The purpose of the speaker classification system is to segment and classify speech data according to different speaker. It improves performance of the speaker classification system by extracting multi-delay feature based on multiple distance microphones. With the number of microphones increases, the multi-delay feature vector dimension grows rapidly. Aiming at this problem, a method is proposed with keeping manifold structure and reducing the computational cost. It uses the multi-component discriminant locality preserving projections algorithm based on multiple distance microphones acoustic merging feature. Experimental results show that Diarization Error Rate(DER) of this algorithm can be reduced to below 20% and is better than traditional methods in most of the data set.
As T-tree has not good cache behaviors and update efficiency, this paper gives some optimization design for T-tree. The structure of T-tree’s node is redesigned according to the technology of cache structure and layout, and the precursor and successor pointers are added into the node structure. That can make T-tree more cache conscious and enhances its ability of range query. When there is a data overflow while updating, multiple data items rather than a data item are transferred between nodes. So the chance of data overflows and rebalancing trees can be reduced and the update performance is improved. Experimental results show that the improved T-tree has a better performance in query and updating and can save 18% main memory space.
With the fast development of Radio Frequency Identification(RFID) and tag usage, security has become the key issue for RFID related applications. In this paper, a new RFID secure protocol is proposed, which uses a synchronized random number created in the backend system, for tag authentication. After tag successful authentication each time, both tag and random number update synchronously. Meanwhile, the performance and security of the protocol are analyzed theoretically. Also the BAN logic is used to prove its security in further. The result show that this protocol can provide authentication between tag and reader, moreover, it can resist attacks such as replay, location tracking, traffic analysis, counterfeiting, denial of service, and illegal access, etc. From the implementation point, it just uses Hash and exclusive XOR operations, which reduces the computing complexity and makes it fitting for low price tags applications.
In order to improve the concurrent access performance of WebGIS system in a cluster environment, two level parallel processing mechanisms are proposed. One level is among each service node in the cluster and the other is within the service node. The calculation method of task response ratio, the request queue buffer mechanism and multi-thread scheduling strategy are focused on to improve system performance. Experimental results show that the WebGIS model can fully use the multi-core computing advantage of server nodes under the concurrent access environment. The more CPU core, the higher concurrent processing capabilities is enhanced.
This paper researches on Hybrid Human Immunodeficiency Virus(HIV) transmission mechanism, and proposes a dynamic mixed scale-free network generation algorithm based on the configuration model for investigating HIV transmission. It takes homosexual transmission network and heterosexual transmission network into the same system according to transmission law between man and man, man and woman, and forms a homosexual and heterosexual mixed-propagation network. In order to reflect the real social network evolution process, this algorithm adds the dynamic nature of network. The configuration model is presented to support scale-free rapid flexible and robust generations of dynamic mixed scale-free networks with given size and power law degree distributions. The dynamics of network is supported through network links dissolution and reformation. This configuration model avoids the shortcomings of the classic BA model. Simulation results show that the algorithm can correctly and effectively reflect the characteristics of the mixed-propagation network.
An improved speech enhancement algorithm is proposed by analysis the human auditory masking properties when a serious problems of residual musical noise brought by the Spectral subtraction in low Signal Noise Ratio(SNR). The gain parameters are adjusted by combined human auditory masking properties with wiener filter. Noise estimation is used by the Minimum Controlled Recursive Averaging(MCRA) algorithm in non-stationary environment. In order to further eliminate the musical noise, the optimal smoothing factor based on Minimum Mean Square Error(MMSE) is used to smooth the enhanced voice. Simulation results show that compared with the improved spectral subtraction and Wiener filtering method, the algorithm can effectively suppress background noise and residual musical noise as well as maintaining speech quality and intelligibility in low SNR.
Self-organizing linked lists perform well for handling requests with strong locality. Non-blocking algorithm can provide better robust and reliable. It uses MTF(Move-to-Front) concurrent rule to do concurrent operation, and presents the non-blocking self-organizing linked list algorithm which supports linearizable search, insert and remove operations using CAS primitive. Experimental results show that this algorithm performs well for long linked lists with high percentages of search operations. With the data sequence in experiment, when the percentage of lookup reaches 90% and the data range becomes 4 096, it provides the best performance.
The original Hidden Number Problem(HNP) usually works with a prime modules or a special form composite number modulus. To break through the limitation, this paper presents a method which can resolve the HNP in regular non-prime modulus case. It promotes the HNP in prime modulus case to the HNP in prime powers modulus case by introducing the Hensel-lifting method and lattice rounding technique in a creative way, and combines the Chinese remainder theorem to receive the achievement. Analysis result shows that the discrete logarithm computational problem under the Hensel-lifting situation can be reduced to the HNP in prime modulus case.
Aiming at the problem of high algorithm complexity, low monitoring efficiency and privacy disclosure for existed monitoring system of elderly people living alone, this paper proposes a monitoring system for elderly people living alone based on omni-directional vision. It uses Omni-directional Vision Sensor(ODVS) to get panoramic image, and designs a Motion History or Energy Images(MHoEI) algorithm for the motion object tracking. In the posture recognition, different human model algorithms are applied in accordance with the imaging characteristics of ODVS. For the behavior analysis, a calibrated one-to-one correspondence is adopted between the ground locations and the environmental factors. Both the reliability of behavior detection and the robustness of behavior detection are significant expanded. In addition, elderly activities are accurately obtained by using ODVS calibration and effective tracking for the human object. Experimental results show that this system has high robustness, instantaneity and accuracy rate of the detection. It is adequate to meet current needs by protecting privacy.