International Journal of Advances in Engineering & Technology, Jan 2012.

©IJAET ISSN: 2231-1963

AUDIO DENOISING USING WAVELET TRANSFORM

B. JaiShankar1 and K. Duraiswamy2
1
Department of Electronics and Communication Engineering, KPR Institute of Engineering
and Technology, Coimbatore, India.
2
Dean, K.S.Rangasamy College of Technology, Tiruchengode, India.

ABSTRACT
Noises present in communication channels are disturbing and the recovery of the original signals from the path
without any noise is very difficult task. This is achieved by denoising techniques that remove noises from a
digital signal. Many denoising technique have been proposed for the removal of noises from the digital audio
signals. But the effectiveness of those techniques is less. In this paper, an audio denoising technique based on
wavelet transformation is proposed. Denoising is performed in the transformation domain and the improvement
in denoising is achieved by a process of grouping closer blocks. The technique exposes each and every finest
details contributed by the set of blocks and also it protects the vital features of every individual block. The
blocks are filtered and replaced in their original positions. The grouped blocks overlap each other and thus for
every element a much different estimation is obtained. A technique based on this denoising strategy and its
efficient implementation is presented in full detail. The implementation results reveal that the proposed
technique achieves a state-of-the-art denoising performance in terms of both signal-to-noise ratio and audible
quality.
KEYWORDS: Wavelet Transformation, Block Matching, Grouping, Denoising, Reconstruction

I. INTRODUCTION
Signal processing applications always disturbed by noise and it seems to be a major problem. A
nonessential signal gets superimposed over an undisturbed signal. If the regularity of noise lessens,
then the method for denoising [12] gets more sophisticated. When a signal pass through equipments
and connecting wires it naturally gets added with a noise. Therefore it results in signal contamination.
Once a signal is polluted, it is essentially difficult to remove it without altering the original signal.
Hence, the basic task in signal processing [15] is denoising of signals. Humming noise from audio
equipments and background environment noise, both serves as the major root cause for pollution in
audio signals. The objective of audio denoising is attenuating the noise, while recovering the
underlying signals. It is accessible in many applications such as music and speech restoration etc
Previous methods, such as Gaussian filters and anisotropic diffusion, denoise the value of a signal
based on the observed values neighbouring points. Various authors proposed many global and
multiscale denoising approaches [15] in order to overcome this locality property. From the beginning
of wavelet transforms in signal processing, it is noticed that the wavelet thresholding focuses a
attention in removing noise from signals and images. To remove the wavelet coefficients smaller than
a given amplitude and to transform the data back into the original domain, the method has to
decompose the noisy data into an orthogonal wavelet basis. A nonlinear thresholding estimator can
compute in an orthogonal basis such as Fourier or cosine.
In denoising of the audio signals, the denoised signal obtained after performing wavelet
transformation is not totally free from noise, some residue of noise left or some other kinds of noise
gets introduced by the transformation that is present in the output signal. Several techniques have

419 Vol. 2, Issue 1, pp. 419-425

International Journal of Advances in Engineering & Technology, Jan 2012.
©IJAET ISSN: 2231-1963
been proposed so far for the removal of noise from an audio signal, yet, the efficiency remains an
issue as well as they have some drawbacks in general. In this article, we propose an audio signal
denoising technique based on an improved block matching technique in transformation domain. The
transformation can achieve a clear representation of the input signal so that the noise can be removed
well by reconstruction of the signal. A biorthogonal 1.5 wavelet transform is used for the
transformation process. A multi dimensional signal vector is generated from the transformed signal
vector and the original vector signal is reconstructed by applying inverse transform. The signal to
noise ratio (SNR) is comparatively higher than SNR level of the noisy input signal thus increasing the
quality of the signal.

