Speaker embedding
A numeric vector that represents a speaker's voice characteristics, used to tell speakers apart and to recognise the same speaker across recordings.
A speaker embedding (sometimes called an x-vector) is a vector that captures the distinctive qualities of a voice. Diarization uses these embeddings to group audio segments by speaker — segments with similar embeddings are treated as the same person.
Because embeddings are comparable across files, they also enable speaker identification: matching “Speaker 1” in one recording to the same person in another. Extracting them adds a little processing time, so some tools make it optional.