A New Type of Wavelet Threshold Denoising Algorithm
According to spectrum subtraction, this paper puts forward a new type of threshold value determination algorithm. Firstly, through the artificial extraction or by the zero point detection method, extract background noise from no sound segment. Secondly, d
- PDF / 1,868,255 Bytes
- 7 Pages / 439.37 x 666.142 pts Page_size
- 24 Downloads / 222 Views
Abstract According to spectrum subtraction, this paper puts forward a new type of threshold value determination algorithm. Firstly, through the artificial extraction or by the zero point detection method, extract background noise from no sound segment. Secondly, do wavelet decomposition with background noise, and determine the threshold value on the basis of each layer’s wavelet decomposition coefficient. Then, we can make a speech enhancement for the speech signal with noise. The simulation results show that this algorithm can effectively remove the noise component and keep the details of the useful signal characteristics very well. More over, the amount of calculation is far less than the traditional threshold algorithm’s. Keywords Wavelet analysis
Threshold de-noising Speech enhancement
1 Introduction The basic idea of using wavelet threshold de-noising algorithm to make a speech enhancement is: when we are making wavelet decomposition to the speech signal with noise, speech signal’s wavelet coefficient is concentrated in the dense area, and the absolute value of wavelet coefficients is relatively large; On the contrary, noise signal wavelet coefficient distributes relatively dispersive, so the absolute value is small [1]. Based on the features of different layers’ wavelet decomposition coefficient, we can do the multi-scale wavelet transform with noise signal, then cut or completely remove wavelet coefficients of noise component for each layer’s coefficient, meanwhile retain speech signal’s wavelet coefficient; After that, we can carry through wavelet reconstruction, so as to achieve the effect of speech enhancement. L. K. Xing (&) S. Qi W. J. Wang College of Electrical and Information Engineering, Anhui University of Science and Technology, Huainan 232001 Anhui, China e-mail: [email protected]
Z. Yin et al. (eds.), Proceedings of The Eighth International Conference on Bio-Inspired Computing: Theories and Applications (BIC-TA), 2013, Advances in Intelligent Systems and Computing 212, DOI: 10.1007/978-3-642-37502-6_31, Ó Springer-Verlag Berlin Heidelberg 2013
249
250
L. K. Xing et al.
There is a critical value between useful signal and noise signal in the high frequency wavelet coefficient. On the basis of experience or some testimony, we can find out the critical value and set it to threshold value. Then, the next work is to do threshold function processing [2]. The ideal threshold is equal to the critical value, so that the noise component can be entirely eliminated. At the same time, the useful signal can also be well reserved. Further more, we can get pure useful signal just through reconstructing wavelet coefficient. However, the noise component’s critical value is quite close to the useful signals in actual applications. At this time, it directly affects the noise reduction effect. If the threshold value is too small, it’s unable to effectively remove noise component, but if too large, it will cause the signal distortion [3]. This paper refers to spectrum subtraction. Using the improved thr
Data Loading...
