Abstract | ||
---|---|---|
Equal error rate (ERR) is a widely used evaluation metric of speaker verification, which reflects the performance of a verification system at a given decision threshold. However, the threshold may not serve the best for all applications, which triggers the need for optimizing the performance at a range of decision thresholds. To fulfill this objective, this paper proposes a back-end metric learning algorithm to directly maximize the partial area under the ROC curve (pAUC) given an interested range of false positive rate. The reason that we aim to maximize the pAUC where the false positive rate keeps low is because the number of imposter trials is much larger than that of true trials in practice. Moreover, wrong predictions of imposter trials will cause a great risk in many applications. Experimental results on NIST SRE data sets illustrate the effectiveness of our algorithm with either the i-vector or the x-vector extractors being used as its front-end. |
Year | DOI | Venue |
---|---|---|
2020 | 10.21437/Odyssey.2020-53 | Odyssey |
DocType | Citations | PageRank |
Conference | 0 | 0.34 |
References | Authors | |
7 | 3 |
Name | Order | Citations | PageRank |
---|---|---|---|
Zhongxin Bai | 1 | 2 | 1.87 |
Xiao-Lei Zhang | 2 | 147 | 14.47 |
Jingdong Chen | 3 | 1460 | 128.79 |