Title Page
Contents
ABSTRACT 6
1. Introduction 7
1.1. Background 7
1.2. Suggestions 9
1.3. Structure of Thesis 10
2. Related Works 11
2.1. Emotion Recognition Technologies 11
2.2. Gender Recognition Technologies 14
2.3. Age Recognition Technologies 16
3. Proposal for Emotion, Gender and Age Recognition Method 19
3.1. System Architecture 20
3.1.1. Emotion Recognition Using Spectral Entropy 20
3.1.2. Gender Recognition Using Spectral Entropy 21
3.1.3. Age Recognition Using Spectral Entropy 22
3.2. Feature Extraction 24
3.2.1. Pitch 25
3.2.2. Entropy 27
3.2.3. Delta and Delta-Delta FFT Spectral Entropy 28
3.2.4. MFB, Delta MFB and Delta-Delta MFB Spectral Entropy 31
3.3. Emotion, Gender and Age Recognition Using GMM 36
4. Experiments and Results 40
4.1. Experimental Environment 40
4.2. Experimental Results 43
4.2.1. Performance Evaluation of Emotion Recognition 43
4.2.2. Performance Evaluation of Gender Recognition 47
4.2.3. Performance Evaluation of Age Recognition 51
5. Conclusion 56
References 57
논문요약 63
List of Tables Table 4.1. Consist of age speech database 41
Table 4.2. Recognition using general feature 43
Table 4.3. Recognition using general and delta general feature 43
Table 4.4. Recognition using general, delta and delta-delta general feature 44
Table 4.5. Experimental results using FFT spectral entropy 45
Table 4.6. Experimental results using FFT and Delta FFT spectral entropy 45
Table 4.7. Experimental results using FFT, Delta and Delta-Delta FFT spectral entropy 45
Table 4.8. Experimental results using MFB spectral entropy 46
Table 4.9. Experimental results using MFB spectral entropy and Delta MFB spectral entropy 46
Table 4.10. Experimental results using MFB, Delta and Delta-Delta MFB spectral entropy 47
Table 4.11. The recognition accuracy of the acoustic analysis approach 48
Table 4.12. The recognition accuracy of the pitch estimation approach 48
Table 4.13. The accuracy of the combining approach 48
Table 4.14. Experimental results using MFB spectral entropy and pitch 49
Table 4.15. Experimental results using MFB spectral entropy, delta MFB spectral entropy and pitch 49
Table 4.16. Experimental results using MFB spectral entropy, delta MFB spectral entropy, delta-delta MFB spectral entropy and pitch 50
Table 4.17. Age groups with spans of 5. years and the amount of used training and test recording 51
Table 4.18. Experimental results using Pitch and FFT spectral entropy 53
Table 4.19. Experimental results using Pitch, FFT spectral entropy and Delta FFT spectral entropy 53
Table 4.18. showed experimental results using FFT spectral entropy based age recognition method. Table 4.19 showed experimental results using FFT spectral entropy 53
Table 4.20. Experimental results using Pitch, FFT spectral entropy, delta FFT spectral entropy, delta-delta MFB spectral entropy, and delta-delta FFT spectral entropy 54
Table 4.21. Experimental results using Pitch and MFB spectral entropy 54
Table 4.22. Experimental results using Pitch, MFB spectral entropy, and delta MFB spectral entropy 55
Table 4.23. Experimental results using Pitch, MFB spectral entropy, delta MFB spectral entropy, and delta-delta MFB spectral entropy 55
List of Figures
Figure 3.1. Block Diagram of Emotion Recognition 20
Figure 3.2. Block Diagram of Gender Recognition 21
Figure 3.3. Block Diagram of Age Recognition 22
Figure 3.4. Delta FFT spectral Entropy and Delta MFB Spectral Entropy 24
Figure 3.5. Basic-AMDF in a voiced speech 26
Figure 3.6. Delta FFT spectral Entropy MATLAB c7ode 30
Figure 3.7. "diff" MATLAB Function 31
Figure 3.8. Mel Scale 32
Figure 3.9. Delta MFB Spectral Entropy MATLAB code 35
Figure 3.10. Block Diagram of GMM Parameter Training Processing 38
Figure 3.11. Block Diagram of Emotion Recognition Processing Using GMM 39
Figure 4.1. Estimation Error Rate when Using Groups with Spans of 5. Years 52