基于深度神经网络和多元损失的说话人识别
DOI:
作者:
作者单位:

作者简介:

通讯作者:

中图分类号:

TP391;TN919.81

基金项目:


Speaker verification based on deep learning and beyond triplet loss
Author:
Affiliation:

Fund Project:

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    生物特征识别技术相对于传统密码等方式具有更高的可靠性,而作为生物特征识别技术的重要研究方向之一的声纹识别方法,研究更精确的声纹识别方法具有更高的研究意义。随着深度学习的发展,深度学习应用在声纹识别技术上成为在声纹识别领域研究的重点。提出一种基于深度神经网络和beyond triplet loss相结合的说话人识别方法,模型通过梅尔频率倒谱系数(MFCC)提取MFCC声学特征,对MFCC声学特征提取说话人声纹特征,然后进行多元损失的模型训练。实验结果表明,DNN-BTL算法在说话人识别领域比高斯混合-隐马尔可夫模型(GMM-HMM)具有更好的识别效果。

    Abstract:

    Biometric recognition technology has higher reliability than traditional cryptography. As one of the important research directions of biometrics, voiceprint recognition method has more research significance to study more accurate voiceprint recognition methods. With the development of deep learning, the application of deep learning in voiceprint recognition technology has become the focus of research in voiceprint recognition field. In this paper, a speaker recognition method based on deep neural network and beyond triplet loss is proposed. The model extracts the acoustic characteristics of MFCC through Mel-frequency cepstral coefficients, and extracts the voiceprint characteristics of the speaker from the MFCC acoustic characteristics, and then carries out the beyond triplet loss model training. Experimental results show that DNN-BTL algorithm has better recognition effect in speaker recognition field than Gussian mixture model-hidden Markov model.

    参考文献
    相似文献
    引证文献
引用本文

关健,王敏.基于深度神经网络和多元损失的说话人识别[J].电子测量技术,2019,42(5):39-43

复制
分享
文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:
  • 最后修改日期:
  • 录用日期:
  • 在线发布日期: 2021-07-29
  • 出版日期: