YrMN11MP

1
2%% 
3%% 
4%% 
5%% 
6%% 
7%% 
8%%
9%% This is a skeleton file demonstrating the use of IEEEtran.cls
10%% (requires IEEEtran.cls version 1.7 or later) with an IEEE journal paper.
11%%
12%% Support sites:
13%% http://www.michaelshell.org/tex/ieeetran/
14%% http://www.ctan.org/tex-archive/macros/latex/contrib/IEEEtran/
15%% and
16%% http://www.ieee.org/
17
18
19
20% *** Authors should verify (and, if needed, correct) their LaTeX system  ***
21% *** with the testflow diagnostic prior to trusting their LaTeX platform ***
22% *** with production work. IEEE's font choices can trigger bugs that do  ***
23% *** not appear when using other class files.                            ***
24% The testflow support page is at:
25% http://www.michaelshell.org/tex/testflow/
26
27
28%%*************************************************************************
29%% Legal Notice:
30%% This code is offered as-is without any warranty either expressed or
31%% implied; without even the implied warranty of MERCHANTABILITY or
32%% FITNESS FOR A PARTICULAR PURPOSE! 
33%% User assumes all risk.
34%% In no event shall IEEE or any contributor to this code be liable for
35%% any damages or losses, including, but not limited to, incidental,
36%% consequential, or any other damages, resulting from the use or misuse
37%% of any information contained here.
38%%
39%% All comments are the opinions of their respective authors and are not
40%% necessarily endorsed by the IEEE.
41%%
42%% This work is distributed under the LaTeX Project Public License (LPPL)
43%% ( http://www.latex-project.org/ ) version 1.3, and may be freely used,
44%% distributed and modified. A copy of the LPPL, version 1.3, is included
45%% in the base LaTeX documentation of all distributions of LaTeX released
46%% 2003/12/01 or later.
47%% Retain all contribution notices and credits.
48%% ** Modified files should be clearly indicated as such, including  **
49%% ** renaming them and changing author support contact information. **
50%%
51%% File list of work: IEEEtran.cls, IEEEtran_HOWTO.pdf, bare_adv.tex,
52%%                    bare_conf.tex, bare_jrnl.tex, bare_jrnl_compsoc.tex
53%%*************************************************************************
54
55% Note that the a4paper option is mainly intended so that authors in
56% countries using A4 can easily print to A4 and see how their papers will
57% look in print - the typesetting of the document will not typically be
58% affected with changes in paper size (but the bottom and side margins will).
59% Use the testflow package mentioned above to verify correct handling of
60% both paper sizes by the user's LaTeX system.
61%
62% Also note that the "draftcls" or "draftclsnofoot", not "draft", option
63% should be used if it is desired that the figures are to be displayed in
64% draft mode.
65%
66\documentclass[journal]{IEEEtran}
67\usepackage{blindtext}
68\usepackage{graphicx}
69
70% Some very useful LaTeX packages include:
71% (uncomment the ones you want to load)
72
73
74% *** MISC UTILITY PACKAGES ***
75%
76%\usepackage{ifpdf}
77% Heiko Oberdiek's ifpdf.sty is very useful if you need conditional
78% compilation based on whether the output is pdf or dvi.
79% usage:
80% \ifpdf
81%   % pdf code
82% \else
83%   % dvi code
84% \fi
85% The latest version of ifpdf.sty can be obtained from:
86% http://www.ctan.org/tex-archive/macros/latex/contrib/oberdiek/
87% Also, note that IEEEtran.cls V1.7 and later provides a builtin
88% \ifCLASSINFOpdf conditional that works the same way.
89% When switching from latex to pdflatex and vice-versa, the compiler may
90% have to be run twice to clear warning/error messages.
91
92
93
94
95
96
97% *** CITATION PACKAGES ***
98%
99%\usepackage{cite}
100% cite.sty was written by Donald Arseneau
101% V1.6 and later of IEEEtran pre-defines the format of the cite.sty package
102% \cite{} output to follow that of IEEE. Loading the cite package will
103% result in citation numbers being automatically sorted and properly
104% "compressed/ranged". e.g., [1], [9], [2], [7], [5], [6] without using
105% cite.sty will become [1], [2], [5]--[7], [9] using cite.sty. cite.sty's
106% \cite will automatically add leading space, if needed. Use cite.sty's
107% noadjust option (cite.sty V3.8 and later) if you want to turn this off.
