Evaluation of the deep nonlinear metric learning based speaker identification on the large scale of voiceprint corpus
Yong Feng, Cai Xinyuan, Ji Ruifang · 2016
Speaker recognition is a valuable biometric recognition technology. Recently, with the development of deep learning, many deep network based speaker recognition algorithms are proposed. However, the evaluations on the speaker recognition algorithms are always performed on a small or middle scale of voiceprint corpus. There are few evaluation reports of speaker recognition on a large scale of voiceprint corpus. In this paper, we try to introduce a deep nonlinear metric learning based speaker identification algorithm and construct a larger scale of voiceprint corpus consisting of about 440 thousand people. We perform the evaluation of the deep nonlinear metric learning based method on the large scale corpus and give the recognition rate results in different scales. The evaluation results prove that, the performance is surely decreased when the scale of the corpus is increased. But for the scale of 400 thousand, the recognition rate of Top 50 is stable at about 95%, which means that it could be used in some real applications.