palign')]
Add a repetition - source and echos - in tiers.
@staticmethod
def __add_repetition(repetition, spk1_tier, spk2_tier, start_idx1, start_idx2, trs_out):
"""Add a repetition - source and echos - in tiers.
:param repetition: (DataRepetition)
:param spk1_tier: (Tier) The tier of speaker 1 (to detect sources)
:param spk2_tier: (Tier) The tier of speaker 2 (to detect echos)
:param start_idx1: start index of the interval in spk1_tier
:param start_idx2: start index of the interval in spk2_tier
:param trs_out: (sppasTranscription) The resulting tiers
:returns: (bool) the repetition was added or not
"""
src_tier = trs_out.find('OR-Source')
echo_tier = trs_out.find('OR-Echo')
or_index = len(src_tier)
s, e = repetition.get_source()
src_begin = spk1_tier[start_idx1 + s].get_lowest_localization()
src_end = spk1_tier[start_idx1 + e].get_highest_localization()
iitime = sppasInterval(src_begin.copy(), src_end.copy())
try:
a = src_tier.create_annotation(sppasLocation(iitime), sppasLabel(sppasTag('S' + str(or_index + 1))))
src_id = a.get_meta('id')
except TierAddError:
return False
logging.debug('==> Source {:d}'.format(or_index))
echo_labels = list()
for s, e in repetition.get_echos():
rep_begin = spk2_tier[start_idx2 + s].get_lowest_localization()
rep_end = spk2_tier[start_idx2 + e].get_highest_localization()
anns = spk2_tier.find(rep_begin, rep_end)
for a in anns:
for lbl in a.get_labels():
echo_labels.append(lbl.copy())
logging.debug(' -> echo {} {}: {}'.format(s, e, lbl))
eetime = sppasInterval(rep_begin.copy(), rep_end.copy())
r = sppasLabel(sppasTag('R' + str(or_index + 1)))
try:
a = echo_tier.create_annotation(sppasLocation(eetime), r)
a.set_meta('source_id', src_id)
except TierAddError:
a = echo_tier.find(rep_begin, rep_end)
if len(a) > 0:
a[0].append_label(r)
anns = spk1_tier.find(src_begin, src_end)
src_labels = list()
for a in anns:
for lbl in a.get_labels():
src_labels.append(lbl.copy())
logging.debug(' => src: {}'.format(lbl))
a = trs_out.find('OR-SrcStrain').create_annotation(sppasLocation(iitime), src_labels)
a.set_meta('source_id', src_id)
a = trs_out.find('OR-SrcLen').create_annotation(sppasLocation(iitime), sppasLabel(sppasTag(len(src_labels), 'int')))
a.set_meta('source_id', src_id)
or_type = 'variation'
if len(repetition.get_echos()) > 1:
or_type = 'split:{:d}'.format(len(repetition.get_echos()))
elif len(src_labels) > len(echo_labels):
or_type = 'reduction'
elif len(src_labels) == len(echo_labels):
equals = True
for ls, le in zip(src_labels, echo_labels):
if ls.get_best() != le.get_best():
equals = False
break
if equals is True:
or_type = 'strict'
a = trs_out.find('OR-SrcType').create_annotation(sppasLocation(iitime), sppasLabel(sppasTag(or_type)))
a.set_meta('source_id', src_id)
all_echos_tier = trs_out.find('OR-AllEchos')
if all_echos_tier is not None:
for tok_idx in repetition.get_all_echos():
ann = spk2_tier[start_idx2 + tok_idx]
r = sppasLabel(sppasTag('R' + str(or_index + 1)))
find_a = all_echos_tier.find(ann.get_lowest_localization(), ann.get_highest_localization(), overlaps=False)
if len(find_a) == 1:
find_a[0].append_label(r)
else:
a = ann.copy()
a.set_labels([r])
all_echos_tier.add(a)
a.set_meta('source_id_of_R{:d}'.format(or_index + 1), src_id)
return True