

In [18]:

    
%pylab inline









    



Populating the interactive namespace from numpy and matplotlib

Diarization evaluation metrics



In [19]:

    
from pyannote.core import Annotation, Segment



In [20]:

    
reference = Annotation()
reference[Segment(0, 10)] = 'A'
reference[Segment(12, 20)] = 'B'
reference[Segment(24, 27)] = 'A'
reference[Segment(30, 40)] = 'C'
reference









    Out[20]:



In [21]:

    
hypothesis = Annotation()
hypothesis[Segment(2, 13)] = 'a'
hypothesis[Segment(13, 14)] = 'd'
hypothesis[Segment(14, 20)] = 'b'
hypothesis[Segment(22, 38)] = 'c'
hypothesis[Segment(38, 40)] = 'd'
hypothesis









    Out[21]:

Diarization error rate



In [22]:

    
from pyannote.metrics.diarization import DiarizationErrorRate
diarizationErrorRate = DiarizationErrorRate()
print("DER = {0:.3f}".format(diarizationErrorRate(reference, hypothesis, uem=Segment(0, 40))))









    



DER = 0.516

Optimal mapping



In [23]:

    
reference









    Out[23]:



In [24]:

    
hypothesis









    Out[24]:



In [25]:

    
diarizationErrorRate.optimal_mapping(reference, hypothesis)









    Out[25]:





{'a': 'A', 'b': 'B', 'c': 'C'}

Details



In [26]:

    
diarizationErrorRate(reference, hypothesis, detailed=True, uem=Segment(0, 40))









    Out[26]:





{'confusion': 7.0,
 'correct': 22.0,
 'diarization error rate': 0.5161290322580645,
 'false alarm': 7.0,
 'missed detection': 2.0,
 'total': 31.0}

Clusters purity and coverage



In [27]:

    
from pyannote.metrics.diarization import DiarizationPurity
purity = DiarizationPurity()
print("Purity = {0:.3f}".format(purity(reference, hypothesis, uem=Segment(0, 40))))









    



Purity = 0.828



In [28]:

    
from pyannote.metrics.diarization import DiarizationCoverage
coverage = DiarizationCoverage()
print("Coverage = {0:.3f}".format(coverage(reference, hypothesis, uem=Segment(0, 40))))









    



Coverage = 0.759