Table 5.

PD vs. HC classification EER (in %) obtained with different segment lengths for the x-vectors extraction.

Classifier	Mismatched	Matched
x-vec + cos	41	39
x-vec + LDA + cos	36	32
x-vec + PLDA	36	33

The classification was performed using the sentence repetition tasks of the male telephone recordings. [1–5 s] segments were used for training and either [25 ms–100 s] (mismatched) or [1–5 s] (matched) segments for test. X-vectors were classified in various ways: with cosine similarity (alone and with LDA) and with PLDA. Bold numbers indicate the best EER for each classifier.