=Paper=
{{Paper
|id=Vol-2416/paper35
|storemode=property
|title=Increasing signal/acoustic interference ratio in telecommunications audio exchange by adaptive filtering methods
|pdfUrl=https://ceur-ws.org/Vol-2416/paper35.pdf
|volume=Vol-2416
|authors=Yuriy Kropotov,Aleksey Belov,Aleksander Proskuryakov
}}
==Increasing signal/acoustic interference ratio in telecommunications audio exchange by adaptive filtering methods ==
https://ceur-ws.org/Vol-2416/paper35.pdf
Increasing signal/acoustic interference ratio in
telecommunications audio exchange by adaptive filtering
methods
Y A Kropotov1, A A Belov1 and A Y Prockuryakov1
1
Murom Institute (branch) "Vladimir State University named after Alexander and Nicholay
Stoletovs", Orlovskaya street, 23, Murom, Russia, 602264
e-mail: kaf-eivt@yandex.ru
Abstract. The paper deals with the issues of increasing signal/noise ratio in
telecommunication audio exchange systems. The study of characteristics of speech signals and
acoustic noises, such as mathematical expectation, dispersion, relative intensity of acoustic
speech signals and various types of acoustic noises and interference is carried out. It is shown
that in the design of telecommunications systems, in particular loudspeaker systems operating
under the influence of external acoustic noise of high intensity, it is necessary to solve the
problem of developing algorithms to effectively suppress the above mentioned interference to
ensure the necessary signal/noise ratio in communication systems. A mathematical model of
the autocorrelation function of the speech signal by using the Lagrange interpolation
polynomial of order 10, considered the creation of adaptive algorithms to suppress acoustic
noise by linear filtering methods. Thus suppression of acoustic noises and hindrances is
possible at the expense of operated change of area of a cutting in the interval from 0 Hz to 300-
1000 Hz, depending on a hindrance conditions.
1. Introduction
The issues of increasing signal/acoustic interference in telecommunications audio exchange are
relevant and widely covered in the domestic [1-4] and foreign literature [5-8]. As the main criteria for
the effectiveness of telecommunications systems, to ensure a reliable exchange of information
between the subscribers of the system, we can cite the signal/acoustic noise ratio and syllable
intelligibility [1].
Reliable exchange of audio information in communication and telecommunication systems is
possible with the provision of signal/external acoustic noise ratio in Ps Pn ≥ 20 dB, which,
accordingly, will provide the necessary syllabic intelligibility S ≥ 93% for the effective, complete
exchange of audio data and operational-command information in loudspeaker systems at
multifunctional facilities of various purposes. For this reason, it is important to continue research
aimed at developing new approaches and algorithms to increase signal-to-acoustic interference ratio
by methods of adaptive filtration with the use of controlled change of the resection area.
V International Conference on "Information Technology and Nanotechnology" (ITNT-2019)
Data Science
Y A Kropotov, A A Belov and A Y Prockuryakov
2. Investigation of the characteristics of speech signals and acoustic noise
The known mathematical model of the function of probability density P(xi) of speech signals,
developed in [9], allows to receive values of mathematical expectation xi and dispersion σ 2 of
speech signals according to expressions
2 n −1 2 n −1
xi = ∑ x ⋅ P( x ) ,
i i σ = 2
∑ x P( x ) ,
2
i i (1)
i = −2 +1
n
i = − ( 2 −1)
n
where xi is the speech signal samples quantized at 2n intervals and normalized relative to the
maximum values in the form of ± x max = 1 V, then we obtain the values xi ≈ 0 and σ 2 = 0 ,126
W/Ohm.
Accordingly, the level of sound pressure intensity I relative to the zero level of intensity
I0=10-12 W/m2 is defined as
I kσ 2
( dB ) = 10 lg ⋅ 1012 , (2)
I0 S
where k is the directivity coefficient of a loudspeaker device, in case of radiation in one direction is
accepted k = 2; S = 4πR 2 is a sphere area, R is a sphere radius in meters.
