Discourse Feature Recognition for Text Dynamic Translation
Meng Li
Guangzhou Huali College, Guangzhou, 511325, China
Keywords: Text, Dynamic Translation, Discourse Feature Recognition, Parameter Modulation, Spectrum Feature Extrac-
tion.
Abstract: In order to improve the ability of dynamic translation of foreign media discourse, this paper puts forward a
method of discourse feature recognition and recognition based on spectral density feature decomposition.
Using time-frequency dynamic feature analysis, the dynamic compensation model of foreign media text trans-
lation discourse signal is established, and the noise parameters of foreign media text translation discourse
signal are analyzed and recognized by combining the frequency spectrum density feature estimation method.
The feature decomposition model of foreign media text translation discourse signal is constructed by multi-
frame speech compensation and parameter modulation method, and the wavelet multi-level feature detection
method is adopted. The anti-interference filtering analysis of foreign media text translation discourse is real-
ized. The conditional coding of text translation discourse signal is carried out by the method of spectral density
dynamic feature decomposition, and the statistical information analysis model of foreign media text transla-
tion discourse signal is established. The signal detection and feature recognition of foreign media text trans-
lation discourse signal are realized by text parameter clustering and envelope amplitude-frequency feature
detection. The test results show that the accuracy of foreign media text translation discourse feature recogni-
tion by this method is high, and the output signal-to-noise ratio of the signal is improved by discourse feature
recognition.
1 INTRODUCTION
With the development of foreign media text and dis-
course recognition technology, the accuracy of for-
eign media text translation and discourse recognition
is required higher and higher. A discourse feature
recognition model oriented to dynamic text transla-
tion is established, and a foreign media text transla-
tion discourse signal collection and feature recogni-
tion model is built by combining signal processing
and speech signal analysis methods (John, 2019). The
feature recognition and detection of foreign media
text translation discourse signals are carried out by
combining anti-interference design methods. Im-
prove the ability of signal detection and recognition,
so as to improve the output clarity of foreign media
text translation discourse signals and reduce the inter-
ference of signals. Studying the feature recognition
technology of foreign media text translation discourse
signals plays an important role in the further develop-
ment of foreign media texts (JIA, LAI, YU et al.,
2021).
The feature recognition of foreign media text
translation discourse signal is based on the automatic
test technology of the signal, and adopts the embed-
ded signal processing method combined with signal
feature analysis to carry out anti-interference pro-
cessing and feature analysis of foreign media text
translation discourse signal. Among the current meth-
ods, the feature recognition methods of foreign media
text translation discourse signal mainly include time-
frequency analysis method, statistical feature analysis
method, differential control method and spectral fea-
ture recognition method, etc (AI, YANG, XIONG,
2018). The error gain control model of foreign media
text translation discourse signal is constructed, and
combined with signal gain control, the foreign media
text translation discourse feature recognition pro-
cessing is realized, and the signal interference is re-
duced. However, the traditional method of foreign
media text translation discourse feature recognition
has poor output signal-to-noise ratio and weak adap-
tive control ability (Wang, Li, Meng, 2022).
Aiming at the disadvantages of traditional meth-
ods, this paper puts forward a method of foreign me-
dia text translation discourse feature recognition and
578
Li, M.
Discourse Feature Recognition for Text Dynamic Translation.
DOI: 10.5220/0011752300003607
In Proceedings of the 1st International Conference on Public Management, Digital Economy and Internet Technology (ICPDI 2022), pages 578-583
ISBN: 978-989-758-620-0
Copyright
c
2023 by SCITEPRESS Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)
recognition based on spectral density feature decom-
position. Firstly, the collected foreign media text
translation discourse signals are preprocessed, includ-
ing noise reduction filtering, anti-interference design,
feature extraction, etc., and then the signals are de-
composed by time-frequency decomposition and em-
pirical mode decomposition to realize foreign media
text translation discourse feature recognition. Finally,
the simulation test shows the superior performance of
this method in improving foreign media text transla-
tion discourse feature recognition and recognition
ability.
