Search

First of all, This paper introduces the construction of NAPS kernel. Then a recognition algorithm which combines the NAPS kernel with general hyper-plane SVM as well as its application to speaker identification is studied, and the problem encountered is given. To overcome the problem, an algorithm based on the NAPS kernel hyper-sphere SVM is proposed and is applied to speaker identification. Experiments and theoretical analysis show that the algorithm proposed is effective.

This paper discusses the FPGA implementation of modified Goertzel Algorithm for the application of DTMF signal detection, which is as effective as that of normal Goertzel Algorithm yet is more resources saved than that of normal Goertzel Algorithm. Detail algorithm simulation and algorithm FPGA implementation are given in the paper.

Vector quantization is a useful method that had been applied to diverse fields such as speaker recognition. This paper describes in detail how to get the optimum vector quantization codebook for the use of speaker recognition. Optimization depends on specific criterion or conditions. The optimization discussed here includes three level optimizations. Level one is locally optimization, level two is globally optimization, level three is personally optimization for speaker recognition.

A novel approach to isolated word recognition based on difference subspace is firstly introduced in this paper. Then a modification is made in order to improve the robusticity of the algorithm. The theoretical analysis on the performance of the improved algorithm is then presented in detail. The results of both theoretical analysis and experimental study show that the improved algorithm is an efficient and robust approach to isolated word recognition.

利用同步辐射光电发射谱研究了Co与CH\-3CSNH\-2处理的S钝化GaAs(100)的界面形成.发现 其界面反应较弱，Co覆盖层达到0.8nm时，形成稳定的界面.GaAs表面上和S原子形成桥 键的Ga原子与Co发生交换反应并扩散到覆盖层中，形成Co—S键.Co覆盖层表面无偏析As的出现，与Co/GaAs(100)界面不同，这表明GaAs表面的S钝化可有效地阻止As原子向覆盖层的扩散.