【摘要】 声纹识别属于一种新型生物识别技术，其综合了生命科学综合研究，计算机技术等多种技术。随着深度学习技术不断的发展，声纹识别技术在案件侦破、智能网联、支付系统上的应用也越来越多，论文针对现有声纹识别系统识别率低，识别效率慢等问题，提出了基于注意力机制改进的DenseNet网络模型作为声学模型，进一步提高声纹识别系统的性能。首先将语音经过预处理和特征提取，进入改进后的DenseNet网络中，最终进入SoftMax函数输出结果，最终经过多组实验验证并进行比对，实验结果表明，使用注意力机制改进的DenseNet网络作为声纹识别系统中的声学模型相较于其他传统声学模型在准确率、AUC上分别提升了4.25%、4.18%，在等错误率上降低了6.09%，证明了该模型对于声纹识别任务上的合理性。更多还原

【Abstract】 Voiceprint recognition belongs to a new type of biological recognition technology,the combination of comprehensive research on life science,computer technology and other technology. With the development of deep learning technology constantly,voiceprint recognition technology in case detection,the application of intelligent snatched,payment system is also more and more,this article in view of the existing recognition rate is low,voiceprint recognition system identification efficiency is slow. An improved DenseNet network model based on attentional mechanism is proposed as an acoustic model to further improve the performance of the voice print recognition system. Firstly,the speech is preprocessed and extracted into the improved DenseNet network,and finally into the SoftMax function to output the results. Finally,the results are verified and compared by multiple groups of experiments. The experimental results show that,compared with other traditional acoustic models,the accuracy and AUC of DenseNet network improved by attention mechanism are improved by 4.25% and 4.18% respectively,and the equal error rate is reduced by 6.09%,which proves the rationality of the model for the voice print recognition task.更多还原