2 SIGNAL ACQUISITION
DETECTION AND
PREPROCESSING
2.1 Signal Sampling Analysis
In order to realize foreign media text translation dis-
course feature recognition based on spectral density
feature decomposition, firstly, the foreign media text
translation discourse signal is sampled, the time-fre-
quency dynamic feature analysis of foreign media
text translation discourse is adopted, and a dynamic
compensation model of foreign media text translation
discourse signal is established. M omni-directional
signal sources are set to collect foreign media text
translation discourse signal, an expected foreign me-
dia text translation discourse signal and p interference
feature quantitie (Zheng, 2022) s. The collected for-
eign media text translation discourse signals are input
into the TV signal detection system in array, and there
is a foreign media text translation discourse signal
with a length of
N
, which can be represented as ar-
ray, and it is represented as
()
N
x
nR
. Through the
encoding and decoding of foreign media text transla-
tion discourse signals, the characteristic marker
model of foreign media text translation discourse sig-
nals is obtained, and the acquisition output of foreign
media text translation discourse signals is expressed
as follows:
1
N
ii
i
x
ss
=
=Ψ
(1)
Wherein,
i
s
is the time domain component of for-
eign media text translation discourse signal,
i
Ψ
is the
phase of foreign media text translation discourse sig-
nal detection, and
s
Ψ
is the frequency domain com-
ponent of foreign media text translation discourse sig-
nal. The acquisition model of foreign media text
translation discourse signal is constructed, and the
best sampling starting point is obtained:
𝐽
*
(𝑚)=𝑚𝑎𝑥
{𝐽
*
(𝜏)+𝐷
(𝜏)+𝐶},
𝐽
*
(0) = 0
(2)
Wherein,
*
()J
τ
is the time delay of foreign me-
dia text translation discourse signal detection,
()
m
D
τ
is the characteristic decomposition level of foreign
media text translation discourse signal, and
C
is the
bandwidth of foreign media text translation discourse
signal. According to the constraint characteristics of
the modulation mode of foreign media text translation
discourse signals, the discrete components of broad-
band multi-frame foreign media text translation dis-
course signals after basic decoding are obtained, and
the discrete characteristic equation is expressed as
follows:
() () ()
()( )
() () ()
111
()
1
()
|, | ,
(|.)
iii
kkk k kk
i
k
i
k
xn sn vn
py X Y px X Y
qx
ω
−−
=+
=
(3)
In the above formula,
()
s
n
represents the dis-
crete sequence of foreign media text translation dis-
course signal,
()
vn
represents the color noise com-
ponent,
()
1
i
k
ω
represents the scale of foreign media
text translation discourse signal,
()
(|.)
i
k
qx
is the prior
probability density of foreign media text translation
discourse signal detection,
()
() ()
11
|,
ii
kkk
px X Y
−−
is the
conditional probability density of foreign media text
translation discourse signal reliability collection, and
SD is the random probability density component. The
domain feature expression is carried out on the broad-
band multi-frame foreign media text translation dis-
course signal, and the joint parameter identification is
carried out on the foreign media text translation dis-
course signal components in Y to construct the con-
ditional coding model of foreign media text transla-
tion discourse signal. The coding output is as follows:
1
(|) (|,,)
l
1i
xy
L
i
i
p
prl
=
α= α
(4)
Wherein,
(|,,)
i
y
i
p
rlα
is the encoded output of for-
eign media text translation discourse frame format,
and L is the sampling sequence length of foreign me-
dia text translation discourse signal. It is assumed that
the foreign media text translation discourse signal is
Discourse Feature Recognition for Text Dynamic Translation
579
decomposed by the adaptive spectrum separation
method, and the sampling model of the foreign media
text translation discourse signal is obtained (Wang,
2021).
2.2 Signal Filtering Pretreatment
By using time-frequency dynamic feature analysis, a
dynamic compensation model of foreign media text
translation discourse signal is established (Lin, 2021),
and the signal anti-interference design is carried out
by combining the estimation method of frequency
spectrum density feature when foreign media text
translation discourse signal is used, and the modula-
tion components of foreign media text translation dis-
course signal are obtained:
𝐸
=
𝐸
,

(5)
𝑃
,
= 𝐸
,
/𝐸
(6)
Wherein,
,
k
E
is the order of foreign media text
translation discourse signal, and
j
E
is the Doppler
frequency offset of foreign media text translation dis-
course. Time-frequency analysis method is used to
filter the foreign media text translation discourse sig-
nal in the interval, and wavelet multi-level feature de-
tector is used to obtain the modulation frequency
spectrum
k
WE
of foreign media text translation dis-
course signal as follows:
𝑊𝐸
=
𝑃
,
𝑙𝑛( 𝑃
,
) (7)
Wherein,
,
j
k
P
is the carrier frequency of foreign
media text translation discourse signal, and
,
ln( )
j
k
P
is the amplitude of filtering detection. The interfer-
ence suppression component is determined by the or-
der determination algorithm, and the modulation pro-
cessing of the broadband multi-frame foreign media
text translation discourse signal is carried out. The in-
put characteristic sequence of the broadband multi-
frame foreign media text translation discourse signal
is
(0),..., ( 1)xxNx=[ ]
, which is a discrete se-
quence of the finite-length broadband multi-frame
foreign media text translation discourse signal.
01nN≤≤
D, the empirical mode decomposi-
tion method is adopted to filter and detect the broad-
band multi-frame foreign media text translation dis-
course signal, and the average sum of the filtered out-
put of foreign media text translation discourse is ob-
tained.
𝑆
= 𝐸
𝑥
(𝑡)
+
𝑠𝑏 (8)
422
() 3 ()
x
K
Ex t E x t b
 
=−
 
(9)
Wherein, 𝐸
𝑥
(𝑡)
is the passband of foreign me-
dia text translation discourse signal transmission,
b
represents multi-frame information gain,
4
()Ext


is the modulation frequency offset of foreign media
text translation discourse, and
s
is the joint probabil-
ity density distribution. According to the filtering re-
sults, the multi-frame speech compensation and pa-
rameter modulation method are used to construct the
feature decomposition model of foreign media text
translation discourse signal, which can improve the
ability of discourse feature recognition and pro-
cessing (ABDUL RAUF, SCHWENK, 2011).
3 DISCOURSE FEATURE
RECOGNITION AND
PROCESSING
3.1 Empirical Mode Decomposition of
Signal
The method of multi-frame speech compensation and
parameter modulation is used to construct the feature
decomposition model of foreign media text transla-
tion discourse signal, and the method of spatial pa-
rameter recognition is used to construct the ambiguity
detection model of foreign media text translation dis-
course signal (MARIE, FUJITA, 2017). After the in-
terference position is determined and zeroed, the dy-
namic amplitude feature recognition model of broad-
band multi-frame foreign media text translation dis-
course signal is obtained. The joint decision function
is:
𝐻
: 𝑥'(𝑡)=𝑤(𝑡)
𝐻
:
𝐸𝑠'(𝑡)+𝑤(𝑡)
0 ≤𝑡≤𝑇 (10)
In the formula, x'(t) and s'(t) are:
𝑥'(𝑡)=𝑥(𝑡)*
(𝑡) (11)
𝑠'(𝑡)=𝑠(𝑡)*
(𝑡) (12)
Wherein,
()wt
is the single-line spectrum of for-
eign media text translation discourse signal,
T
is the
time sampling interval,
E
is the distributed energy
spectrum of foreign media text translation discourse,
𝑥'(𝑡) is the convolution of foreign media text transla-
ICPDI 2022 - International Conference on Public Management, Digital Economy and Internet Technology
580
tion discourse signal,
(𝑡) is the filter transfer func-
tion of foreign media text translation discourse signal,
the M-order phase polynomial is determined, and the
improved empirical mode decomposition method is
adopted. The scale decomposition coefficient of the
broadband multi-frame foreign media text translation
discourse signal is
(1) ( ) ( 1)
0
, , 2
jj
NNNN jJ
==
, and
each sub-signal is described by
𝑃
=1, so the out-
put probability density characteristic quantity of the
broadband multi-frame foreign media text translation
discourse signal satisfies. The energy set of the broad-
band multi-frame foreign media text translation dis-
course signal is described by {𝑃
, 𝑃
,...,𝑃
}, and the
single-line spread characteristic decomposition
method is adopted. The residual frequency offset of
the broadband multi-frame foreign media text trans-
lation discourse signal after discrete orthogonal
wavelet transform is obtained, and {𝑃
, 𝑃
,...,𝑃
} is
used to represent the finite time sequence of the
broadband multi-frame foreign media text translation
discourse signal, that is, the discrete sequence output
by empirical mode decomposition of the foreign me-
dia text translation discourse signal is:
() ( )
[0,..., 1]XXN=−X
(13)
Wherein,
N
is the discrete sequence length of
foreign media text translation discourse signal. Ac-
cording to the empirical mode decomposition of for-
eign media text translation discourse signal, the fre-
quency parameters of foreign media text translation
discourse signal are estimated as follows:
1
0
1
ˆ
()
2
p
i
ii
i
f
nian
π
=
=
(14)
Wherein, 𝑖 is the sampling sequence point of for-
eign media text translation discourse signal, and 𝑎
is
the sampling track of foreign media text translation
discourse signal, 𝑛

is Z transform. At the receiving
end, through information output identification, the
time domain discrete components of foreign media
text translation discourse signals are expressed as fol-
lows:
{
}
12
,,,
n
F
xx x=
(15)
Thus, the discrete components of the signal are
obtained by empirical mode decomposition. Accord-
ing to the constraint feature quantity of foreign media
text translation discourse signal modulation mode, the
signal discrete sequence of foreign media text trans-
lation discourse signal modulation is obtained, and
the feature decomposition model of foreign media
text translation discourse signal is constructed by
multi-frame speech compensation and parameter
modulation method, so as to improve the feature
recognition ability of the signal (ZHANG, LI, YANG
et al., 2020).
