The Algorithm of Voiceprint Recognition Model based DNN-RELIANCE
Jing Zhang · 2020 International Conference on Computer Engineering and Application (ICCEA) · 2020
The traditional voiceprint recognition model (Gaussian hybrid model, hidden Markov model) is very dependent on the scale of speaker speech. Higher recognition performance often requires a large number of trained speaker speech to achieve, and has such shortcomings as environmental noise sensitivity and long training time, besides, it is difficult in convergence, the application of voiceprint recognition in practice is limited accordingly. The deep neural network has the ability of autonomous learning, and it can extract the deep speaker speech features that are not sensitive to environmental noise according to the target, and the extracted speaker speech features can be well classified and identified with the powerful pattern classification ability. So the deep neural network (DNN) is used as the voiceprint recognition model in this paper. However, the experiment shows that the DNN-based voiceprint recognition system still has a low accuracy rate of rejection of counterfeiters. For this reason, the trust degree and label distance are introduced, and the two-order judgment structure based on DNN-RLIANCE algorithm is proposed. Experiments show that the algorithm has short training and recognition time but high recognition accuracy, especially the system's correct rejection rate for impostors is improved.