108% cite.sty is already installed on most LaTeX systems. Be sure and use
109% version 4.0 (2003-05-27) and later if using hyperref.sty. cite.sty does
110% not currently provide for hyperlinked citations.
111% The latest version can be obtained at:
112% http://www.ctan.org/tex-archive/macros/latex/contrib/cite/
113% The documentation is contained in the cite.sty file itself.
114
115
116
117
118
119
120% *** GRAPHICS RELATED PACKAGES ***
121%
122\ifCLASSINFOpdf
123  % \usepackage[pdftex]{graphicx}
124  % declare the path(s) where your graphic files are
125  % \graphicspath{{../pdf/}{../jpeg/}}
126  % and their extensions so you won't have to specify these with
127  % every instance of \includegraphics
128  % \DeclareGraphicsExtensions{.pdf,.jpeg,.png}
129\else
130  % or other class option (dvipsone, dvipdf, if not using dvips). graphicx
131  % will default to the driver specified in the system graphics.cfg if no
132  % driver is specified.
133  % \usepackage[dvips]{graphicx}
134  % declare the path(s) where your graphic files are
135  % \graphicspath{{../eps/}}
136  % and their extensions so you won't have to specify these with
137  % every instance of \includegraphics
138  % \DeclareGraphicsExtensions{.eps}
139\fi
140% graphicx was written by David Carlisle and Sebastian Rahtz. It is
141% required if you want graphics, photos, etc. graphicx.sty is already
142% installed on most LaTeX systems. The latest version and documentation can
143% be obtained at: 
144% http://www.ctan.org/tex-archive/macros/latex/required/graphics/
145% Another good source of documentation is "Using Imported Graphics in
146% LaTeX2e" by Keith Reckdahl which can be found as epslatex.ps or
147% epslatex.pdf at: http://www.ctan.org/tex-archive/info/
148%
149% latex, and pdflatex in dvi mode, support graphics in encapsulated
150% postscript (.eps) format. pdflatex in pdf mode supports graphics
151% in .pdf, .jpeg, .png and .mps (metapost) formats. Users should ensure
152% that all non-photo figures use a vector format (.eps, .pdf, .mps) and
153% not a bitmapped formats (.jpeg, .png). IEEE frowns on bitmapped formats
154% which can result in "jaggedy"/blurry rendering of lines and letters as
155% well as large increases in file sizes.
156%
157% You can find documentation about the pdfTeX application at:
158% http://www.tug.org/applications/pdftex
159
160
161
162
163
164% *** MATH PACKAGES ***
165%
166%\usepackage[cmex10]{amsmath}
167% A popular package from the American Mathematical Society that provides
168% many useful and powerful commands for dealing with mathematics. If using
169% it, be sure to load this package with the cmex10 option to ensure that
170% only type 1 fonts will utilized at all point sizes. Without this option,
171% it is possible that some math symbols, particularly those within
172% footnotes, will be rendered in bitmap form which will result in a
173% document that can not be IEEE Xplore compliant!
174%
175% Also, note that the amsmath package sets \interdisplaylinepenalty to 10000
176% thus preventing page breaks from occurring within multiline equations. Use:
177%\interdisplaylinepenalty=2500
178% after loading amsmath to restore such page breaks as IEEEtran.cls normally
179% does. amsmath.sty is already installed on most LaTeX systems. The latest
180% version and documentation can be obtained at:
181% http://www.ctan.org/tex-archive/macros/latex/required/amslatex/math/
182
183
184
185
186
187% *** SPECIALIZED LIST PACKAGES ***
188%
189%\usepackage{algorithmic}
190% algorithmic.sty was written by Peter Williams and Rogerio Brito.
191% This package provides an algorithmic environment fo describing algorithms.