3.2 Realization Of Discourse Feature
Recognition
The amplitude of modulation mode of foreign media
text translation discourse signal obtained by fre-
quency domain parameter identification is:
2
,
1
() ( , ) arg[ ( )]
2
ab
d dadb
yt ab Z f
df
a
ρ
π
=−

(16)
In the above formula, 𝑓(𝑡) is the estimated value
of modulation parameters of foreign media text trans-
lation discourse signal, 𝜌(𝑎, 𝑏) is the ambiguity of
foreign media text translation discourse signal, and
a
is the broadband multi-frame length of foreign media
text translation discourse signal. The modulation
model of foreign media text translation discourse sig-
nal is constructed, and the modulation output expres-
sion is as follows:
()
() ( )+( )
jf
Ku Z f Af e
θ
=
(17)
Wherein,
𝑍(𝑓) represents the real part of foreign
media text translation discourse signal,
𝐴(𝑓) repre-
sents the i333maginary part, and
𝑒
()
represents the
frequency offset. According to the foreign media text
translation discourse signal modulation processing,
the output expression of discourse feature recognition
can be expressed as:
2
1
() ( )
2
jft
Yu S f e df
π
π
−∞
=
(18)
Wherein,
𝑆(𝑓) represents the energy spectrum
component of foreign media text translation dis-
course, and
𝑓 represents the modulation frequency of
foreign media text translation discourse. The steep
gradient of modulation of foreign media text transla-
tion discourse signal is calculated. Based on the intel-
ligent recognition method, the conditional coding of
foreign media text translation discourse signal is car-
ried out, and the statistical information analysis
model of foreign media text translation discourse sig-
nal is established. Through text parameter clustering
and envelope amplitude-frequency feature detection,
the signal detection and feature recognition of foreign
media text translation discourse signal are realized.
Discourse Feature Recognition for Text Dynamic Translation
581
4 SIMULATION TEST
In order to verify the application performance of this
method in foreign media text translation discourse
feature recognition, Matlab is used for experimental
analysis, and the sample number of foreign media text
translation discourse signal is set to 20, the signal
frame length is 2000~4000, and the chip length is
4016. See Table 1 for the initial parameter settings of
the signal.
Table 1: Initial parameter setting of signal.
Signal
sequence
Bandwidth
/Kbps
Interference
intensity
/dB
Sampling
frequency
/KHz
Sample 1 12.034 3.772 21.000
Sample 2 12.592 3.476 21.516
Sample 3 12.695 3.592 21.370
Sample 4 12.016 3.194 21.886
Sample 5 12.439 3.310 21.262
Sample 6 12.548 3.468 21.982
Sample 7 12.316 3.014 21.372
Sample 8 12.537 3.462 21.695
Sample 9 12.184 3.249 21.661
Sample 10 12.525 3.649 21.775
Sample 11 12.168 3.052 21.585
Sample 12 12.761 3.486 21.181
Sample 13 12.339 3.479 21.558
Sample 14 12.538 3.696 21.358
Sample 15 12.209 3.502 21.312
Sample 16 12.088 2.981 21.577
Sample 17 12.674 3.066 14.176
Sample 18 12.034 3.772 21.000
Sample 19 12.592 3.476 21.516
Sample 20 12.695 3.592 21.370
Figure 1: Original signal.