Then the relative intensity of the acoustic speech signal Iр calculated by means of expression (2)
has the following value
Ip
( dB ) = 103dB ,
I0
in the case of σ 2 = 0,126 W, k = 2, R=1 m.
The relative intensities for different acoustic noises and acoustic disturbances are also calculated
using the values of mathematical expectation and dispersions of different types of acoustic
disturbances determined by formulas (1), in which the mathematical model of the function of
probability density of acoustic disturbances is applied [10]. Values of relative intensities for different
acoustic signals are presented in Table 1.
Table 1. Relative intensities for different acoustic signals.
I Ps
Type of signal P (W) (dB) (dB)
I0 Pn
Speech signal 0.126 103 –
Quantum noise 0.5∙10-5 59 44
Acoustic disturbances 3.5∙10-4 85 18
The noise of the sea 0.0687 90.4 12.6
Wind noise 0.0953 91.8 11.2
The noise in the engine room 0.111 92.5 10.5
Table 1 shows that in the presence of acoustic noise such as sea noise, wind noise, noise in the
engine room Ps Pn is in the range of 10,5÷18 dB. At such signal/interference relations, the syllable
intelligibility can be reduced to 65% for the case of Russian speech transmission. Reliable reception of
the transmitted speech information by the subscriber under such conditions is significantly
complicated.
According to the research in [10], for the correct reception of the transmitted voice message
through a noisy channel, it is necessary to provide a ratio Ps Pn of at least 20 dB. Therefore, in the
design of telecommunications systems, in particular, hands free systems operating under the influence
of external acoustic noise, the task is to create algorithms to effectively suppress the above mentioned
interference to ensure the necessary ratio of Ps Pn ≥ 20 dB.
Studies of the spectral functions of speech signals and external noise interference in [9] have shown
that the spectrum of the most common interference - external acoustic noise - is shifted relative to the
V International Conference on "Information Technology and Nanotechnology" (ITNT-2019) 272
Data Science
Y A Kropotov, A A Belov and A Y Prockuryakov
spectrum of speech signals in the low-frequency region, which suggests that increasing the ratio Ps Pn
can be applied linear filtering methods. Another factor allowing to apply linear filtration to suppress
acoustic noise is the possibility of increasing the cutting area in the range from F=0 Hz to Fl=300-
1000 Hz, according to studies in [10]. From these studies it can be seen that with the increase of the
lower boundary of the reproduced frequencies up to F=1000 Hz the value of S practically does not
change, it takes the value not less than 94% which is acceptable.
Thus, the above factors allow to consider the possibility of designing adaptive algorithms of speech
signal processing and suppression of acoustic interference due to controlled change of the cutting area
in the range from 0 Hz to (300÷1000) Hz, depending on the interference situation.
3. Development of an adaptive filtering algorithm with control of the resection area
Consider as a model of adaptive filtering a bandpass filter of rectangular shape with a floating low-
frequency cut Fl in the AFC channel. Varying Fl within 0≤Fl≤Fh, where Fh is a fixed high-frequency
section of the AFC filter. The AFC of the bandpass filter K(f) is specified as
1, at Fl ≤ f ≤ Fh ,
K( f ) =
0 , at other values f ;
where 0≤Fl≤Fh, Fh=6000 Hz.
With the known spectral function of the speech signal G ( f ) , Ps at the bandpass filter output is
defined as
Fh Fh
∫
Ps = K ( f )G( f )df = G( f )df , ∫
2
Fl Fl
where the second part of the Ps calculation expression is rewritten according to the K 2 ( f ) = 1
integration limits of Fl to Fh .
Correspondingly changing Fl within the range of 0≤ Fl ≤ Fh , we obtain the dependence of Ps on the
value of low-frequency Fl AFC of the channel. Write down the spectral function of the energy
spectrum of the speech signal in the form
N −1 k −1
G( ∆f ⋅ g ) = ∑ ∑ R( kT )e π
g =0 k =0
− j 2 ∆f ⋅ g ⋅kT
,
where ∆f = Fd is the sampling interval of the spectral function by frequency, N is number of speech
N
signal samples at the final interval, g is the number of the frequency discrete component of the spectral
function reference is within 0 ≤ g ≤ N-1, R(kT) is the voice signal ACF is represented as a grid
function in Table 2 for one of the voice signal implementations.
