An improved normalized gain-based score normalization technique for spoof detection algorithm

A spoof detection algorithm supports the speaker verification system to examine false claims by an imposter through careful analysis of input test speech. The scores are employed categorize genuine and spoofed samples effectively. Under mismatch conditions, acceptance ratio increases can be reduced appropriate score normalization techniques. In this article, we using normalized Discounted Cumulative Gain (nDCG) norm derived from ranking speaker’s log-likelihood scores. proposed scoring technique smoothens decaying process due logarithm with added advantage ranking. baseline employs Constant Q-Cepstral Co-efficient (CQCC) as base features a Gaussian Mixture Model (GMM) based classifier. computed ASVspoof 2019 dataset for without conditions. techniques including Zero (Z-norm) Test (T-norm) also considered. is found perform better in terms improved Equal Error Rate (EER) 0.35 against 0.43 (no normalization) wrt synthetic attacks development data. Similarly, improvements seen case replay attack EER 7.83 nDCG-norm 9.87 no (no-norm). Furthermore, tandem-Detection Cost Function (t-DCF) 0.015 no-norm 0.010 normalization. Additionally, t-DCF 0.195 0.17 performance satisfactory when evaluated evaluation data 8.96 9.57 while 9.79 11.04 attacks. Supporting EER, 0.1989 0.2636 attacks; attacks, 0.2284 0.2454 no-norm. increase accuracy overall system.