Dissertation > Excellent graduate degree dissertation topics show

Based on artificial neural network model voice conversion algorithm

Author: ChenZhi
Tutor: ZhangLingHua
School: Nanjing University of Posts and Telecommunications
Course: Signal and Information Processing
Keywords: Voice conversion Spectral envelope conversion Pitch frequency conversion Artificial Neural Network Model Quantum Genetic Algorithm
CLC: TN912.34
Type: Master's thesis
Year: 2011
Downloads: 30
Quote: 2
Read: Download Dissertation

Abstract


Voice conversion is performed by a speaker (source speaker) voice personality characteristics into another speaker (target speaker) voice personality characteristics, while maintaining the content of speech and emotional features of a technology change. After conversion to get the sound you want to sound like the target speaker's voice, while keeping the contents of the source speaker's speech and emotional characteristics unchanged. The technology has important theoretical value and good prospects. This paper mainly focus on voice conversion of baseband conversion and spectral envelope trajectory parameters into these two key technologies research, the main work and innovations are as follows: (a) the current trajectory of existing baseband conversion algorithms compare the experimental study, Current conversion algorithms are found in a simple linear transformation, and in fact between the two speakers baseband nonlinear relationship trajectory. For the problems of traditional conversion algorithm (linear transform baseband conversion), proposed RBF neural network baseband track conversion algorithm, divided by the fundamental frequency trajectory data segments of equal length, and then be modeled separately, looking between the source and target mapping rules in order to achieve the fundamental frequency track conversions. Subjective and objective tests show that: the algorithm not only improves the accuracy of the conversion parameters, but also enhance the naturalness of synthesized speech. (2) The traditional voice conversion algorithms are the segment information parameters and suprasegmental information open to independent parameters, namely the conversion, and finally with the synthesized speech translation. And more and more studies show that hidden between large correlation parameters can be extracted from one another parameter, be open to independent conversion method, respectively, is bound to undermine both the links between affect conversion results. For this problem, the fundamental frequency parameter and the spectral information parameters combine to form short-term spectral parameters combined together, as the characteristic parameter for training and conversion. Experiments show that: the improved transition effects are superior to the traditional algorithm under the same conditions. (3) through the traditional neural networks for voice conversion algorithm, transition effects found to affect the neural network is a key factor in the central value of the hidden layer and the weight matrix. Algorithms for solving the key factor to improve and enhance the accuracy of the conversion is bound to improve neural network transition effects, based on this analysis, the use of quantum genetic algorithm optimization neural network, thus achieving voice conversion. Experimental results show that the subjective and objective tests: improved algorithm based on the conversion of voice both to improve the degree of similarity between the target voice, but also enhances the synthesized speech intelligibility.

Related Dissertations

  1. Research on DNA Encoding Based on Quantum Computing,Q75
  2. The Application Research of Optimal Design of Electromagnetic Devices,TM503.1
  3. Study on Optimization Modeling and Intelligent Control of Alcohol-based Fuel Boiler Combustion System,TK223.2
  4. Research of QGA on the Mechanical Optimization Problems,TP18
  5. Research on Intelligent Optimization Algorithm and Its Application in Communication,TN92
  6. The Research on Intrusion Detection Based on Quantum Genetic Algorithm Optimization BP Network,TP393.08
  7. The IDS detection algorithm and technology research,TP393.08
  8. Research on Multi-Point Non-Sequential Deformation Forecast Model Based on BP Artificial Neural Network,P258
  9. Quantmtum Gentic Algorithm and Its Application Covngive Radio Spectrum Allgication Based on Geme Theory,TN925;O225
  10. Research on Power Control in Congnitive Radio Based on Quantum Genetic Algorithm,TN925
  11. Quantum Intelligent Algorithm and Its Application,TN912.34
  12. Research on Energy-Optimization of Wireless Sensor Networks Based on Quantum Genetic Algorithm,TP212.9
  13. Voice Changer alto section features research modeling and transformation,TN912.3
  14. Research on Technologies of Voice Conversion Based on Gaussian Mixture Model,TN912.3
  15. Quantum Clone Genetic Algorithm Based Multi-User Detecter,TN929.533
  16. Application Research on Fault Diagnosis of Power Electronic Circuits Based on BP Neutral Network Optimized by QGA,TN710
  17. The Application of an Improved PSO Based on the Quantum Genetic Algorithm in the Submersible Path-planning,TP18
  18. Research on Uncertain Production Scheduling Based on Improved Quantum Genetic Algorithm,TP18
  19. SME credit risk management research,F276.3
  20. Research on the Gas Forecast Model Owing to That BP Neural Networks and Quantum Genetic Algorithm,TP183

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