【24h】

Tibetan acoustic model research based on TDNN

机译:基于TDNN的藏语声学模型研究

获取原文

摘要

Deep neural network (DNN) has been significantly improved in Tibetan speech recognition tasks, however, it still requires improvement when compared with that in Mandarin, English, or other languages. This paper examines a Tibetan acoustic model based on deep neural network and extracts the i-Vector features by modeling the speaker in the feature space. After combining the MFCCs and i-Vector features, we train a time-delayed neural network (TDNN) based Tibetan acoustic model, compared to deep neural network, it can get better performance. At the same time, we study the transfer learning from Mandarin to Tibetan and prove its effectiveness.
机译:深度神经网络(DNN)在藏文语音识别任务中已得到显着改进,但是与普通话,英语或其他语言相比,它仍需要改进。本文研究了基于深度神经网络的藏族声学模型,并通过在特征空间中对说话人建模来提取i-Vector特征。结合MFCC和i-Vector功能后,我们训练了基于时延神经网络(TDNN)的藏语声学模型,与深层神经网络相比,它可以获得更好的性能。同时,我们研究了从普通话到藏语的迁移学习,并证明了其有效性。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱:kefu@zhangqiaokeyan.com

京公网安备:11010802029741号 ICP备案号:京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有
  • 客服微信

  • 服务号