• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文会议>Pacific-Rim conference on multimedia >Speech Intelligibility Enhancement in Strong Mechanical Noise Based on Neural Networks
【24h】

Speech Intelligibility Enhancement in Strong Mechanical Noise Based on Neural Networks

机译：基于神经网络的强机械噪声中语音清晰度的提高

获取原文
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

Speech intelligibility is a significant factor for successful speech communication. To enhance the intelligibility, many methods have been proposed, mainly by operating the speech signal such as increasing the amplitude or modifying the speech spectrum. However, their effects are limited when the background noise is extremely strong. In this paper, we purpose a preprocessed noise cancellation model to enhance the speech intelligibility by predicting the cancelling signal and superimposing it into the speech signal. We build a deep neural network (DNN) model to make the prediction algorithm have better accuracy. Finally, the effectiveness of the algorithm was verified by objective and subjective tests, the average of signal-to-noise ratio (SNR) improved 4.5 dB, the average of speech intelligibility index (SO) increased 5.4% and the average of comparison mean opinion score (CMOS) rose 1.16 on a variety of test cases.
机译：语音清晰度是成功进行语音交流的重要因素。为了提高清晰度，已经提出了许多方法，主要是通过操作语音信号来进行的，例如增加幅度或修改语音频谱。但是，当背景噪声非常强时，它们的作用会受到限制。在本文中，我们采用了一种预处理的噪声消除模型，通过预测消除信号并将其叠加到语音信号中来增强语音清晰度。我们建立了一个深度神经网络（DNN）模型，以使预测算法具有更好的准确性。最后，通过主观和客观测试验证了算法的有效性，信噪比（SNR）平均值提高了4.5 dB，语音清晰度指数（SO）平均值提高了5.4％，比较平均意见的平均值在各种测试案例中，分数（CMOS）上升了1.16。

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