192% You can use the algorithmic environment in-text or within a figure
193% environment to provide for a floating algorithm. Do NOT use the algorithm
194% floating environment provided by algorithm.sty (by the same authors) or
195% algorithm2e.sty (by Christophe Fiorio) as IEEE does not use dedicated
196% algorithm float types and packages that provide these will not provide
197% correct IEEE style captions. The latest version and documentation of
198% algorithmic.sty can be obtained at:
199% http://www.ctan.org/tex-archive/macros/latex/contrib/algorithms/
200% There is also a support site at:
201% http://algorithms.berlios.de/index.html
202% Also of interest may be the (relatively newer and more customizable)
203% algorithmicx.sty package by Szasz Janos:
204% http://www.ctan.org/tex-archive/macros/latex/contrib/algorithmicx/
205
206
207
208
209% *** ALIGNMENT PACKAGES ***
210%
211%\usepackage{array}
212% Frank Mittelbach's and David Carlisle's array.sty patches and improves
213% the standard LaTeX2e array and tabular environments to provide better
214% appearance and additional user controls. As the default LaTeX2e table
215% generation code is lacking to the point of almost being broken with
216% respect to the quality of the end results, all users are strongly
217% advised to use an enhanced (at the very least that provided by array.sty)
218% set of table tools. array.sty is already installed on most systems. The
219% latest version and documentation can be obtained at:
220% http://www.ctan.org/tex-archive/macros/latex/required/tools/
221
222
223%\usepackage{mdwmath}
224%\usepackage{mdwtab}
225% Also highly recommended is Mark Wooding's extremely powerful MDW tools,
226% especially mdwmath.sty and mdwtab.sty which are used to format equations
227% and tables, respectively. The MDWtools set is already installed on most
228% LaTeX systems. The lastest version and documentation is available at:
229% http://www.ctan.org/tex-archive/macros/latex/contrib/mdwtools/
230
231
232% IEEEtran contains the IEEEeqnarray family of commands that can be used to
233% generate multiline equations as well as matrices, tables, etc., of high
234% quality.
235
236
237%\usepackage{eqparbox}
238% Also of notable interest is Scott Pakin's eqparbox package for creating
239% (automatically sized) equal width boxes - aka "natural width parboxes".
240% Available at:
241% http://www.ctan.org/tex-archive/macros/latex/contrib/eqparbox/
242
243
244
245
246
247% *** SUBFIGURE PACKAGES ***
248%\usepackage[tight,footnotesize]{subfigure}
249% subfigure.sty was written by Steven Douglas Cochran. This package makes it
250% easy to put subfigures in your figures. e.g., "Figure 1a and 1b". For IEEE
251% work, it is a good idea to load it with the tight package option to reduce
252% the amount of white space around the subfigures. subfigure.sty is already
253% installed on most LaTeX systems. The latest version and documentation can
254% be obtained at:
255% http://www.ctan.org/tex-archive/obsolete/macros/latex/contrib/subfigure/
256% subfigure.sty has been superceeded by subfig.sty.
