def transcription():
transcription = Transcription()
transcription.add_edge(3, 5, speech="hi there, I'm Penny",
speaker='Penny')
transcription.add_edge(5, 5.5)
transcription.add_edge(5.5, 7, speech="hi. I'm Leonard", speaker='Leonard')
transcription.add_edge(T('A'), T('B'), summary='Leonard meets Penny')
transcription.add_edge(7, T('C'))
return transcription
def _duration(self, lines, start, end):
"""Iterate dialogue lines + (estimated) timeranges"""
if self.duration:
length = np.array([len(line)+1 for line in lines])
ratio = 1. * np.cumsum(length) / np.sum(length)
end_times = start + (end - start) * ratio
else:
end_times = [T() for line in lines[:-1]] + [end]
start_time = start
for line, end_time in six.moves.zip(lines, end_times):
yield line, start_time, end_time
start_time = end_time
def read(self, path, **kwargs):
T.reset()
transcription = Transcription()
previousNode = TStart
arc = []
for line in self.iterlines(path, encoding='utf-8'):
# skip comments
if re.search(r'^;;', line):
continue
fields = line.strip().split()
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
word = fields[4]
# remove punctuations
if not self.punctuation:
word = re.sub(r'[\.!,;?":]+', ' ', word)
word = re.sub(r' +', ' ', word)
word = word.strip()
if not word:
continue
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
confidence = float(fields[5])
start_time = round(float(fields[2]), 3)
duration = round(float(fields[3]), 3)
end_time = float(start_time)+float(duration)
end_time = round(end_time, 3)
if duration == 0:
node_start = previousNode
node_end = T()
if len(arc) == 2:
transcription.remove_edge(arc[0], arc[1])
transcription.add_edge(arc[0], node_end, **arc_data)
node_inter = T()
transcription.add_edge(node_end, node_inter, speech=word, confidence=confidence)
transcription.add_edge(node_inter, arc[1])
arc.append(node_end)
arc.append(node_inter)
node_end=arc[1]
elif len(arc) > 2:
node_anc_start = arc[0]
node_anc_end = arc[1]
transcription.remove_edge(arc[len(arc)-1], node_anc_end)
transcription.add_edge(arc[len(arc)-1], node_end, speech=word, confidence=confidence)
transcription.add_edge(node_end, node_anc_end)
arc.append(node_end)
node_end=arc[1]
else:
addEdge = True
node_start = T(start_time)
node_end = T(end_time)
if previousNode.drifting:
if not transcription.has_edge(previousNode, node_start):
transcription.add_edge(previousNode, node_start)
else:
if node_start < previousNode:
node_start = previousNode
elif node_start > previousNode:
transcription.add_edge(previousNode, node_start)
if node_start.anchored and node_end.anchored:
if node_start == node_end:
addEdge = False
node_start = previousNode
node_end = T()
if len(arc) == 2:
transcription.remove_edge(arc[0], arc[1])
transcription.add_edge(arc[0], node_end, **arc_data)
node_inter = T()
transcription.add_edge(node_end, node_inter,
speech=word,
confidence=confidence)
transcription.add_edge(node_inter, arc[1])
arc.append(node_end)
arc.append(node_inter)
node_end = arc[1]
elif len(arc) > 2:
node_anc_start = arc[0]
node_anc_end = arc[1]
transcription.remove_edge(arc[len(arc) - 1],
node_anc_end)
transcription.add_edge(arc[len(arc) - 1], node_end, speech=word, confidence=confidence)
transcription.add_edge(node_end, node_anc_end)
arc.append(node_end)
node_end = arc[1]
else:
arc = [node_start, node_end]
arc_data = {'speech': word, 'confidence': confidence}
if addEdge:
transcription.add_edge(node_start, node_end,
speech=word, confidence=confidence)
previousNode = node_end
transcription.add_edge(previousNode, TEnd)
self._loaded = transcription
return self