II. LITERATURE REVIEW

Michael T. Johnson et al. [6] have demonstrated the application of the Bionic Wavelet Transform
(BWT), an adaptive wavelet transform derived from a non-linear auditory model of the cochlea, to the
task of speech signal enhancement. Results measured objectively by Signal-to-Noise ratio and
Segmental SNR and subjectively by Mean Opinion Score were given for additive white Gaussian
noise as well as four different types of realistic noise environments. Enhancement is accomplished
through the use of thresholding on the adapted BWT coefficients and the results were compared to a
variety of speech enhancement techniques, including Ephraim Malah filtering, iterative Wiener
filtering, and spectral subtraction, as well as to wavelet denoising based on perceptually scaled
wavelet packet transform decomposition. Overall results indicated that SNR and SSNR improvements
for the proposed approach were comparable to those of the Ephraim Malah filter, with BWT
enhancement giving the best results of all methods for the noisiest conditions. Subjective
measurements using MOS surveys across a variety of 0 db SNR noise conditions indicate that the
quality enhancement competitive with but still lower than results for Ephraim Malah filtering and
iterative Wiener filtering, but higher than the perceptually scaled wavelet method.
Mohammed Bahoura and Jean Rouat [7] have proposed a new speech enhancement method based on
time and scale adaptation of wavelet thresholds. The time dependency was introduced by
approximating the Teager energy of the wavelet coefficients, while the scale dependency was
introduced by extending the principle of level dependent threshold to wavelet packet thresholding.
The technique does not require an explicit estimation of the noise level or of the a priori knowledge of
the SNR, as was usually needed in most of the popular enhancement methods. Performance of the
proposed method was evaluated on the recorded speech in real conditions (plane, sawmill, tank,
subway, babble, car, exhibition hall, restaurant, street, airport, and train station) and artificially added
noise. MEL-scale decomposition based on wavelet packets was also compared to the common
wavelet packet scale. Comparison in terms of signal-to-noise ratio (SNR) was reported for time
adaptation and time–scale adaptation of the wavelet coefficients thresholds. Visual inspection of
pectrograms and listening experiments were also used to support the results. Hidden Markov Models
speech recognition experiments were conducted on the AURORA-2 database and showed that the
proposed method improved the speech recognition rates for low SNRs.
Ching-Ta and Hsiao-Chuan Wang [8] have proposed a method based on critical-band decomposition
which converts a noisy signal into wavelet coefficients (WCs), and enhanced the WCs by subtracting
a threshold from noisy WCs in each subband. The threshold of each subband is adapted according to
the segmental SNR (SegSNR) and the noise masking threshold. Thus residual noise could be
efficiently suppressed for a speech-dominated frame. In a noise-dominated frame, the background
noise could be almost removed by adjusting the wavelet coefficient threshold (WCT) according to the
SegSNR. Speech distortion could be reduced by decreasing the WCT in speech-dominated subbands.
The proposed method could effectively enhance noisy speech which was infected by colored-noise.
Its performance was better than other wavelet-based speech enhancement methods in their
experiments.
Marián Képesia and Luis Weruaga [11] have proposed new method for time–frequency analysis of
speech signals. The analysis basis of the proposed Short-Time Fan-Chirp Transform (FChT) was
defined univocally by the analysis window length and by the frequency variation rate, that parameter
being predicted from the last computed spectral segments. Comparative results between the proposed
Short-Time FChT and popular time–frequency techniques reveal an improvement in spectral and

420 Vol. 2, Issue 1, pp. 419-425

International Journal of Advances in Engineering & Technology, Jan 2012.
©IJAET ISSN: 2231-1963
time–frequency representation. Since the signal can be synthesized from its FChT, the proposed
method was suitable for filtering purposes.
Nanshan Li et al. [13] have proposed an audio denoising algorithm on the basis of adaptive wavelet
soft-threshold which is based on the gain factor of linear filter system in the wavelet domain and the
wavelet coefficients teager energy operator in order to progress the effect of the content-based songs
retrieval system. Their algorithm integrated the gain factor of linear filter system and nonlinear energy
operator with conventional wavelet soft-threshold function. Experiments demonstrated that their
algorithm had important outcome in inhibiting Gaussian white noise and pink noise in audio samples
and enhancing the accuracy of the songs retrieval system.

III. PROPOSED METHOD

In this section, the proposed audio denoising technique for the removal of unwanted noises from any
audio signal is explained. It is considered that the audio signal is polluted by Additive White Gaussian
Noise (AWGN) and the polluted signal is subjected to the removal of noise using the proposed
denoising technique. The processes performed in the proposed denoising technique are explained in
detail as follows (i) Transformation of the noisy signal to wavelet domain, (ii)Generation of a set of
closer blocks (iii) Generating a multidimensional array (iv) reconstructing the denoised audio signal.
In the proposed work, initially, the noisy audio signal is subjected to Wavelet Transformation.
Wavelet transformation produces a few significant coefficients for the signals with discontinuities.
Audio signals are smooth with a few discontinuities. Hence, wavelet transformation has better
capability for representing these signals when compared to other transformations. Once the signal is
transformed to wavelet domain, set of closer blocks are synthesized from it.

