The (speech-to-text technology) process of splitting mono audio into two channels, according to speaker. Higher diarization scores indicate that the speech recognizer separated speaker phrases more accurately.