According to the parameter settings in Table 1,
foreign media text translation discourse feature
recognition processing is performed, and the original
signal is shown in Fig 1, and the signal after feature
recognition is shown in Fig 2.
Figure 2: Discourse feature recognition.
By comparing the results of Fig 1 and Fig 2, it can
be seen that the peak gain of the foreign media text
translation speech signal is obviously expressed by
this method, which indicates that the foreign media
text translation speech signal has strong anti-interfer-
ence ability, and the output signal-to-noise ratio
(SNR) is tested, and the comparison result is shown
in Fig 3. By analyzing the results of Fig 3, it can be
seen that the output signal-to-noise ratio (SNR) of the
foreign media text translation speech signal is higher
than that of the traditional method.
Figure 3: Contrast test of output signal-to-noise ratio.
ICPDI 2022 - International Conference on Public Management, Digital Economy and Internet Technology
582
5 CONCLUSIONS
In this paper, a model of foreign media text transla-
tion discourse signal collection and feature recogni-
tion is constructed, and the feature recognition and
detection of foreign media text translation discourse
signal are carried out by combining anti-interference
design methods, so as to improve the ability of signal
detection and recognition, and thus improve the abil-
ity of foreign media text translation discourse feature
recognition. In this paper, a method of speech feature
recognition and recognition of foreign media text
translation based on spectral density feature decom-
position is proposed. Based on the intelligent recog-
nition method, the conditional coding of text transla-
tion speech signal is carried out, the statistical infor-
mation analysis model of foreign media text transla-
tion speech signal is established, and the speech fea-
ture recognition is realized by combining signal fil-
tering and detection recognition. The test shows that
the output signal-to-noise ratio (SNR) of the feature
recognition processing of multi-frame foreign media
texts is high and the feature recognition effect is good.
REFERENCES
ABDUL RAUF S,SCHWENK H. Parallel sentence gener-
ation from comparable corpora for improved SMT[J].
Machine Translation, 2011,25(4):341-375.
AI Z L,YANG M M, XIONG D Y. Data augmentation for
neural machine translation[J]. Journal of Chinese Infor-
mation Processing,2018,32(7):30-36.
Haixing Wang. Exploration on the Innovative Path of Col-
lege Students' Mental Health Education under the Back-
ground of Epidemic Situation--Take the Psychological
Salon as an Example[J]. Advances in Educational
Technology and Psychology, 2021, 5(8):66-70.
John K. Aderibigbe and Sihle Mahola and Tendia
Chimucheka. Influence of entrepreneurship risk per-
ceptions and aversion on entrepreneurial intention
among university students in South Africa[J].
AFFRIKA Journal of Politics, Economics and Society,
2019, 9(1): 239-252.
JIA C X, LAI H, YU Z T, et al. Pseudo-parallel corpus gen-
eration for Chinese-Vietnamese neural machine trans-
lation based on pivot language[J]. Computer Engineer-
ing and Science,2021, 43(3):543-550.
MARIE B, FUJITA A. Efficient extraction of pseudo-par-
allel sentences from raw monolingual data using word
embeddings[C]//Proceedings of the 2017 55th Annual
Meeting of the Association for Computational Linguis-
tics. Stroudsburg:ACL,2017:392-398.
Peng Wang, Dan Li, Xiangyin Meng. The Management
Mode of College Students' Mental Health Based on
Wireless Sensor Network[J]. International Journal of
Frontiers in Sociology, 2022, 40(2):45-49.
Yi Lin. Analysis on the Innovative Strategies of College
Students' Mental Health Education under the Back-
ground of We-Media[J]. International Journal of Edu-
cation and Teaching Research, 2021, 2(3):32-38.
ZHANG X L,LI X,YANG Y T,et al. Analysis of bi-direc-
tional reranking model for Uyghur-Chinese neural ma-
chine translation[J]. Acta Scientiarum Naturalium Uni-
versitatis Pekinensis, 2020,56(1):31-38.
Zheng Wanting. Cluster Analysis Algorithm in the Analy-
sis of College Students' Mental Health Education.[J].
Applied bionics and biomechanics, 2022, 63(6):19-21.
Discourse Feature Recognition for Text Dynamic Translation
583