99 lines
3.4 KiB
Python
99 lines
3.4 KiB
Python
#!/usr/bin/env python
|
|
# (c) 2015--2018 NovoLanguage, author: David A. van Leeuwen
|
|
|
|
## These classes can be initialized with dictionaries, as they are returned by the python spraaklab recognition system.
|
|
|
|
class Segment(object):
|
|
def __init__(self, segment):
|
|
self.begin = segment["begin"]
|
|
self.end = segment["end"]
|
|
self.begintime = segment.get("beginTime", self.begin / 100.0)
|
|
self.endtime = segment.get("endTime", self.end / 100.0)
|
|
self.label = segment["label"]
|
|
self.score = segment["score"]
|
|
if "llh" in segment:
|
|
self.llh = segment["llh"]
|
|
if "phones" in segment:
|
|
self.type = "word"
|
|
self.phones = Segmentation(segment["phones"], ["sil"])
|
|
if hasattr(self.phones[0], "llh"):
|
|
self.minllh = min([s.llh for s in self.phones]) ## the current word llh for error detection
|
|
else:
|
|
self.type = "phone"
|
|
|
|
def __repr__(self):
|
|
res = "%8.3f -- %8.3f score %8.3f " % (self.begintime, self.endtime, self.score)
|
|
if hasattr(self, "llh"):
|
|
res += "llh %8.3f " % self.llh
|
|
res += self.label.encode("utf8")
|
|
return res
|
|
|
|
def export(self):
|
|
r = {"begin": self.begin, "end": self.end, "label": self.label, "score": self.score, "type": self.type}
|
|
if hasattr(self, "llh"):
|
|
r["llh"] = self.llh
|
|
if hasattr(self, "phones"):
|
|
r["phones"] = self.phones.export()
|
|
return r
|
|
|
|
class Segmentation(object):
|
|
def __init__(self, segments, sils=["<s>", "</s>", "!sil"]):
|
|
"""Create a segmentation from a spraaklab recognition structure.
|
|
segments: an array of words (or phones), represented by a dict with
|
|
"begin", "end", "label", "score", and "llh" keys. Words can also have
|
|
"phones" which is another array of segments."""
|
|
self.segments = [Segment(s) for s in segments]
|
|
if self.segments:
|
|
self.type = self.segments[0].type
|
|
else:
|
|
self.type = None
|
|
self.sils = sils
|
|
self.orig = segments ## in case we want to have access to the original recognition structure
|
|
|
|
def __getitem__(self, item):
|
|
return self.segments[item]
|
|
|
|
def __repr__(self):
|
|
ns = len(self.segments)
|
|
res = "Segmentation with %d %s%s" % (ns, self.type, "" if ns==1 else "s")
|
|
for seg in self.segments:
|
|
res += "\n " + repr(seg)
|
|
return res
|
|
|
|
def __len__(self):
|
|
return len(self.segments)
|
|
|
|
def score(self, skip=None):
|
|
if not skip:
|
|
skip = self.sils
|
|
s = 0.0
|
|
for seg in self.segments:
|
|
if seg.label not in skip:
|
|
s += seg.score
|
|
return s
|
|
|
|
def llhs(self, skip=None):
|
|
if not skip:
|
|
skip = self.sils
|
|
return [seg.llh for seg in self.segments if hasattr(seg, "llh") and seg.label not in skip]
|
|
|
|
def llh(self, skip=None):
|
|
return sum(self.llhs(skip))
|
|
|
|
def minllh(self, skip=None):
|
|
llhs = self.llhs(skip)
|
|
if llhs:
|
|
return min(llhs)
|
|
else:
|
|
return None
|
|
|
|
def labels(self, skip=None):
|
|
if not skip:
|
|
skip = self.sils
|
|
return [seg.label for seg in self.segments if seg.label not in skip]
|
|
|
|
def sentence(self, skip=None):
|
|
return " ".join(self.labels(skip))
|
|
|
|
def export(self):
|
|
return [seg.export() for seg in self.segments] |