Figure1. Generation of a Set of Closer blocks

3.1. Synthesizing Closer Blocks from the Noisy Signal
The Bior 1.5 wavelet transformation is applied to the input noisy signal and a transformed signal is
obtained as output through this process. Designing biorthogonal wavelets allows more level of
choices than orthogonal wavelets. One additional level of choice is the possibility to generate
symmetric wavelet functions. For the process of transformation, the vector noisy audio signal is
reshaped into a matrix of size same as that of the transformation coefficient matrix. The noisy audio
signal is transformed to biorthogonal wavelet transformation domain and it is represented as a vector
signal that eases the operation.

Figure2.Block Representation of the noisy input signal

The process is then followed by the calculation of the L2 norm distance for each block generated in
the vector signal. The transformed initial block is kept as the reference block. The distance between

421 Vol. 2, Issue 1, pp. 419-425

International Journal of Advances in Engineering & Technology, Jan 2012.
©IJAET ISSN: 2231-1963
the reference block and all the other grouped blocks are calculated. Similarly, the process is repeated
in the same fashion for all the blocks with the consideration of every block as a reference block.

Figure3.Generation of multidimensional vector

The obtained temporary vector signal is then transformed using Haar’s transformation. The process of
reshaping is carried out to perform the transformation process. Haar’s Transformation deals with a 2-
point mean and difference operation. Every element in the transformed vector block is compared with
a threshold value such that if the element’s value is less than the particular element is replaced with
‘0’ and replaced back in the temporary block. If the element’s value of the transformed block is
greater than the threshold then the value of the transformed element is not changed. The process is
repeated for all the elements in the reference block and their respective grouped set of closer blocks.
The reconstruction of the signal is shown in figure 4.

Figure4.Reconstruction of the audio signal

IV. IMPLEMENTATION RESULTS

For testing the performance of the proposed technique a dog barking signal is taken as input. The
input signal is then contaminated with AWGN noise for the testing purpose. The input noisy signal
with respect to its length n =12000 is represented in Figure 5(a) .The AWGN was generated and
added to the input signal. The SNR of the noisy audio signal was 7.77 dB at a noise level σ of 0.047.
A linear combination of the generated noise and the original signal is used as the primary input for the
block matching technique. Figure 5(b) shows the input signal corrupted by white noise. The denoised
speech signal achieved through the block matching technique is shown in Figure 5(c). The SNR value
of the denoised signal with block matching technique was 12.85 dB.

Figure5 (a).Original audio signal

422 Vol. 2, Issue 1, pp. 419-425

International Journal of Advances in Engineering & Technology, Jan 2012.
©IJAET ISSN: 2231-1963

Figure5 (b).Original audio signal with noise

Figure5(c).Denoised audio signal

V. CONCLUSIONS
This paper presented an audio denoising technique based on block matching technique. The technique
was based on the denoising strategy and its efficient implementation was presented in full detail. The
implementation results have revealed that the process of block matching has achieved a state-of-the-
art denoising performance in terms of both peak signal-to-noise ratio and subjective improvement in
the audible quality of the audio signal. Grouping of the similar blocks improved the efficient
operation of the technique. The blocks were filtered and replaced in their original positions from
where they were detached. The grouped blocks were overlapping each other and thus for every
element a much different estimation was obtained that were combined to remove noise from the input
signal. The reduction in the noise level interprets that the technique has protected the vital unique
features of each individual block even when the finest details were contributed by grouped blocks. In
addition the technique can be modified for various other audio signals as well as for other problems
that can be benefit from highly linear signal representations.
REFERENCES
[1] Qiang Fu and Eric A. Wan, 2003. "Perceptual Wavelet Adaptive Denoising of Speech", In:
Proc. European Conf. on Speech Commun. and Technology, pp: 577-580.
[2] Alyson K. Fletcher, Vivek K Goyal and Kannan Ramchandran, 2003. "Iterative Projective
Wavelet Methods for Denoising", Proc. Wavelets X, part of the 2003 SPIE Int. Symp. on
Optical Science & Technology, Vol. 5207, pp: 9-15, San Diego, CA August.
[3] Claudia Schremmer, Thomas Haenselmann and Florian Bomers, 2001. "A Wavelet Based
Audio Denoiser", In Proc. IEEE International Conference on Multimedia and Expo
(ICME'2001), pp: 145-148.
[4] Sylvain Durand and Jacques Froment, 2001. "Artifact Free Signal Denoising With Wavelets",
Proceedings IEEE International Conference on Acoustics, Speech, and Signal Processing

