Anomaly detection, as a part of network security, is an important question, which has attracted much attention. The characteristics of data mining make it suitable for anomaly detection. Cluster analysis is a kind of ...
详细信息
ISBN:
(纸本)9781467386005
Anomaly detection, as a part of network security, is an important question, which has attracted much attention. The characteristics of data mining make it suitable for anomaly detection. Cluster analysis is a kind of data mining technology and it can divide records into different clusters, which is convenient for anomaly detection. Traditional K-manes is affected by the selection of initial centers, the number of clusters and isolated points. We combine information entropy and dd algorithm to improve K-means and use Kdd CUP99 data set to analysis the performance. From twice experiences, we find that improved K-means has higher detection rate and lower false positive rate than traditional K-means.
The estimation of the a priori signal-to-noise ratio (SNR) is a very significant issue for many speech enhancement algorithms. The widely-used decision-directed (dd) algorithm largely depresses the musical noise, but ...
详细信息
ISBN:
(纸本)9781479958368
The estimation of the a priori signal-to-noise ratio (SNR) is a very significant issue for many speech enhancement algorithms. The widely-used decision-directed (dd) algorithm largely depresses the musical noise, but the estimated a priori SNR suffer from one frame delay which results in the degradation of speech quality. In this paper, we propose a novel algorithm to a priori SNR estimation which solves the above problem while keeping the advantage of the dd approach. First, a momentum term is added and incorporated into the traditional dd approach to accelerate the tracking speed for the a posteriori SNR. Then a self-adaptive momentum factor is achieved in the minimum-mean-squared-error (MMSE) sense to improve the allover performance of the proposed algorithm. Simulation experiment results show that our proposed algorithm brings significant improvement compared to the dd and fixed momentum factor algorithms under various noisy types and levels.
暂无评论