1. Rajeev Aggarwal, Jai Karan Singh, Vijay Kumar Gupta, Sanjay Rathore, Mukesh Tiwari, Dr.Anubhuti Khare International Journal of Computer Applications (0975 – 8887) Volume 20– No.5, April 2011 Presenter Chia-Cheng Chen 1

2.  Introduction  Multiresolution Analysis Using Filter Bank  Modifier Universal Threshold  Soft and Hard Thresholding  Results and Discussion 2

3. 3

4.  Discrete-wavelet transform (DWT) based algorithm are used for speech signal denoising. Here both hard and soft thresholding are used for denoising.  Analysis is done on noisy speech signal corrupted by babble noise at 0dB, 5dB, 10dB and 15dB SNR levels. 4

5.  The DWT uses Multi resolution filter banks and special wavelet filters for the analysis and reconstruction of signals.  It analyse the signal at different frequency bands with different resolutions, decompose the signal into a coarse approximation and detail information. 5

6.  Two-level wavelet decomposition tree 6

7.  Two-level wavelet reconstruction tree 7

8.  Wavelet thresholding is applied to the approximation coefficient (Vn-1) and detail coefficients (Wn,Wn-1). 8

9.  In this paper, we removed the babble noise from noisy signal which contain the noise contents of babble noise.  We want to find threshold value that will use to remove noise from noisy signal, but also recover the original signal efficiently. 9

10.  Developed by Donoho and Jonstone and it is called as universal threshold ◦ Where N denotes number of samples of noise and is standard deviation of noise. 10

11.  Again Universal threshold was modified with factor “k ‟ in order to obtain higher quality output signal: 11

12.  The soft and hard thresholding methods are used to estimate wavelet coefficients in wavelet threshold denoising. ◦ Hard thresholding ◦ Soft thresholding 12

13. 13 Thr=0.4 Z = (-1, 1, 100)

14.  We implemented babble noise removal algorithm in Matlab 7.10.0 (R2010a).  Speech signal is corrupted by babble noise at 0dB, 5dB, 10dB and 15dB SNR levels. 14

15. 15

16.  We choose 5-level DWT and db5 wavelet. Improved threshold value is obtained by replacing threshold “thr ‟ (2) with 16

17. 17

18. 18

19. 19

20. 20

21.  Speech denoising is performed in wavelet domain by thresholding wavelet coefficients.  We found that by using modified universal threshold, we can get the better results of de-noising, especially for low level noise. 21