ORC-WER (Optimal Reference Combination Word Error Rate)

Optimal reference combination word error rate scores a multi-speaker transcript by finding the assignment of reference utterances to output streams that minimizes total word error, without holding the system to a consistent speaker identity across the whole recording.

It isolates whether the words survived, where cpWER also penalizes attributing them to the wrong person. The two therefore measure different things and their figures are not directly comparable, so an evaluation should always state which one it is reporting.