International Journal of Engineering Trends and Technology (IJETT) – Volume 4 Issue 9- Sep 2013 A Review on Speech Denoising Using Wavelet Techniques KamyaDubey#, Prof. Vikas Gupta* Department of Electronics and Communication, T.I.T, R.G.P.V., Bhopal Abstract- The use of wavelet transform for denoising speech signals contaminated with common noises.The basic principles of wavelet transform hold good promise an alternative to the Fourier transform. Based on the fact that noise is the main factor that limit the capacity of data transmission in telecommunications and that also affect the accuracy of the results in the signal measurement systems, whereas, modeling and removing noise and is at the core of theoretical and practical considerations in communications and signal processing. In this paper Wavelet for noise removal has been discussed. Keywords: Wavelet,Daubechies,Coiflet,De-noising. I. INTRODUCTION Speech is a very basic way for humans to conveyinformation to one another. With a bandwidth of only 4 kHz, speech can convey information with the emotion of a human voice. People want to be able to hear someone's voice from anywhere in the world, as if the person was in the same room. As a result a greater emphasis is being placed on the design of new and efficient speech coders for voice communication and transmission. Human Speech signal are in acoustic signal form so for the purpose of communication it’s a necessary to convert into the electrical signal with the help of instruments called ‘Transducers’. Electrical representation of speech signal has same properties, such as: 1. It is a one-dimensional signal, with time as its independent variable. 2. It is random in nature. 3. It is non-stationary, i.e. the frequency spectrum is not constant in time. 4. Although human beings have an audible frequency range of 20Hz - 20 kHz, The wavelet transform is similar to the Fourier transform. For the FFT, the basis functions are sines and cosines. For the wavelet transform, the basis functions are more complicated called wavelets, mother wavelets or analyzing wavelets and scaling function. In wavelet analysis, the signal is broken into shifted and scaled versions of the original (or mother) wavelet. The fact that wavelet transform is a multi resolution analysis makes it very suitable for analysis of nonstationary signals such as audio signals. II. WAVELETS A wavelet is a finite energy signal defined over specific interval of time [1]. The main interest in wavelets is their ability to represent a given signal at different. Wavelets are used to analyze signals in much the same way as complex exponentials (sine and cosine) used in Fourier analysis of signals. Unlike Fourier, wavelets can be used to analyze nonstationary, time-varying, or transient signals [9] [10]. This is an important aspect, since speech signals are considered to be non-stationary. A given signal is represented by using translated and scaled versions of a mother wavelet . Wavelets are characterized by scale and position, and are useful in analyzing variations in signals in terms of scale and position. Because of the fact that the wavelet size can vary, it has advantage over the classical signal processing transformations to simultaneously process time and frequency data. The general relationship between wavelet scales and frequency is to roughly match the scale. At low scale, compressed wavelets are used. They correspond to fast-changing details, that is, to a high frequency. At high scalethe wavelets are stretched. They correspond to slow changing features, that is, to a low frequency. A. Continuous Wavelet Transform The wavelet transform is a two parameter expansion of a signal in terms of a particular wavelet basis function [1]. Given Ψ(t) called the mother wavelet; all other baby wavelets are obtained by simple scaling and translation of Ψ(t). Ψa,t(t) = (1/√a) Ψ[(t-b)/a]. Where a and b are the scaling and the translation parameter respectively. A nice approach to the CWT representation is first to inspect the Fourier transform represented mathematically by: F(ω) = ∫s(t)e-jωtdt…………..(1) Replacing the complex exponential in the Fourier transform with Ψa,t(t) yields: ISSN: 2231-5381 http://www.ijettjournal.org Page 3882 International Journal of Engineering Trends and Technology (IJETT) – Volume 4 Issue 9- Sep 2013 C(S,U) = ∫√a Ψ[(t-b)/a]dt……….