Hardware Implementation of Speech Enhancement Algorithms
Hardware Implementation of Speech Enhancement Algorithms
Hardware Implementation of Speech Enhancement Algorithms
1258
Abstract
This paper implements a VLSI based Adaptive noise
canceller (ANC) using Least Mean-Square (LMS) and
Normalized Least Mean-Square (NLMS). Adaptive filters
are capable of adapting their filter coefficients as per the
variations in characteristics of input signal and noise to
achieve a noise free signal. These algorithms have been
implemented on Digital Signal Processor. Platform used for
implementation of algorithms is DSP processor
TMS320C6713.
Primary
Signal
d
Reference Signal
Adaptive Filter
-
III.
I.
II.
LMS ALGORITHM
INTRODUCTION
LMS algorithm is most general & simple form of
filtering. The LMS shows advantage of less computational
complexity & simple implementation [4]. LMS involves the
initial estimation of coefficient vector and updating in a
particular direction to achieve the desired response. At each
point the filter changes its coefficient from present state to
next state of coefficient vector. The filter coefficients are
adjusted such that the mean square error (MSE) should be
minimized with the objective of minimum of function
[5].The coefficient of the adaptive filter at minimum error
occurs at W = W0, represents that the current filter
coefficients equal the optimal coefficient values. Optimum
weight of filter can be given as
0 =
(1)
(2)
www.ijsret.org
International Journal of Scientific Research Engineering & Technology (IJSRET), ISSN 2278 0882
Volume 3, Issue 8, November 2014
+ 1 = J
(3)
(4)
(5)
(6)
IV.
1259
& error
With the terms of equation (5) & (6) equation (4) is used to
represent the equation of LMS algorithm [7]. The boundary
of operation for the adaptive filters depends on the step size
for the algorithm to be stable the operation boundary is
given by
2
0<<
,
(7)
.().()
()()
(8)
(9)
(10)
www.ijsret.org
International Journal of Scientific Research Engineering & Technology (IJSRET), ISSN 2278 0882
Volume 3, Issue 8, November 2014
1260
REFERENCES
Name
SNR in dB
Noisy Signal
Filtered Signal for LMS
Filtered Signal for NLMS
11.6323 dB
15.9174 dB
15.9924 dB
V.
CONCLUSION
www.ijsret.org