Table 2. Grid function of the voice signal ACF.
kh, kl k0=0 k1=7 k2=25 k3=54 k4=93 k5=130
R(kh) R(k0)=0,126 R(k1) = 0,115 R(k2)=0,037 R(k3) =-0,048 R(k4) = -0,016 R(k5) = -0,025
kh, kl k6=182 k7=182 k8=228 k9=245 k10=253
R(kh) R(k6)=0,026 R(k7)=0,008 R(k8) =-0,003 R(k9) =-0,001 R(k10)=0
Mathematical model of speech signal ACF is represented by approximation of interpolation
polynomial Lagrangian of the tenth order in the form of
H =10 H =10
k − kr
R( k ) = ∑ R( k ) ∏ k − k ,
k =0
h
r =0 q r
r ≠q
where k is the current delay multiple of the sampling period Т, kq is the delay in the node with the
number q, r is current delays of other nodes at r≠q.
Similarly, the power of acoustic noise Pn is calculated as
V International Conference on "Information Technology and Nanotechnology" (ITNT-2019) 273
Data Science
Y A Kropotov, A A Belov and A Y Prockuryakov
Fh
∫
Pn = GП ( f )df .
Fl
The spectral function of acoustic noise GП ( ∆f ⋅ g ) looks like
N −1 K −1
GП ( ∆f ⋅ g ) = ∑∑ R ( kT )l π
g =0 K =0
П
− j 2 ∆f ⋅g ⋅kT
,
where Rп(kТ) is ACF interference is also represented by its approximation by an interpolation
Lagrangian polynomial.
The process of filtration of the spectral functions of the speech signal G(f) and interference Gп(f)
and, accordingly, the change in the ratio Ps Pn depending on the change in Fl within
300 Hz ≤ Fl ≤ 1000 Hz is experimentally simulated and the results are illustrated in Figure 1. Figure 1
clearly shows that the area of the integrative function of Gп(f) corresponding to the Pn decreases faster
with changes in Fl from 300 Hz to 1000 Hz than the area of the integrative function of G(f). Figure 1
also demonstrates that the basic energy of the spectral function G(f) remains in the range from 0 Hz to
Fl, outside the bandwidth of the rectangular filter.
Figure 1. Illustration of the model of filtration of the spectral functions of the speech signal G(f) and
the interference Gп(f) depending on the change in the value of Fl.
Thus, the ratio of the values of Ps and Pn at the output of the rectangular filter shows the
dependence of the degree of suppression of acoustic noise relative to the speech signals on the value of
the resection region from 0 Hz to Fl in the AFC. In the discrete case of representation of spectral
functions, we obtain the ratio Ps/Pn in the form of
gh
Ps
∑ G( g ⋅ ∆f )
=
gl
gh
,
Pn
∑ G ( g ⋅ ∆f )
gl
П
where at duration of the analysis interval τ segm , the number of samples in the analysis interval
τ segm
N= , ∆f = Fd of the spectral function sampling interval by frequency, g = Fl is the number of
Т N ∆f
l
frequency interval for low-frequency section of AFC Fl.
Thus, changing g l in the ratio expression Ps/Pn will get a function of changing this ratio at the
output of the rectangular filter depending on the width of the suppression area, which is in the range
from 0 to Fl =gl ∆f .
V International Conference on "Information Technology and Nanotechnology" (ITNT-2019) 274
Data Science
Y A Kropotov, A A Belov and A Y Prockuryakov
The obtained results of studies of the effect of change in Fl on Ps Pn a fixed Fh=6000 Hz are shown
in Figure 2. In researches cases for three various realizations of speech signals concerning various
external noise hindrances are considered.
a) Wind noise b) The noise in the engine room
c) The noise of the sea
Figure 2. Dependence of the ratio Ps P n at the output of a rectangular filter on the value of Fl, at a
fixed value of Fh=6000 Hz.
