The enrollment phase for speaker verification

[JerryPeng21cuhk]

This experiment seems to provide an "utterance v.s. utterance" trial. That is to determine whether two utterances are uttered by the same speaker. However, for standard SV, it requires an enrollment phase. For ivector-based approaches, that is to score the similarity between the vector provided by all utterances uttered by a speaker in enrollment set and the vector from a test utterance.