Speaker-Count Error

Speaker-count error measures whether a system found the right number of people in a recording. It is usually computed as the absolute difference between the number of speakers detected and the number actually present, divided by the number present, then averaged across recordings. A four-speaker file where a system finds three scores 25%.

It is reported alongside diarization error rate because the two can disagree: a system can place turn boundaries well and still be wrong about how many people are in the room. Counting accuracy typically degrades as the number of speakers grows, so results are more informative broken out by speaker count than given as a single average.