def process_utterance(self, utt, make_label=True):
utt_data = []
utt_questions = defaultdict(int)
nodelist = utt.xpath(self.config["target_nodes"])
if nodelist == []:
print(
'WARNING: FeatureDumper\'s target_nodes matches no nodes: %s' %
(self.config["target_nodes"]))
for node in nodelist:
self.htk_state_xpath = None ## make sure this is none.
self.start_time_xpath = None
self.end_time_xpath = None
## for phone!:--
node_data, node_questions = self.get_node_context_label(node)
statelist = node.xpath('.//' + self.state_tag)
assert statelist != []
for (i, state) in enumerate(statelist):
state_ix = i + 2
state_node_data = "%s[%s]" % (node_data, state_ix)
start_time = state.attrib.get(self.start_attribute,
'_NA_') ## no time at runtime!
end_time = state.attrib.get(self.end_attribute, '_NA_')
if not (start_time == "_NA_" or end_time == "_NA_"):
start_time = string.ljust(str(ms_to_htk(start_time)), 10)
end_time = string.ljust(str(ms_to_htk(end_time)), 10)
state_node_data = "%s %s %s" % (start_time, end_time,
state_node_data)
utt_data.append(state_node_data)
##utt_questions.update(node_questions)
## Sum the dictionaries' values:
for question in node_questions:
utt_questions[question] += node_questions[question]
if make_label:
label_file = utt.get_filename(self.config["output_filetype"])
writelist(utt_data, label_file, uni=True)
return (utt_data, utt_questions
) ## for writing utterance-level labels,
def get_node_context_label(self, node):
# print('in MFD get_node_context_label')
context_vector = node.get_context_vector(self.contexts)
mapped_context_vector = []
for (i, (name, value)) in enumerate(context_vector):
if i in self.mappers:
# mapped_names = [name + '=' + field_name for field_name in self.mappers[i].feature_names]
mapped_values = self.mappers[i].lookup(value)
# mapped_context_vector.extend(zip(mapped_names, mapped_values))
mapped_context_vector.extend(mapped_values)
else:
mapped_context_vector.append(value)
## add numbers:
assert len(mapped_context_vector) == self.number_of_features
context_vector = zip(range(self.number_of_features), self.mapped_feature_names, mapped_context_vector)
# At this point, context_vector looks like this:
#
# [(0, u'll_segment:cmanner=affric', 0.0), (1, u'll_segment:cmanner=approx', 0.0), (2, u'll_segment:cmanner=fric', 0.0), (3, u'll_segment:cmanner=lateral', 0.0), (4, u'll_segment:cmanner=nasal', 1.0), (5, u'll_segment:cmanner=stop', 0.0), (6, u'll_segment:cplace=alveolar', 0.0),
if self.context_separators == "numbers":
formatted_context_vector = ["%s:%s" % (number, value)
for (number, name, value) in context_vector]
formatted_context_vector = "/".join(formatted_context_vector)
formatted_context_vector = "/" + formatted_context_vector + "/"
else:
if self.context_separators == "spaces":
separator = " "
elif self.context_separators == "commas":
separator = ","
else:
sys.exit("'%s' not a recognised separator" % (self.context_separators))
formatted_context_vector = [str(value) for (number, name, value) in context_vector]
formatted_context_vector = separator.join(formatted_context_vector)
if self.htk_monophone_xpath:
## PREpend an extra monophone feature -- appending will screw up
## extraction of sentence level contexts, which (currently) are
## assumed to be at the end of the model name:
htk_monophone = node.safe_xpath(self.htk_monophone_xpath)
formatted_context_vector = "-%s+%s" % (htk_monophone, formatted_context_vector)
## Don't need to add this to context questions -- just used for
## extracting monophones, not context clustering.
## TODO: find a neater way to handle this? Don't rely on HTK's
## inbuilt monophone extractor in the HTS-Training script?
if self.htk_state_xpath:
## Increment to start state count at 2 as in HTK
htk_state = node.safe_xpath(self.htk_state_xpath)
formatted_context_vector = "%s[%s]" % (formatted_context_vector, htk_state + 1)
if self.start_time_xpath and self.end_time_xpath:
start_time = node.safe_xpath(self.start_time_xpath)
end_time = node.safe_xpath(self.end_time_xpath)
## safe_xpath will give _NA_ when times are absent (i.e at runtime) --
## in this case, omit times:
if not (start_time == "_NA_" or end_time == "_NA_"):
start_time = string.ljust(str(ms_to_htk(start_time)), 10)
end_time = string.ljust(str(ms_to_htk(end_time)), 10)
formatted_context_vector = "%s %s %s" % (start_time, end_time, formatted_context_vector)
return formatted_context_vector
def get_node_context_label(self, node):
"""
依据Utterance对象中的一个节点(如，segment节点)，产生上下文标签。注：节点类型仍为Utterance对象
:param node: segment节点
:return: 上下文标签
"""
context_vector = node.get_context_vector(self.contexts)
## add numbers:
context_vector = [(number, name, value) for \
(number, (name, value)) in enumerate(context_vector)]
node_questions = defaultdict(int) ## {}
for triplet in context_vector:
node_questions[triplet] += 1 ## store question's triplet as a key -- sort later
## Count can be used for filtering infreq. questions
if self.context_separators == "numbers":
formatted_context_vector = ["%s:%s" % (number, value)
for (number, name, value) in context_vector]
formatted_context_vector = "/".join(formatted_context_vector)
formatted_context_vector = "/" + formatted_context_vector + "/"
else:
if self.context_separators == "spaces":
separator = " "
elif self.context_separators == "commas":
separator = ","
else:
sys.exit("'%s' not a recognised separator" % (self.context_separators))
formatted_context_vector = [str(value) for (number, name, value) in context_vector]
formatted_context_vector = separator.join(formatted_context_vector)
if self.htk_monophone_xpath:
## PREpend an extra monophone feature -- appending will screw up
## extraction of sentence level contexts, which (currently) are
## assumed to be at the end of the model name:
htk_monophone = node.safe_xpath(self.htk_monophone_xpath)
formatted_context_vector = "-%s+%s" % (htk_monophone, formatted_context_vector)
## Don't need to add this to context questions -- just used for
## extracting monophones, not context clustering.
## TODO: find a neater way to handle this? Don't rely on HTK's
## inbuilt monophone extractor in the HTS-Training script?
if self.htk_state_xpath:
## Increment to start state count at 2 as in HTK
htk_state = node.safe_xpath(self.htk_state_xpath)
formatted_context_vector = "%s[%s]" % (formatted_context_vector, htk_state + 1)
if self.start_time_xpath and self.end_time_xpath:
start_time = node.safe_xpath(self.start_time_xpath)
end_time = node.safe_xpath(self.end_time_xpath)
## safe_xpath will give _NA_ when times are absent (i.e at runtime) --
## in this case, omit times:
if not (start_time == "_NA_" or end_time == "_NA_"):
start_time = string.ljust(str(ms_to_htk(start_time)), 10)
end_time = string.ljust(str(ms_to_htk(end_time)), 10)
formatted_context_vector = "%s %s %s" % (start_time, end_time, formatted_context_vector)
return (formatted_context_vector, node_questions)
