摘要
最常用的先验信噪比估计算法—DD(Decision directed)算法的估计性能依赖于平滑因子对音乐噪声和语音失真的折衷平衡。当平滑因子取值较大的时候,语音失真严重,但是对音乐噪声抑制彻底;当平滑因子取值较小的时候,语音失真较小,但是音乐噪声严重。为了解决这个问题,本文根据语音存在概率融合两个取平滑因子不同值的DD算法,以把这两种算法的优点结合在一起,组成了一个新的算法。仿真结果显示,本文提出的新算法估计的增强语音的音乐噪声抑制彻底且语音失真较小。
The DD （ decision directed ） method widely used for estimating the a prior signal-noise-ratio （ SNR ） suffers from the compromise balance between musical noise effect and speech distortion reduction. When the value of the smoothing factor is set close to one, the enhanced speech will have a serious speech distortion and less musical noise; while the value of the smoothing factor is set close to zero, the enhanced speech has less speech distortion, and a serious musical noise. In order to solve this problem, this paper proposes a new approach in order to retain the advantages of the less musical noise and small speech distortion, which is based on speech presence probability and combine two DD methods with different smoothing factors. The simulation results show that the enhanced speech based on the modified method proposed in this paper are both thoroughly suppress musical noise and smaller speech distortion.
作者
沈锁金
刘伟
高颖
SHEN Suo-jin LIU Wei GAO Ying(School of Science and Technology for Opto-Electronics Information of Yantai University, Yantai 264005, Chin)
出处
《中国集成电路》
2016年第12期26-30,共5页
China lntegrated Circuit
基金
烟台大学研究生科技创新基金资助重点项目(编号:YDZD1605)
关键词
先验信噪比
平滑因子
语音存在概率
语音失真
音乐噪声
a priori signal-to-noise ratio
smoothing factor
, speech presence probability
speech distortion
musical noise