4. Conclusion
From the diagrams in Figure 2 it can be seen that external acoustic interference of the "wind noise"
type, if set to Fl=500 Hz, is suppressed by -(17÷23) dB. At the influence of acoustic noise of the type
"noise in the engine room" and at Fl =700÷1000 Hz, such acoustic noise is suppressed up to -12 dB.
At the influence of acoustic interference of "sea noise" type at Fl = 800÷1000 Hz, this interference is
suppressed by (11÷15) dB [10]. The obtained results of acoustic noise suppression studies show that
the linear filtration method can provide the necessary ratio of Ps Pn ≥20 dB and, accordingly, the
necessary syllabic legibility of S ≥ 93% in the telecommunications system of voice information
exchange.
5. References
[1] Zheleznyak V K, Makarov Y K and Khoreev A A 2000 Some methodical approaches
toevaluation of efficiency of speech information protection Special technique 4 39-45
[2] Sapozhkov M A 1962 Speech signal in cybernetics and communications (Moscow: Svyazizdat)
p 452
[3] Martin R, Heute U and Antweiler C 2008 Advances in Digital Speech Transmission (London:
John Wiley and Son) p 572
[4] Ryabenkyi V S 2012 Mathematical model of the external noise suppression devices in the
subarea of space Mathematical modeling 24(8) 3-31
[5] McAulay R and Malpass M 1980 Speech enhancement using a soft-decision noise suppression
filter IEEE Trans, on Acoustics, Speech, and Signal Processing 28(2) 137-145
V International Conference on "Information Technology and Nanotechnology" (ITNT-2019) 275
Data Science
Y A Kropotov, A A Belov and A Y Prockuryakov
[6] Cohen I, Benesty J and Gannot S 2010 Speech processing in modern communication (Berlin,
Heidelberg: Springer) p 342
[7] Hansler E and Schmidt G 2006 Topics in acoustic echo and noise control: Selected methods for
the cancelation of acoustic echoes, the reduction of background noise, and speech processing
(Berlin, Heidelberg: Springer) p 642
[8] Kahrs M and Brandenburg K 2002 Applications of digital signal processing to audio and
acoustics (New York: Kluwer Academic Publisher) p 572
[9] Kropotov Y A and Belov A A 2016 Application method of barrier functions in the problem of
estimating the probability density of the parameterized approximations 13th International
Scientific-Technical Conference on Actual Problems of Electronic Instrument Engineering 69-
72
[10] Kolpakov A A and Kropotov Y A 2017 Advanced mixing audio streams for heterogeneous
computer systems in telecommunications CEUR Workshop Proceedings 1902 32-36
[11] Kropotov Y A and Proskuryakov A Y 2007 Mathematical model of probability law for the
amplitudes of speech waveforms in the exponential basises 17th International Crimean
Conference - Microwave and Telecommunication Technology 364-366
[12] Breining C 1999 Acoustic echo control. An application of very-high-order adaptive filters
Signal Processing Magazine 16(4) 42-69
[13] Boll S 1979 Suppression of acoustic noise in speech using spectral subtraction IEEE
Transactions on Acoustics, Speech, and Signal Processing 27(2) 113-120
[14] Diniz P S 2008 Adaptive filtering algoriths and practical implementation (New York: Springer)
p 627
[15] Hakansson L 2002 Noise Reduction in Speech Applications (Boca Raton, Florida: CRC press) p
384
[16] Kropotov Y A, Belov A A and Proskuryakov A Y 2018 Method for forecasting changes in time
series parameters in digital information management systems Computer Optics 42(6) 1093-
1100 DOI: 10.18287/2412-6179-2018-42-6-1093-1100
V International Conference on "Information Technology and Nanotechnology" (ITNT-2019) 276