欢迎您！东篱公司
退出
-
申报数据库
-
立项数据库
-
成果数据库
-
项目获奖数据库
中文摘要：
在熵的基础上,引人数学上的负熵,并提出一种基于负熵特征的语音端点检测算法。算法利用平稳噪声的长时平稳特性,并通过合理假设,从噪声幅度谱中提取隐藏的高斯随机信息,在此信息基础上应用近似负熵算法构造负熵特征。与熵特征不同处在于,对平稳噪声负熵特征值趋近于零,并且与噪声信号幅度无关,基于这两种特性可以利用噪声的先验统计信息预先设定阈值,构造鲁棒性能较高的语音端点检测算法。实验表明,即使在噪声信号类型、幅度、信噪比改变或者无法正确的获取噪声后验信息的情况下,新算法依然能够保持较高的噪声检测正确率。
英文摘要：
In this paper,we introduced mathematical negentropy on the basis of entropy,and developed a Voice Activity Detection algorithm employing negentropy feature. In the proposed algorithm, the procedure of feature construction is as follows：first, the spectral statistics of current frame is derived from frames nearby according to the noise nature of long-term stationarity; then, the Gaussian se- quence with zero mean and unit variance,which hidden in spectrum,is abstracted from current frame based on a reasonable hypothesis; and last, the feature is constructed through applying approximate negentropy method to the sequence. Different from the entropy, the value of negentropy feature is approach to zero and unrelated to the amplitude of noise signal,since the threshold can be decide by the priori information. As a result, the proposed algorithm can work well in complex environments even when the type, amplitude, and SNR of noise signal are all varied or the posterior information can＇ t be obtained correctly. And the experimental results demonstrate the robustness of the proposed algorithm.
同期刊论文项目
同项目期刊论文
Decoupled 2D direction of arrival estimation using compact uniform circular arrays in the presence o
期刊信息
位置：