| 摘要: |
| 利用连续可变学习速率处理回声抵消双话情况时，随着近端语音能量的提高，学习速
率的估计偏差增大，导致残留回声增加。提出了一种利用短时能量比显示检测与连续学
习速率相结合的改进双话处理算法。该算法利用近端与远端语音的短时能量比，对学习速率
估计中的泄露因子参数进行自适应修正，从而调整连续学习速率。实验证明，该算法使得回
声抵消双话情况下，自适应滤波器发散程度下降，语音质量得到提升。在近远端能量比-6
～6 dB范围内，回声返回损失增加度（ERLE）提高0～11 dB,平均意见得分（MOS）
提高0.02～0.45分。 |
| 关键词: 回声抵消 双话检测 泄露因子 能量比检测 |
| DOI： |
|
| 基金项目: |
|
| An improved method to handle double talk using energy ratio detection combined with continuous learning rate |
| HUANG Ying,TANG Kun,CUI Hui-juan |
| (Tsinghua National Laboratory for Information Science and Technology，Department of Electronic Engineering，Tsinghua University，Beijing 100084，China) |
| Abstract: |
| When handling double talk in echo cancellation with continuous variable
learning rate,estimation of learning rate is biased as energy of near end sign
al improves,thus causing more residual echo. This paper proposes an improved me
thod to handle double talk using explicit detection of energy ratio combined wit
h variable continuous learning rate. Short time energy ratio of near end to far
end signal is used to adaptively modify leakage parameter in learning rate estim
ation, thus adjusting learning rate. Experimental results show that this algorit
hm can decrease divergence of adaptive filter and improve sound quality in doubl
e-talk situation of echo cancelation. With energy ratio of near to far end signa
l between -6 dB and 6dB, ERLE(Echo Return Loss Enhancement) is improved by 0
～11 dB and MOS(Mean Option Score) is improve
d by 002～045 point. |
| Key words: echo cancellation double talk detection leakage parameter energy ratio detection |
《电讯技术》编辑部 