Loading...

Analysis and Detection of Pitch Contour

Choupan, Jeiran | 2010

2421 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: English
  3. Document No: 41052 (52)
  4. University: Sharif University of Technology, International Campus, Kish Island
  5. Department: Science and Engineering
  6. Advisor(s): Ghorshi, Mohammad Ali; Mortazavi, Mohammad
  7. Abstract:
  8. Pitch or fundamental frequency is one of the components of speech production. Pitch detection is the process of determining the period of the vocal cords closure or in other words, the time duration of one glottal closing, opening and returning phase. There are different algorithms for pitch detection purposes, each of which has some advantages and disadvantages. Pitch detection algorithms can be classified into event pitch detection and non-event pitch detection. Dyadic wavelets transform is an example of event pitch detection. Modified higher order moment, which is based on the Autocorrelation function, is non-event pitch detection. Dyadic wavelets transform is a fast and simple pitch detection method; however, it has less accuracy compared to modified higher order moment. On the other hand, modified higher order moment has high computational complexity and is time consuming. In this thesis, we propose a pitch detection method based on Dyadic Wavelets Transform, enhanced with Modified higher order moment. This method is called Modified Dyadic Wavelets Transform and applies Wavelet Transform to the whole speech signal in the first step. Then, Modified higher order moment is applied only to a specific part of transformed speech signal to improve the accuracy while keeping the computational complexity low. Therefore, the proposed method is more accurate than Dyadic wavelets transform, and is faster than the Modified higher order moment method. Finally, our evaluations indicate that the new pitch detection method improves the Dyadic wavelets transform method in terms of accuracy and robustness to noise. It was also found that modified DyWT has about 50% reduction in overall error compared to that of DyWT and can be very useful for real-time purposes
  9. Keywords:
  10. Speech ; Pitch Detection ; Autocorrelation Function ; Dyadic Wavelets Transform ; Modified Higher Order Moment

 Digital Object List

 Bookmark

  • 2.pdf
    • Chapter 1
    • Introduction
      • 1.1 Pitch Detection
      • 1.2 Motivation
      • 1.3 Contribution
      • 1.4 Thesis Content
      • Chapter 2
      • Speech Production and Perception
      • 2.1 Speech Production and Various Properties of Speech Signals
        • 2.1.1 Lungs
        • 2.1.2 Larynx
        • 2.1.3 Vocal Tract
      • 2.2 Speech Sound Classification
      • 2.3 Vowel Production Model
      • 2.4 Pitch Overview
        • 2.4.2 Pitch Theory
        • 2.4.3 Pitch Tracking and Detection
        • 2.4.4 The Difficulty of Pitch Tracking
        • 2.4.5 History of Pitch Determination
        • 2.4.6 Applications of pitch estimation
      • 2.5 Summary
    • Chapter 3
    • Related Work
      • 3.1 Pre-Processing
        • 3.1.1 Centre Clipping
        • 3.1.2 Low-Pass Filtering
        • 3.1.3 Inverse Filtering
        • 3.1.4 Windowing
      • 3.2 Non-Event Pitch Detection methods
        • 3.2.1 Time Domain Waveform Similarity Method
          • 3.2.1.1 Zero-Crossing
          • 3.2.1.2 Autocorrelation Function (ACF)
          • 3.2.1.3 Average Magnitude Difference Function Method (AMDF)
        • 3.2.2 Frequency Domain Spectral Similarity Methods
          • 3.2.2.1 Cepstrum Method
      • 3.3 Event Pitch Detectors
        • 3.3.1 Wavelet Transform
      • 3.4 Post-Processing
      • 3.5 Summary
    • Chapter 4
    • Method
      • 4.1 Algorithm Outline
        • 4.1.1 Window Selection
        • 4.1.2 Dyadic Wavelet Transform
        • 4.1.5 Determination of Pitch Period
        • 4.1.6 Post-Processing
      • 4.2 Block Diagram of Modified DyWT
      • 4.3 Result
      • 4.4 Summary
    • Chapter 5
    • Evaluation and Results
      • 5. 1 Introduction
      • 5.2 Brief Description of the Test Database
      • 5.3 Evaluation Measures and Types
        • 5.3.1 Accuracy
        • 5.3.2 Robustness
        • 5.3.3 Computational Complexity
      • 5.4 Comparison between Different Methods
        • 5.4.1 Optimal Higher Order Moment Evaluation
          • 5.4.1.1 Accuracy
          • 5.4.1.2 Robustness
          • 5.4.1.3 Computational Complexity
        • 5.4.2 Comparison between Modified 4th Order Moment and DyWT
          • 5.4.2.1 Accuracy
          • 5.4.2.2 Robustness
          • 5.4.2.3 Computational Complexity
        • 5.4.3 Comparison between Modified DyWT and DyWT
      • 5.5 Summary
    • Chapter 6
    • Conclusion
...see more