Dissertation > Excellent graduate degree dissertation topics show

Research on Speaker Characteristic Parameter Extraction in the Noise Environment

Author: JiangCheng
Tutor: JiangZhiFang
School: Shandong University
Course: Software Engineering
Keywords: Speaker Recognition Pitch Detection LMS algorithm Voice activity detection Adaptive time-frequency parameters
CLC: TN912.3
Type: Master's thesis
Year: 2009
Downloads: 79
Quote: 0
Read: Download Dissertation

Abstract


Speaker recognition is an important part of the speech signal processing, which is one of the current research focus. It is a speech waveform reflects speak physiological, psychological and behavioral characteristics of the voice parameters to automatically recognize the identity of the speaker biometric authentication technology. Speaker recognition lies in the personal feature extraction and recognition model, which personal characteristics extraction is the most critical. Since the introduction of various uncertain noise in the real environment, generally makes the characteristic parameter extraction becomes more difficult, so that the speaker recognition system performance dropped significantly. Therefore, the extraction of characteristic parameters of the speaker in a noisy environment is very important significance. In this paper, on the basis of analysis of the speaker characteristic parameters to study the three noise environments, the new speaker feature parameter extraction algorithm, given a specific algorithm processes were simulated in MATLAB software environment. Experimental results show that the new detection algorithm is significantly better than the conventional method, to reduce the detection error robustness. First, the article simply describes the basics of the digitized voice signal processing, voice signal, the frequency domain processing method to analyze the common pitch period, endpoint detection and Mel down the performance of the pedigree and the number of speaker characteristic parameters. The researchers then the speaker characteristic parameters of pitch period extraction method, traditional autocorrelation function (ACF) and the average magnitude difference function (AMDF) the advantages and disadvantages. LMS adaptive filtering, proposed an improved based LMS adaptive filtering and ACF / AMDF weighted square features pitch detection algorithm. Experimental results show that the detection results of the algorithm in a noise environment significantly superior to the conventional method, to effectively suppress the influence of the resonance peak, improved pitch detection accuracy, and the calculation complexity is low, suitable for speech synthesis and coding the real time processing. Endpoint detection method of short-time average energy, short-time average zero-crossing rate method and spectral entropy method is described in detail. Combined with spectral subtraction enhancement principle, proposed an improved speech endpoint detection based on spectral subtraction and adaptive sub-band energy spectrum entropy algorithm. The algorithm uses an improved spectral subtraction enhancement processing of the speech signal with noise, and then extract the voice signal adaptive sub-band energy spectrum entropy characteristics of voice endpoint detection. The simulation results show that the algorithm has a good performance of the detection algorithm is simple, strong environmental adaptability. Further, from the human auditory system, the Mel scale filter bank, when a the Mel domain of adaptive frequency parameters of the algorithm process. The experimental results show that this parameter can effectively distinguish between speech and noise frequency band, the recognition rate than traditional MFCC parameters improved significantly less dependent on the signal-to-noise ratio, and has a good noise robustness. Finally, the proposed algorithm needs to be improved in the future work, made a brief introduction and outlook on the direction of future research, pointed out that the prospects for the development of the next speaker feature parameter extraction.

Related Dissertations

  1. Second-order Circulation Statistics-based Array Signal Processing,TN911.7
  2. Phase-coded signal pulse technology research and implementation,TN957.51
  3. 14-bit 100 MB samples / sec pipelined ADC Research and Design,TN792
  4. MFCC -based speech recognition system to improve research and design,TN912.34
  5. Complex channel speaker recognition technology,TN912.34
  6. Research on Speaker Recognition System under the Visual C++ 6.0,TN912.34
  7. Research on Adaptive Narrowband Suppression Techniques in Spread Spectrum Communication Systems,TN914.42
  8. Recognition of Pulse Spectral Signal and Design of Software,TN24
  9. Research on the Algorithm of Speeth Pitch Detection,TN912.3
  10. Mixing characteristics and Gaussian mixture model - based speaker recognition,TN912.34
  11. Design and Implementation of Speaker Recognition System Based on Windows CE,TN912.34
  12. Study of Extraction and Optimization Characteristic Parameters in Speaker Recognition,TN912.34
  13. The Research Based on the Text Irrelevant Speaker Distinguishes,TN912.34
  14. Research and Implementation on Real-time Speaker Recognition Algorithmin in Multiplexing Parallel Model,TN912.34
  15. Research on Shunt Active Power Filter Based on Adaptive Algorithm,TN713.8
  16. A Study for Discretization of Real Value Attributes and LMS Algorithm,TP18
  17. The Design of Computer Software in Piano Tuning,TP368.1
  18. The Development of Speaker Recognition System Based on Support Vector Machine,TN912.34
  19. Minority language recognition research based telephone voice,TN912.34
  20. Research on Text-independence of Open-set Speaker Recognition,TN912.34
  21. Speaker Recognition Research in Noisy Environment,TN912.34

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