def get_DER(all_reference, all_hypothesis):
"""
calculates DER, CER, FA and MISS
Args:
all_reference (list[Annotation]): reference annotations for score calculation
all_hypothesis (list[Annotation]): hypothesis annotations for score calculation
Returns:
DER (float): Diarization Error Rate
CER (float): Confusion Error Rate
FA (float): False Alarm
Miss (float): Miss Detection
< Caveat >
Unlike md-eval.pl, "no score" collar in pyannote.metrics is the maximum length of
"no score" collar from left to right. Therefore, if 0.25s is applied for "no score"
collar in md-eval.pl, 0.5s should be applied for pyannote.metrics.
"""
metric = DiarizationErrorRate(collar=0.5, skip_overlap=True)
for reference, hypothesis in zip(all_reference, all_hypothesis):
metric(reference, hypothesis, detailed=True)
DER = abs(metric)
CER = metric['confusion'] / metric['total']
FA = metric['false alarm'] / metric['total']
MISS = metric['missed detection'] / metric['total']
metric.reset()
return DER, CER, FA, MISS
def get_DER(all_reference, all_hypothesis):
"""
calculates DER, CER, FA and MISS
Args:
all_reference (list[Annotation]): reference annotations for score calculation
all_hypothesis (list[Annotation]): hypothesis annotations for score calculation
Returns:
DER (float): Diarization Error Rate
CER (float): Confusion Error Rate
FA (float): False Alarm
Miss (float): Miss Detection
"""
metric = DiarizationErrorRate(collar=0.25, skip_overlap=True)
DER = 0
for reference, hypothesis in zip(all_reference, all_hypothesis):
metric(reference, hypothesis, detailed=True)
DER = abs(metric)
CER = metric['confusion'] / metric['total']
FA = metric['false alarm'] / metric['total']
MISS = metric['missed detection'] / metric['total']
metric.reset()
return DER, CER, FA, MISS