978-1-4673-7231-2/15/$31.00 ©2015 IEEE
Blind Audio Source Separation Using Wiener
Filtering Approach
Pardeep Sharma
1
, Rajesh Mehra
2
, Naveen Dubey
3
1,2,3
Department of Electronics and Communication Engineering, National Institute for Technical Teachers Training
and Research
Sector – 26, Chandigarh, 16001, India
1
Pardeep_4992@yahoo.in
2
rajeshmehra@yahoo.com
3
naveendubey@rkgitw.edu.in
Abstract- Audio Source separation techniques are used for better
reception of sound and speech signals. Wiener filtering tool is
best and one of the principally used method in separation of the
audio signal from mixture of source signals. As the STFT utilizes
short-duration stationarity in time frequency domain, we use
Wiener filtering mask which does not depends on the consistency
of the output for the voice gram and is different from the STFT
in time-frequency domain. Short-time Fourier transform (STFT)
in time-frequency domain is used if processing is done on audio
signals. In this paper a technique for blind audio source
separation using Wiener filtering algorithm is presented and
result reflects that it serves good quality of separation in
comparison of classical ICA algorithms like fast ICA, JADE. So
it gives the SIR value 6.68% and 39.22% higher than that of fast
ICA and JADE respectively.
Keywords- Blind audio source separation, Wiener filtering,
Independent component analysis (ICA), Short time Fourier
transforms.
I. INTRODUCTION
ASS basically known as blind audio source separation or
blind audio signal separation used to recover the
individual/independent source signals from the mixture of
signals. Suppose R(t)= [r
1(t)
, r
2(t),
----, r
n(t)
] be the source signals
and M(t) = [m
1(t)
, m
2(t), ------
, m
n(t)
] be the mixed signal received
at different microphones located at different positions. So that
the actual mixing process received at the mixer end will be
expressed by M(t) = A(t)*R(t). Let take an example of the
group discussion problem, where voices from individual
persons are mixed and it is difficult to listen to the one
actually speaking and need to be separate out the desired one.
So for this a processing system can be developed to obtain the
desired voice from the rest of the speakers.
In audio scenario, the signals which are non stationary, audio
processing can be done by using short-time Fourier transform
(STFT). Here the signal is taken in time-frequency domain
[1]. An application which grows abruptly in various fields i.e.
speech processing, telecommunication and many more [2].
Here the separated signal is the mixture form of interference
which includes the voice interference or a noise generated in
the path of communication and the estimation of the variance
of that signal is not desired as compared to the presented
signal [3]. In digital signal processing, there are different
Wiener algorithms, which include maximum likelihood
estimation and negentropy maximization. These are the
computational methods for separating a mixed signal into
additive non-Gaussian sub-elements. The signal comes out of
this method is complex and this complexity is greater than its
applied source signal. The problem of this blind convolutive
mixture is sort-out from the algorithm popularly known as
Wiener filtering approach. This is based on fixed point
algorithm iteration to maximize nongaussianity to a great
extent as a measure of statistical independence.
In this paper we propose implementation of blind
audio source separation using the dedicated approach
popularly known as Wiener filtering, which is used to recover
two or more audio source signals obtained from unknown
linear combinations. This work proposes a Wiener filtering
technique for ICA algorithm which is having better quality of
separation and fast convergence in terms of signal to
interference ratio. The rest of this paper is formed as. In the
next section, we define blind audio source separation (BASS).
Section III describes the Weiner filtering technique and
algorithm for blind audio source separation. Section IV
produces the Results and discussion. The conclusion of this
paper and improved separation effect are given in Section V.
II. BASS TECHNIQUE
The purpose of BASS is to recover audio (source) signal from
a mixture of different audio signal generated by a number of
B