표제지

목차

국문초록 9

제1장 서론 10

제2장 확률 모형 12

2.1. 확률 과정 정의 12

2.2. 신호 분석 모델링 13

2.2.1. 음원의 정의 13

2.2.2. 성도 필터 14

제3장 음원 이론과 신호 변환 16

3.1. 음원 필터 이론 16

3.2. 푸리에 변환 17

3.3. Z-변환 22

3.4. 웨이블릿 변환 23

제4장 푸리에 변환 계산 27

4.1. 시간 영역의 푸리에 계산 성질 27

4.1.1. 시간 적분 법칙 27

4.1.2. 시간 이동 법칙 27

4.2. 주파수 영역의 푸리에 계산 성질 28

4.2.1. 주파수 미분 법칙 28

4.2.2. 푸리에 변환쌍 법칙 28

제5장 음성 분석과 실증 연구 30

5.1. 음성 필터 30

5.2. 캡스트럼 피크 현저성 33

5.3. 실증 연구 36

제6장 결론 44

참고 문헌 47

영문초록 49

Table 5.1. Descriptive statistics of Data 36

Table 5.2. Comparison of results by smoothing measures 40

Table 5.3. Comparison of results by measures of feature extractions for all participants 41

Table 5.4. Comparison of results of sex group by measures of feature extractions, for Gender ＝ Male, Female 42

Table 5.5. Comparison of results of age group by measures of feature extractions; for age : lower(age ≤ 30); middle(30 ≤ age ＜ 70); upper(age ≥ 70). 43

Table 6.1. Descriptive statistics of CPP, Group ＝ Normal, Depressed 44

Table 6.2. Results of evaluation metrics with regression 45

Table 6.3. Summary of results of classification evaluation metrics 46

Fig. 2.1. Schemstic diagram of human speech production 14

Fig. 3.1. LTI system of speech 16

Fig. 3.2. Spectrum of voice wave 18

Fig. 3.3. Power spectrum of voice wave 19

Fig. 3.4. Mel spectrogram of voice wave 20

Fig. 3.5. Mel Frequency Cepstral Coeffceint of voice wave 20

Fig. 3.6. Real cepstrum of voice wave (top) and complex cepstrum of voice wave (bottom) 21

Fig. 3.7. Morlet function (top) and Mexican Hat function (bottom) 23

Fig. 3.8. Morlet function on Continous Wavelet Transform 24

Fig. 3.9. Mexican Hat function on Continous Wavelet Transform 25

Fig. 3.10. Frequency pass filter of Discrete Wavelet Transform 25

Fig. 3.11. Discrete Wavelet Transform, level ＝ 1, 2, 3 26

Fig. 5.1. Cepstral Peak Prominence for quefrency, quefrency ＝ 0.05 s, 0.1 s, 0.2 s, 0.4 s (from the top left) 34

Fig. 5.2. Cepstral Peak Prominence for sampling rate, sampling rate ＝ 8000 Hz, 16000 Hz, 22050 Hz, 44100 Hz (from the top left) 35

Fig. 5.3. Schematic diagram for analysis 36

Fig. 5.4. Cepstral Peak Prominence of linear regession 37

Fig. 5.5. CPP of linear regession on depressive patient (left) CPP of linear regession on normal (right) 37

Fig. 5.6. Real cepstrum smoothed with lowess 38

Fig. 5.7. CPP of Lowess on depressive patient (left) and CPP on Lowess of normal (right) 38

Fig. 5.8. Real cepstrum smoothed with wavelet 39

Fig. 5.9. CPP on wavelet of depressive patient (left) and CPP on wavelet of normal (right) 39

Fig. 6.1. Boxplot of CPP with group, Group ＝ Nomal, Depressed 44