Voiceprint
A mathematical representation of a voice used to recognize a speaker.
A voiceprint is a compact mathematical representation, often called an embedding, derived from a sample of someone speaking. It captures the characteristics that make a voice distinctive while discarding the literal words, so two recordings of the same person produce similar voiceprints even if they say different things.
Voiceprints are compared by measuring how close two vectors are. They are model-specific, which means a voiceprint created by one system cannot be read by another; migrating between vendors requires re-enrolling speakers from audio.