(2) In other words with wavelet transform, reference to frequency is replaced by reference to scale [8], [9][10]. B. Discrete Wavelet Transform In our application the discrete wavelet transform is applied.By choosing scale and position based on power of 2, CWT isreduced to DWT without any loss in energy. The scalingparameter is discrete and dyadic, a = 2-j. The translation isdiscretized with respect to each scale by using τ = k2-jT [5].Ψj,k(t) = (2j/2) Ψ[(2jt-kT)].The integer k represents the translation of the waveletfunction; it indicates time in wavelet transform. Integer j,however, is an indication of the wavelet frequency orspectrum shift and generally referred to as scale. The DWTtransforms a discrete input signal vector into two sets ofcoefficients the approximation CA containing low frequencyinformation and the detail coefficients CD containing highfrequency information. Fig. 1 shows a level 2 DWTdecomposition of an input signal s (t) [8] [9] [10]. Figure.3: IDWT Reconstruction (Synthesis) III. NOISE Noise is defined as an unwanted signal that interferes with the communication or measurement of another signal. A noise itself is an information-bearing signal that conveys information regarding the sources of the noise and the environment in which it propagates. Noise can be categorized of different types .In which white noise is defined as an uncorrelated random noise process with equal power at all frequencies. Random noise has the same power at all frequencies in the range of ∞ it would necessarily need to have infinite power, and it is therefore an only a theoretical concept. However, a band-limited noise process with a flat spectrum covering the frequency range of a band-limited communication system is practically considered a white noise process. Figure .1: Plots of Different Daubechies Orthogonal Wavelets Figure .2: DWT Decomposition (Analysis) ISSN: 2231-5381 http://www.ijettjournal.org Page 3883 International Journal of Engineering Trends and Technology (IJETT) – Volume 4 Issue 9- Sep 2013 Figure.4: Sine wave with added white noise. IV.RESULT used as denoising technique. We have compared different wavelet families: Daubechies and Coiflets, and we used cross-correlation to determine the best fit between an original signal and the processed one. By using Coiflet 5, Daubechies 9 and 10 we obtained the best result because they have a higher correlation at zero. An audio signal is selected and with wavelets coif5, db9 and db10.denoising and cross correlation is performed. Figure shows the original audio signal, audio signal with noise and the recovered signal using the wavelet. Figure.5:Cross correlation of original signal and signal de-noise REFERENCES [1] Y. T. Chan “Wavelet Basics”, Kluwer Academic Publishers, 1995. [2] C. Taswell, “Speech Compression with Cosine and Wavelet Packet NearBest Bases”, ICASSP-96. Conference Proceedings, IEEE International Conference on Acoustics, Speech and Signal Processing, Vol.1, pp: 566 – 568, 1996. [3] E. Fgee, W. J. Phillips, W.Robertson “Comparing Audio Compression Using Wavelets With Other Audio Compression Schemes”, IEEE, CCECE Conference Edmonton, Alberta, Canada. May 9-12, pp: 698701, 1999. [4] N. M. Hosny, S. H. El-Ramly, M. H. El-Said, “Novel Techniques for Speech Compression Using Wavelet Transform”, ICM '99. 11th International Conference on Microelectronics, pp: 225 – 229, 1999. [5] http://www.mathworks.com/academia MATLAB Wavelet Toolbox 2. [6] O. Rioul and M. Vetterli, “Wavelets and Signal Processing”, IEEE Signal Process. Mag. Vol 8, pp. 14-38, Oct. 1991. [7] J. I. Agbinya, “Discrete Wavelet Transform Techniques in Speech Processing,” IEEE Digital Signal Processing Applications Proceedings,IEEE, New York, pp: 514 – 519, 1996. [8] J. F. Koegel Buford, “Multimedia Systems”, ACM Press, 1994. [9] G. Strang and T. Nguyen, “Wavelets and Filter Banks”, WellesleyCambridge Press, 1996. [10] J. S. Walker, “Wavelets and their Scientific Applications”, Chapman and Hall/CRC, 1999.International Journal of Biological and Life Sciences 1:4 2005 234. Figure.6: Original audio signal, audio signal with noise and the recovered signal V.CONCLUSION Practical approach in how to put into practice wavelets in noisy audio data to improve clarity and signal retrieval has been discussed. analysis.Fourier transform based spectral analysis is the dominant analytical tool for denoising of audio signal. However, Fourier transform cannot provide any information of the spectrum changes with respect to time. Fourier transform assumes the signal is stationary, but audio signal is always non-stationary. To overcome this wavelet is ISSN: 2231-5381 http://www.ijettjournal.org Page 3884