423 Vol. 2, Issue 1, pp. 419-425

International Journal of Advances in Engineering & Technology, Jan 2012.
©IJAET ISSN: 2231-1963
(ICASSP '01), Vol.6, pp: 3685-3688.
[5] Imola K. Fodor and Chandrika Kamath, 2003. "Denoising Through Wavelet Shrinkage: An
Empirical Study", Journal of Electronic Imaging, Vol. 12, pp: 151-160.
[6] Michael T. Johnson, Xiaolong Yuan and Yao Ren, 2007. "Speech Signal Enhancement through
Adaptive Wavelet Thresholding", Speech Communications, Vol. 49, No. 2, pp: 123-133.
[7] Mohammed Bahoura and Jean Rouat, 2006. "Wavelet speech enhancement based on time–scale
adaptation", Speech Communication, Vol. 48, No. 12, pp: 1620-1637.
[8] Ching-Ta and Hsiao-Chuan Wang, 2003. "Enhancement of single channel speech based on
masking property and wavelet transform", Speech Communication, Vol. 41, No 2-3, pp: 409-
427.
[9] Ching-Ta and Hsiao-Chuan Wang, 2007. "Speech enhancement using hybrid gain factor in
critical-band-wavelet-packet transform", Digital Signal Processing, Vol. 17, No. 1, pp: 172-
188.
[10] Erik Visser, Manabu Otsuka and Te-Won Lee, 2003. "A spatio-temporal speech enhancement
scheme for robust speech recognition in noisy environments", Speech Communication, Vol. 41,
No. 2-3, pp: 393-407.
[11] Marián Képesia and Luis Weruaga, 2006. "Adaptive chirp-based time–frequency analysis of
speech signals", Speech Communication, Vol. 48, No. 5, pp: 474-492.
[12] Claudia Schremmer, Thomas Haenselmann and Florian Bomers, 2001. "A Wavelet Based
Audio Denoiser," Proceedings of IEEE International Conference on Multimedia and Expo, pp:
145-148, Tokyo, Japan.
[13] Nanshan Li and Mingquan Zhou, 2008. "Audio Denoising Algorithm Based on Adaptive
Wavelet Soft-Threshold of Gain Factor and Teager Energy Operator," Proceedings of IEEE
International Conference on Computer Science and Software Engineering, Vol. 1, pp: 787-790.
[14] Enqing Dong and Xiaoxiang Pu, 2008. "Speech denoising based on perceptual weighting
filter," Proceedings of 9th IEE International Conference on Signal Processing, pp: 705-708,
October 26-29, Beijing.
[15] Amit Singer, Yoel Shkolnisky, and Boaz Nadler, 2009. "Diffusion Interpretation of Nonlocal
Neighborhood Filters for Signal Denoising", SIAM J. Imaging Sciences, Vol. 2, No. 1, pp: 118–
139.
[16] Guoshen Yu, Stephane Mallat and Emmanuel Bacry, 2007. "Audio Signal Denoising with
Complex Wavelets and Adaptive Block Attenuation," Proceedings of IEEE International
Conference on Acoustics, Speech and Signal Processing, Vol. 3, pp: 869-872, April 15-20,
Honolulu, HI.
[17] Guoshen Yu, Stéphane Mallat and Emmanuel Bacry, 2008. "Audio Denoising by Time-
Frequency Block Thresholding", IEEE Transactions On Signal Processing, Vol. 56, No. 5, pp:
1830-1839.
[18] D. L. Donoho and I. M. Johnstone, 1994. “Ideal spatial adaptation via wavelet shrinkage,”
Biometrika, Vol. 81, pp: 425–455.
[19] D. L. Donoho, 1995. “De-noising by soft-thresholding,” IEEE Trans. Inform. Th., Vol. 41, pp:
613–627.
[20] R. R. Coifman and D. L. Donoho, 1995. “Translation-invariant de-noising, in Wavelets and
Statistics”, A. Antoniadis and G. Oppenheim, eds., Springer Lecture Notes in Statistics 103, pp:
125–150, Springer-Verlag, New York.
[21] Jonathan Berger and Charles Nichols, “Brahms at the piano,” Leonardo Mus. Journal, vol. 4, pp. 23–
30, 1994.
[22] David L. Donoho, “Nonlinear Wavelet Methods for Recovering Signals, Images, and Densities
from Indirect and Noisy Data,” http://wwwstat.stanford.edu/˜donoho/Reports/, 1993.
[23] Ingrid Daubechies, Ten Lectures on Wavelets, vol. 61, SIAM. Society for Industrial and Applied
Mathematics, Philadelphia, Pennsylvania, 1992, ISBN 0-89871-274-2.
[24] St´ephaneMallat, AWavelet Tour of Signal Processing,Academic Press, San Diego, CA, USA, 1998,
ISBN 0-12-466605-1.
[25] M. Lang, H. Guo, J.E. Odegard, and C.S. Burrus,“Nonlinear processing of a shift invariant DWT for
noise reduction,” SPIE,Mathematical Imaging: WaveletApplications for Dual Use, April 1995.
[26] Curtis Roads, The ComputerMusic Tutorial, TheMIT Press, 1996.
[27] Claudia Schremmer, Thomas Haenselmann, and Florian B¨omers, “Wavelets in Real–Time Digital
Audio Processing: A Software For Understanding Wavelets in Applied Computer Science,” in
Workshop on Signal Processing Applications (WoSPA), http://www.sprc.qut.edu.au/wospa2000/,
December 2000, Signal Processing Research Center (SPRC) and IEEE.
[28] Claudia Schremmer, “The Wavelet Tool,” http://-www.informatik.uni-mannheim.de/˜cschremm/-

