the script 'forced_alignment_novo.py' which is to run novo_api on Python 3.6 environment is added.

novoapi_for_python3x/asr/segments/__init__.py

@@ -0,0 +1,4 @@
+#!/usr/bin/env python
+
+from .segments import Segmentation
+from .praat import seg2tg

novoapi_for_python3x/asr/segments/praat.py

@@ -0,0 +1,77 @@
+#!/usr/bin/env python
+# (c) 2015--2018 NovoLanguage, author: David A. van Leeuwen
+
+import codecs
+
+def print_header(output, begin, end, nr_tiers):
+    print >> output, 'File type = "ooTextFile"'
+    print >> output, 'Object class = "TextGrid"'
+    print >> output, ''
+    print >> output, 'xmin = %s' % begin
+    print >> output, 'xmax = %s' % end
+    print >> output, 'tiers? <exists>'
+    print >> output, 'size = %d' % nr_tiers
+    print >> output, 'item []:'
+
+
+def print_info_tier(output, title, begin, end, label):
+    print >> output, '\titem [%d]:' % 0
+    print >> output, '\t\tclass = "IntervalTier"'
+    print >> output, '\t\tname = "%s"' % title
+    print >> output, '\t\txmin = %s' % begin
+    print >> output, '\t\txmax = %s' % end
+    print >> output, '\t\tintervals: size = %d' % 1
+
+    print >> output, '\t\tintervals [1]:'
+    print >> output, '\t\t\txmin = %s' % begin
+    print >> output, '\t\t\txmax = %s' % end
+    print >> output, '\t\t\ttext = "%s"' % label
+
+
+#def print_tier(output, title, begin, end, segs, (format, formatter)):
+def print_tier(output, title, begin, end, segs, format, formatter):
+    print >> output, '\titem [%d]:' % 0
+    print >> output, '\t\tclass = "IntervalTier"'
+    print >> output, '\t\tname = "%s"' % title
+    print >> output, '\t\txmin = %s' % begin
+    print >> output, '\t\txmax = %s' % end
+    print >> output, '\t\tintervals: size = %d' % len(segs)
+
+    count = 1
+    for seg in segs:
+        #print seg
+        print >> output, '\t\tintervals [%d]:' % count
+        print >> output, '\t\t\txmin = %s' % repr(int(seg['begin']) / 100.0)
+        print >> output, '\t\t\txmax = %s' % repr(int(seg['end']) / 100.0)
+        string = '\t\t\ttext = "' + format + '"'
+        print >> output, string % formatter(seg['label'])
+        count += 1
+
+
+def seg2tg(fname, segments):
+    if not segments:
+        return
+    output = codecs.open(fname, "w", encoding="utf-8")
+
+    confidences = []
+    word_labels = []
+    phones = []
+
+    for s in segments:
+        conf = s.llh if hasattr(s, "llh") else s.score
+        confidences.append({'begin': s.begin, 'end': s.end, 'label': conf})
+        word_labels.append({'begin': s.begin, 'end': s.end, 'label': s.label})
+        for p in s.phones:
+            phones.append({'begin': p.begin, 'end': p.end, 'label': p.label})
+
+
+    begin = repr(int(segments[0].begin) / 100.0)
+    end = repr(int(segments[-1].end) / 100.0)
+
+    nr_tiers = 3
+    print_header(output, begin, end, nr_tiers)
+    print_tier(output, "confidence", begin, end, confidences, ('%.3f', lambda x: x))
+    print_tier(output, "words", begin, end, word_labels, ('%s', lambda x: x))
+    print_tier(output, "phones", begin, end, phones, ('%s', lambda x: x))
+
+    output.close()

novoapi_for_python3x/asr/segments/segments.py

@@ -0,0 +1,99 @@
+#!/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]