摘要:In this paper, KL discreminant information of the modified Hidden Markov Model is proposed as a distortion measure for coarse classification in speech recognition. The feasibility of the classification method is proved, the clustering algorithm is given. The relationship between the rate of correct classification and the subset overlaping rate is shown in the experiments.
摘要:In this paper, we study three types of noisy speech enhancement algorithms based on Minimum Mean-Square Error Short-Time Spectral Estimation: Spectral Amplitude Estimation (MMSE-STSA), LOG-Spectral Amplitude Estimation (LOG-MMSE-STSA). and Relative Spectral Amplitude Estimation (MMSE-REL-STSA). On the basis of theoretical analysis, experimental studies are conducted. Computer simulations show that SNR improvement of 3.4-12dB can be obtained after enhancement processing when the original SNR of noisy speech degraded by Additional White Gaussian Noise ranges from 5dB to-10dB. Among the three algorithms, MMSE-LOG-STSA provides the best SNR improvement, and the SNR results coincide with informal listening.We also study the reduction of residual noise accompanied by the enhanced speech. By using the center clipping and gain function modification of estimator, the residual noise can be reduced significantly.
关键词:speech signal processing;speech enhancement;Short-time spectral estimation
摘要:In this paper, a new beamforming algorithm is proposed for reducing the signal cancellation in the presence of correlated interference. A new concept, the Split-Polartiy Transformation (SPT), is introduced for decorrelating desired signal from interference. The design of SPT matrices is discussed. The proposed algorithm is applicable to arbitrarily configured arrays and robust against directional mismatch and array model errors to within some tolerance. Computer simulation results have demonstrated the effectiveness of SPT-LCMV beamforming algorithm.
Li Chunsheng, Li Jingwen, Zhou Yinqing, He Junxiang, Feng Shizhang
Issue 4, Pages: 21-25(1993)
摘要:Based on the eigenstructure decomposition method, a new kind of approach to high resolution space-time two dimensional signal spectrum estimation is presented in this paper. The signal subspace and the noise subspace are formed under eigenstructure decomposition of the space-time signal covariance matirx. By using the orthogonal properties of the two subspaces, we can gain a high resolution space-time spectrum. The presented method. Isapplied to phased array radars and can accurately get the information about the velocities and azimuth angles of spatial moving targets. The computer simulation results illustrate its availability.
关键词:Space-time two-dimensional signal processing;spectrum estimation;array signal processing;radar signal processing
摘要:In this paper the power spectral density of p-adic stationary stochastic prosess under the sence of Chrestenson transform and its maximum entropy estimator are studied. The relationship formula between the power spectral density and the entropy rate and the normal equations of maximum entropy spectral estimator which is expressed in finite form are obtained. When the number of sampling data is p, the maximum entropy estimator can be directly expressed by the known finite autocorrelation data. These results are much different from those of Fourier’s.
摘要:Under assumption on separable input processes, using the block-oriented model of nonlinear systems, a generalized Wiener-Hopf equation on adaptive nonlinear systems are derived in this paper. Based on this, the analytical solutions of inverse model of some nonlinear systems are obtained. Similar to common LMS algorithm, a useful algorithm used in inverse adaptive modeling of the nonlinear system is proposed. It is shown by simulations that weight values converge to the analytical solutions closely to the proposed algorithm.
摘要:A neural network model with the global convergence is proposed and studied in this paper. The dynamical property of this model is described by dual differential equations and it can find out several stable equilibrium points from any initial point in the state space. This model Ican be extended easily to suit for the problems which need the global convergence such as the global optimization problems.
关键词:neural networks;global convergence;Dual dynamic;energy function
摘要:This paper presents a matching and estimation algorithm for 3-D rigid body motions, integrating the procedures of the point match and motion estimation into a global optimization process. It solves the problems of matching, segmenting and estimating in the 3-D motions, extends the idea of first-match-then-estimate, and is suitable for real-time processing. A large number of experimental results have been described to demonstrate the efficiency of the algorithm.
关键词:3-D rigid body motion;Point match;Motion parameter estimation;Global optimization
摘要:With AI technology in searching, Xu, et al. presented a global optimum feature selection algorithm, BF, which is computationally superior to B&B one. Based on the analysis of search tree TB used in BF, this paper proposes a new structure of search tree, Tb, on which there are fewer nodes than on TB. A feature selection algorithm on Tb, referred as to BF, is designed, of which the computational complexity is lower than BF’s, without increasing storage and with global optimum property.
摘要:This paper puts forward to a multi- structure element technique for morphological shape decomposition. Its performances are beffer than single ones’. The basis of the method proposed relies upon the generaton of set of non-over lapping segments of an image via repeated basic morphological operations. The maximum core decomposition is chosen to guarantee the uniqueness. The searching scheme for optimal decomposition is based on heuristic search strategy. The image coding via morphotogical transformation is also discussed and the experimental results have shown the greot advantages of multi-element technique.
摘要:A new hybrid coding scheme with subsampled DPCM and interpolative DP-CM is proposed in this paper. It may be utilized to 140 Mb/s transmission of HDTV signals without involving high speed complex processing. Computer simulation results show that high quality of reconstructed pictures can be obtained by the proposed coding scheme.
摘要:In this paper, the throughput of unslotted FH-CDMA system is evaluated. As the channel traffic and the number of frequency slots, denoted by μ and q respectiv ely, approach infinity while μ/q remains constant, it is shown that the throughput of unslotted FH-CDMA system is the same as that of slotted system, and the maximum throughput of synchronous frequency-hopping system is twice that of asynchronous system. Besides, the numerical results show that it is a better way to employ unslotted system as compared with slotted system.
摘要:Output queueing is the optimal switch of fast packet switching in performance. In this paper, the queueing performance of output queueing with arbitrary priority classes is studied. We derive the statistical characteristics, such as average queue length of any priority packet in independent and dependent packet arrival cases. It is found that for an N xN output queueing switch, the queueing performance of both cases tends to be the same if N is large enough. A numerical iteration is presented to get the stationary queue length distribution of high and low priority packets with queueing equation in two dimensional Markov process. The analyses are proved correct by computer simulation.
摘要:HF channel’s burst-interference may destroy the equalization badly and makes the equalizer ineffective. This paper presents two methods to prevent equalizer from out of working and make equalization recover quickly. So the whole communication system can work successively. Computer simulation proves effectiveness of both methods.
摘要:<正> 第三届中国国际通信学术会议(1992 Interna-tional Con ference on Communication techno-logy,ICCT’92)于1992年9月16日至18日在北京清华大学召开。这次国际会议由中国电子学会、中国通信学会和清华大学联合发起,并得到IEEE通
摘要:Different Schemes of intrafield prediction coding of HDTV component signals Y, U, V are discussed in this paper. According to the “Large Error Criterion”, the coding scheme is determined. And two nonlinear quantizers of luminance and chrominance with 16 levels are designed by the subjective tests. The subjective tests show that it is difficult to distinguish the reconstructed picture from the original one.
摘要:In the paper, a biological neuron is simulated by means of MOS circuits, and a artificial neural network is constructed. In addition, a method of adjusting and measuring the neural circuit is given. The result of measurement shows that these circuits can simulate the biological function of neural cells quite well.
摘要:In this paper, it will be verified that many current spectral estimation methods are corresponding to a same mathematic problem. A common matrix form for spectral estimation methods will be given and a series of new approaches to spectral estimation can be extended from it. As an example, procedures based on the eigenvector space of the correlation matrix are investigated.
关键词:spectral estimation;Common matrix form;Eigen-space spectral estimation;singular value decomposition