Dissertation > Excellent graduate degree dissertation topics show

Research on Time-Frequency Modeling and Enhancement for Speech Signal Based on Matching Pursuit

Author: GuoZuo
Tutor: YuFengQin
School: Jiangnan University
Course: Detection Technology and Automation
Keywords: Local cosine base model Matching Pursuit Algorithm Speech Enhancement Time-frequency analysis
CLC: TN912.3
Type: Master's thesis
Year: 2008
Downloads: 121
Quote: 2
Read: Download Dissertation

Abstract


In modern signal processing, speech signal is a typical class of non-stationary signals. The type of signal frequency changes over time, the traditional time domain and frequency domain analysis methods are not able to fully reflect the signal characteristics of time-frequency analysis is a powerful tool for analysis and processing of such non-stationary signals. It represents the signal to reveal the time-frequency distribution in a joint of the time domain and frequency domain information as a joint function of time and frequency, clearly we describe a distribution relationship of the signal frequency changes over time. From the time-frequency analysis point of view, the obvious difference in time domain and frequency domain characteristics and noise of the voice signal time domain, frequency domain characteristics, through the time-frequency modeling method can achieve the purpose of speech enhancement. In this paper, the voice signal modeling of local cosine bases. Construct local cosine bases dictionary, the segmentation of the timeline to adapt to the voice signal frequency changes in the structure, but also a higher time-frequency resolution. When using local cosine bases atomic frequency distribution of the speech signal time-frequency analysis. The distribution of inheritance Wegener - Willy aggregation has the advantage of time-frequency distribution of high-frequency, while avoiding the weaknesses of the existing cross terms in the Wigner - Willy frequency distribution. Simulation test, to verify the feasibility and effectiveness of using local cosine base model of the speech signal modeling. In a variety of enhancement algorithm based speech production model, how accurately extract model parameters is always difficult. This selection of adaptive signal decomposition method - matching pursuit method frequency model parameter extraction. Select the appropriate dictionary matching pursuit decomposition of the noisy signal, the noise in the time-frequency plane diluted speech component is relatively gathered at a regional time-frequency plane. Match tracking algorithm with adaptive parameters at each step will extract the atoms with the strongest correlation time to be decomposed signal frequency. The simulation results show that the local cosine bases of speech signal modeling based on matching pursuit, allows time-frequency characteristics of the speech signal can be gathered on the atoms in the dictionary, and then extract as much as possible from the noisy speech speech component and this speech enhancement method does not require a priori information about the statistical properties of the signal and noise. Matching Pursuit decomposition can be achieved by controlling the number of iterations to the purpose of filtering noise. However, when the signal-to-noise ratio is too low, the noise atomic energy atomic energy than voice, so matching pursuit decomposition noise atom is mistaken when voice atoms to break out. In this paper, the matching pursuit decomposition and subspace methods combined, with the noisy speech signal vector space can be constituted by a signal-plus-noise subspace and a pure noise subspace. Can use the signal subspace processing technology, the elimination of the pure noise subspace, and then to decompose the speech signal, speech enhancement. The simulation results show that the subspace matching pursuit decomposition effectively achieve the purpose of speech enhancement in the case of low signal-to-noise ratio and processing of colored noise.

Related Dissertations

  1. Research on Key Technologies of Digital Hearing Aids Based on Auditory Masking,TN912.3
  2. Single channel speech enhancement algorithm,TN912.35
  3. Band Entropy Method and Its Application to Fault Diagnosis of Rolling Bearings,TH165.3
  4. The Rearch on the Technology of MELPe Speech Coder,TN912.3
  5. Research on Subspace-Based Speech Enhancement,TN912.35
  6. Research of Speech Enhancement Algorithm under Non-stationary Environments,TN912.35
  7. Study and Application of Speech Enhancement Algorithm Based on Adaptive Spectral Estimation,TN912.35
  8. Design and Implementation of the seismic sequence of time-frequency analysis system,TP311.52
  9. Motor Imagery EEG Classifying Based on Parallel Factor Model,R318.04
  10. Analysis and Signal Extraction of Simulated Data for Moonquake Exploration,P184.5
  11. The Method of Time-frequency Analysis on Brier Slope Area Earthquake Research on the Application of Data Processing,P631.44
  12. Research on Ionosphere Contamination of High Frequency Radar Echoes and Time-Frequency Analysis Technology,TN958.93
  13. Study on Mechanical Fault Diagnosis Methods Based on local Mean Decompostion,TH165.3
  14. Heart sound noise reduction and segmentation method of heart sounds,R318.04
  15. Algorithm of Speech Enhancement Based on Auditory Masking and Implementation Based on DSP,TN912.35
  16. Speech Enhancement Algorithm Research Base on Short—term Spectrum Estimation,TN912.35
  17. Research on Speech Enhancement Under Noise of Music,TN912.35
  18. Research on the Recognition Algorithm of FH Signals,TN914.41
  19. Research and Application of Speech Enhancement Arithmetic,TN912.35
  20. Left and right hand motor imagery EEG feature extraction and classification,TN911.7
  21. The Research of Speech Enhancement Based on Wavelet Transform and Neural Network,TN912.35

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