Dissertation > Excellent graduate degree dissertation topics show

RBF neural network optimized for speech recognition research

Author: LiGaoYun
Tutor: ZhangXueYing
School: Taiyuan University of Technology
Course: Signal and Information Processing
Keywords: RBF neural network Adaptive Genetic Algorithm Fuzzy Systems Fuzzy rules Membership function Speech Recognition
CLC: TN912.34
Type: Master's thesis
Year: 2009
Downloads: 140
Quote: 7
Read: Download Dissertation


RBF neural network has a simple structure, classification performance, learning speed, generalization ability and avoiding local minima, etc., so in speech recognition field to get more and more attention and application. For the RBF neural networks, basis function neural network RBF centers important parameters, their values ??on the network performance is greatly affected. However, the traditional RBF neural network learning algorithm often converges to local optimum value. RBF neural networks and fuzzy logic system can achieve a good complement, enhance learning neural network generalization ability. Therefore, in order to further improve the performance of the traditional RBF network, this paper on the basis of the previous work proposes two improved algorithms. First, we use the global search performance with training RBF neural network genetic algorithm to determine the basis function center. Genetic algorithm crossover and mutation probability of selection is the impact of genetic algorithm behavior and performance of the key, directly affect the convergence of the algorithm. Therefore, this paper introduces natural number coding genetic algorithm, based on the specific circumstances of individual adaptively adjust the crossover probability and mutation probability. Using genetic algorithms to optimize the RBF network, enhanced network global optimization capability, improved network pattern recognition performance. The improved algorithm is applied to speech recognition system, the experimental results show that the recognition result of using this method over using K-means clustering algorithm selected RBF network centroid recognition results. Secondly, the paper constructs based on structural equivalence type fuzzy RBF neural network speech recognition system. Proposed a structure equivalent fuzzy RBF neural network structure and learning algorithm, using five neural network structure to achieve fuzzy systems and fuzzy inference rules, neural network parameters correspond to all nodes and membership functions of fuzzy systems and reasoning process. The network uses the RBF neural network and fuzzy inference system equivalent feature that can automatically determine the number of fuzzy rules and membership functions, how to solve the fuzzy system automatically generates and adjust membership functions and fuzzy rules puzzle. Based on the number of words to be recognized automatically select the number of fuzzy rules, using supervised clustering and gradient descent method to train the system parameters in two steps, the convergence rate than all the parameters are trained using gradient descent algorithm or the membership function center uses K-means clustering algorithm is much more to learn fast and high recognition accuracy. Simultaneously with the RBF neural network algorithm are compared, experiments show that the method has a high recognition rate and robustness, and adaptability to different speech characteristics, further research can be used as the basis for anti-noise speech recognition. Finally, this paper introduces the theory of wavelet transform, wavelet function as structural equivalence fuzzy RBF neural network membership function, the experimental results show that the neural network can also get good recognition results.

Related Dissertations

  1. Multiple ANN/HMM Hybrid Used in Speech Recognition,TN912.34
  2. The Design of a DSP-Based Robot Speech Command Recgnition System,TN912.34
  3. The Design and Research of Health Management Based on Smartphone Environment,TN929.53
  4. Study on Risk Identification and Evaluation of Manufacturing Green Products R & D,F205;F224
  5. Application of the Fuzzy Technology in Temperature Control,TP273.4
  6. Research on Image Segmentation Method Based on Measure of Medium Truth Degree,TP391.41
  7. The Research and Design of Pressure Cooker System Based on Fuzzy Neural Network,TM925.5
  8. Mobile audio and video interactive platform of business execution,TN915.09
  9. A new generation of versatile disc tracking servo control study,TP333.4
  10. AC Drive Sliding Control Algorithm,TP273
  11. Study on Fault-tolerant Guaranteed Cost Control for Nonlinear Systems,TP13
  12. Situaiton Assessment Using Probabilistic Graphical Models,E917
  13. The Application Research for Fuzzy Synthesis Evaluation in Green Building Evaluation,TU201.5
  14. Based on RBF artificial neural network in the application of PCB drilling process,TN405
  15. Networked Control System Fault Diagnosis and Fault Tolerant Control,TP273
  16. Linear guide system Elevator single electromagnetic levitation RBF neural network sliding mode control,TP273
  17. Optimization algorithm based on artificial intelligence Melt Index Prediction Modeling Optimization,TQ325.14
  18. Path Planning of Robot Systems,TP242
  19. Priority -order matching method of billet steel companies,F426.31;F224
  20. Five kinds of papaya salt tolerance of plants,S685.99
  21. Robust and adaptive unmanned helicopter Tolerant Control Technology,V249.1

CLC: > Industrial Technology > Radio electronics, telecommunications technology > Communicate > Electro-acoustic technology and speech signal processing > Speech Signal Processing > Speech Recognition and equipment
© 2012 www.DissertationTopic.Net  Mobile