PLDA (Probabilistic Linear Discriminant Analysis)
A calibrated scoring method that decides whether two voiceprints are the same speaker.
PLDA, or Probabilistic Linear Discriminant Analysis, is a statistical method used to decide whether two voiceprints come from the same speaker. Instead of a raw similarity number, it produces a calibrated score that accounts for how much voices naturally vary, which makes match decisions more reliable across different speakers and conditions.
In voice biometrics, PLDA scoring is often used alongside, or instead of, plain cosine similarity because its calibration translates into more trustworthy accept and reject thresholds.
