Dissertation > Excellent graduate degree dissertation topics show

Application of MMCE Algorithm to Factor Analyzed Probability Statistic Models

Author: WuYanQu
Tutor: ZengYiCheng
School: Xiangtan University
Course: Physical Electronics
Keywords: Factor analysis MMCE algorithm FAGMM+MMCE model FAHMM+MMCE model
CLC: TN912.34
Type: Master's thesis
Year: 2009
Downloads: 35
Quote: 0
Read: Download Dissertation

Abstract


The speaker recognition based on models is an effective method in the field of speaker recognition, and probability statistical models belong to such methods. GMM and HMM models are typical probability statistical models. At present, the most extensive research has been obtained, and there are increasingly important research significance and practical value.In recent years, the speaker probability statistical models has become a major area of research hot spots. However, when applied to the actual environment, there are many problems including: the intra-related correlation of feature vector; the poor classification for too many model parameters in large text; complex calculations and large overhead of the system, as well as the slow training speed; the poor optimization and inflexible of training algorithm classification, and so on. In response to these problems, we carry out the following tasks:The probability statistical models commonly used in speaker recognition: Hidden markov model (HMM) and Gaussian mixture model (GMM) are discussed, and their advantages and disadvantages of basic algorithms are deeply studied and discussd in detail. On this basis, the minimum classification error(MCE) algorithm and a detailed analysis of the advantages and disadvantages of the algorithm are introduced. We attempt to put forward the modified MMCE algorithm for the problems of the widely used MCE in GMM and HMM. In order to increase training flexibility, duplicate discriminant of the loss function is replaced by an orderly group, avoiding duplication and complexity of the comparison calculation.Realization process of MMCE algorithm and parameters mediation are derived. The formation process and the concept of combinating factor analysis and GMM/HMM into FAGMM and FAHMM are disscused. The EM and the MMCE training algorithm are derived, and the combination of MMCE and FAGMM/FAHMM is achieved, which forms the new models of FAGMM+MMCE and FAHMM+MMCE. The speaker experiments based on 50 personal voice library show that: MMCE has better performance and faster training than MCE and EM algorithm in the speaker model. Moreover, the experiments of anti-noise performance of the FAGMM+MMCE and FAHMM+MMCE models also show that: we separately verify the recognition performance of the model under the white noise with different signal to noise ratio and various types of common real noise environment, and they have better anti-noise performance than other methods.In this thesis, the new FAGMM+MMCE and FAHMM+MMCE models are studied. The models not only enchance the recognition rate and training speed, but also obtain better anti-noise performance.

Related Dissertations

  1. Studieson Effects of Soybean Species on Yuba and Initial Establishment of Quality Evalution System for Yuba,TS214.2
  2. Chinese hog industry ( chemicals ) International Competitiveness,F752.6;F224
  3. Based on Factor Analysis of the chemical industry trade Financial Performance Evaluation of Listed Companies,F224
  4. Study on Risk Identification and Evaluation of Manufacturing Green Products R & D,F205;F224
  5. An Study on the Efficiency of the Listed Real Estate Companies Based on Factor Analysis Method,F224
  6. A Research on the Evaluation of Financial Competitiveness for Listed Companies Related to Agriculture,F832.51;F224
  7. Research on State Diagnosis on Fan Based on Factor Analysis and BP Neural Network,F426.61
  8. Project management system implementation of key factors,F270.7;F224
  9. Affect the effectiveness of classroom teaching high school chemistry factor analysis and empirical research,G633.8
  10. Shenyang scale industrial enterprises technological innovation ability evaluation,F425;F224
  11. Investment in high-tech industries in Hebei Province and Countermeasures of Environmental Assessment,F276.44
  12. A Study on M&A Performance and International Competitiveness of Zhejiang’s Listed Company,F224
  13. Automotive Competitiveness Analysis and Evaluation,F224
  14. Drug on the chemical composition of dried ginger - galangal volatile oil and Salvia miltiorrhiza fingerprinting studies,R284
  15. Empirical Research of Our Country Commercial Bank Management Performance Evaluation,F224
  16. Study on Mental Health and Influencing Factors of the Three Gorges Migrants in Shandong Province,R395.1
  17. The Evaluation of Liaoning Software Industrial Competitiveness,F426.672
  18. The Study on the Development and Countermeasure of Xinjiang’s Cotton Logistics in China,F326.6
  19. The Factor Analysis of Hebei Province Absorbing Rural Surplus Labors,F323.6
  20. An Empirical Research on Affacting Factors of FDI of Hebei Province,F832.6
  21. Credit Evaluation and Risk Management of Local Government Investment and Financing Platform,F124

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