424 Vol. 2, Issue 1, pp. 419-425

International Journal of Advances in Engineering & Technology, Jan 2012.
©IJAET ISSN: 2231-1963
wavelets/WaveletAudioTool/, 2000.
[29] Manojit Roy, V. Ravi Kumar, B.D. Kulkarni, John Sanderson, Martin Rhodes, and Michel van der
Stappen, “Simple denoising algorithmusing wavelet transform,”AIChE Journal, vol. 45, no. 11, pp.
2461–2466, 1999.
[30] Florian B¨omers, “Wavelets in Real–Time Digital Audio Processing: Analysis and Sample
Implementations,” M.S. thesis, Universit¨at Mannheim, http://www.bome.com/personal/thesis.pdf,
2000.

AUTHORS PROFILE

B. Jai Shankar received B.E. degree in Electronics and Communication Engineering from
Government College of Engineering, Salem and M.E. degree in Applied Electronics from
Kongu engineering College, Erode. He worked in K.S.R College of Engineering, Tiruchengode
for three years. Currently he is working as lecturer in Kumaraguru College of Technology,
Coimbatore since 2008. His research interest includes Digital Signal Processing, Image
Processing and Wavelets.

K. Duraiswamy received his B.E. degree in Electrical and Electronics Engineering from
P.S.G. College of Technology, Coimbatore in 1965 and M.Sc.(Engg) from P.S.G. College of
Technology, Coimbatore in 1968 and Ph.D. from Anna University in 1986. From 1965 to 1966
he was in Electricity Board. From 1968 to 1970 he was working in ACCET, Karaikudi. From
1970 to 1983, he was working in Government College of Engineering Salem. From 1983 to
1995, he was with Government College of Technology, Coimbatore as Professor. From 1995 to
2005 he was working as Principal at K.S. Rangasamy College of Technology, Tiruchengode
and presently he is serving as DEAN of K.S. Rangasamy College of Technology, Tiruchengode, India. Dr. K.
Duraiswamy is interested in Digital Image Processing, Computer Architecture and Compiler Design. He
received 7 years Long Service Gold Medal for NCC. He is a life member in ISTE, Senior member in IEEE and a
member of CSI.

425 Vol. 2, Issue 1, pp. 419-425