검색 상세

Beamforming based sound source separation in noisy environment

  • 이택진 (Lee, TaekJin, 대학원 영상정보처리협동과정 영상정보처리전공)

초록/요약

This thesis proposes a sequential least squares method of estimating the relative transfer function (RTF) which is used in sound source separation. In the conven-tional method, the batch least squares method is employed to estimate the RTF which can not trace the changes in the environment. The proposed estimation method for the RTF focuses on tracing the changing data to achieve an accurate estimation. In addition, it can reduce the memory requirement by using a recursive form. For more accurate estimation, the initialization of the RTF is decided based on the decision rule described in Chapter 3. By conducting experiments with real and artificial data, the results show that the proposed method exhibits improved performance in sound source separation compared to the conventional methods. Signal enhanced factor (SEF) improves by an average of d dB over the baseline. With respect to mean squared error (MSE), the proposed method shows an enhancement of 0.35 over the conventional method.

more

목차

Chapter 1 Introduction = 1
1.1 Theoretical background and research goals = 1
1.2 Contributions = 4
1.3 Organization of thesis = 5
Chapter 2 Related works = 6
2.1 Relative transfer function estimation = 6
2.2 Steered response power-phase transform (SRP-PHAT) = 9
Chapter 3 Beamforming based sound source separation = 13
3.1 Dual source transfer function generalized sidelobe canceller = 13
3.2 Sound source separation using the DTF-GSC = 18
Chapter 4 Sequential estimation of RTF = 22
4.1 Problem statements = 22
4.2 Sequential RTF Estimation Algorithm = 24
4.3 RTF initialization = 28
4.4 Summary of the proposed sound source separation = 33
Chapter 5 Experiments and results = 34
5.1 Experiment settings = 34
5.2 Performance evaluation = 36
5.3 Results and discussions = 37
Chapter 6 Concluions = 46
References = 48

more
반출 Meta View 목록