257
258
259
260%\usepackage[caption=false]{caption}
261%\usepackage[font=footnotesize]{subfig}
262% subfig.sty, also written by Steven Douglas Cochran, is the modern
263% replacement for subfigure.sty. However, subfig.sty requires and
264% automatically loads Axel Sommerfeldt's caption.sty which will override
265% IEEEtran.cls handling of captions and this will result in nonIEEE style
266% figure/table captions. To prevent this problem, be sure and preload
267% caption.sty with its "caption=false" package option. This is will preserve
268% IEEEtran.cls handing of captions. Version 1.3 (2005/06/28) and later 
269% (recommended due to many improvements over 1.2) of subfig.sty supports
270% the caption=false option directly:
271%\usepackage[caption=false,font=footnotesize]{subfig}
272%
273% The latest version and documentation can be obtained at:
274% http://www.ctan.org/tex-archive/macros/latex/contrib/subfig/
275% The latest version and documentation of caption.sty can be obtained at:
276% http://www.ctan.org/tex-archive/macros/latex/contrib/caption/
277
278
279
280
281% *** FLOAT PACKAGES ***
282%
283%\usepackage{fixltx2e}
284% fixltx2e, the successor to the earlier fix2col.sty, was written by
285% Frank Mittelbach and David Carlisle. This package corrects a few problems
286% in the LaTeX2e kernel, the most notable of which is that in current
287% LaTeX2e releases, the ordering of single and double column floats is not
288% guaranteed to be preserved. Thus, an unpatched LaTeX2e can allow a
289% single column figure to be placed prior to an earlier double column
290% figure. The latest version and documentation can be found at:
291% http://www.ctan.org/tex-archive/macros/latex/base/
292
293
294
295%\usepackage{stfloats}
296% stfloats.sty was written by Sigitas Tolusis. This package gives LaTeX2e
297% the ability to do double column floats at the bottom of the page as well
298% as the top. (e.g., "\begin{figure*}[!b]" is not normally possible in
299% LaTeX2e). It also provides a command:
300%\fnbelowfloat
301% to enable the placement of footnotes below bottom floats (the standard
302% LaTeX2e kernel puts them above bottom floats). This is an invasive package
303% which rewrites many portions of the LaTeX2e float routines. It may not work
304% with other packages that modify the LaTeX2e float routines. The latest
305% version and documentation can be obtained at:
306% http://www.ctan.org/tex-archive/macros/latex/contrib/sttools/
307% Documentation is contained in the stfloats.sty comments as well as in the
308% presfull.pdf file. Do not use the stfloats baselinefloat ability as IEEE
309% does not allow \baselineskip to stretch. Authors submitting work to the
310% IEEE should note that IEEE rarely uses double column equations and
311% that authors should try to avoid such use. Do not be tempted to use the
312% cuted.sty or midfloat.sty packages (also by Sigitas Tolusis) as IEEE does
313% not format its papers in such ways.
314
315
316%\ifCLASSOPTIONcaptionsoff
317%  \usepackage[nomarkers]{endfloat}
318% \let\MYoriglatexcaption\caption
319% \renewcommand{\caption}[2][\relax]{\MYoriglatexcaption[#2]{#2}}
320%\fi
321% endfloat.sty was written by James Darrell McCauley and Jeff Goldberg.
322% This package may be useful when used in conjunction with IEEEtran.cls'
323% captionsoff option. Some IEEE journals/societies require that submissions
324% have lists of figures/tables at the end of the paper and that
325% figures/tables without any captions are placed on a page by themselves at
326% the end of the document. If needed, the draftcls IEEEtran class option or
327% \CLASSINPUTbaselinestretch interface can be used to increase the line
328% spacing as well. Be sure and use the nomarkers option of endfloat to
329% prevent endfloat from "marking" where the figures would have been placed
330% in the text. The two hack lines of code above are a slight modification of
331% that suggested by in the endfloat docs (section 8.3.1) to ensure that
332% the full captions always appear in the list of figures/tables - even if
333% the user used the short optional argument of \caption[]{}.
334% IEEE papers do not typically make use of \caption[]'s optional argument,
335% so this should not be an issue. A similar trick can be used to disable
336% captions of packages such as subfig.sty that lack options to turn off
337% the subcaptions:
338% For subfig.sty:
339% \let\MYorigsubfloat\subfloat
340% \renewcommand{\subfloat}[2][\relax]{\MYorigsubfloat[]{#2}}
341% For subfigure.sty:
342% \let\MYorigsubfigure\subfigure
343% \renewcommand{\subfigure}[2][\relax]{\MYorigsubfigure[]{#2}}
344% However, the above trick will not work if both optional arguments of
345% the \subfloat/subfig command are used. Furthermore, there needs to be a
346% description of each subfigure *somewhere* and endfloat does not add
347% subfigure captions to its list of figures. Thus, the best approach is to
348% avoid the use of subfigure captions (many IEEE journals avoid them anyway)
349% and instead reference/explain all the subfigures within the main caption.
350% The latest version of endfloat.sty and its documentation can obtained at:
351% http://www.ctan.org/tex-archive/macros/latex/contrib/endfloat/
352%
353% The IEEEtran \ifCLASSOPTIONcaptionsoff conditional can also be used
354% later in the document, say, to conditionally put the References on a 
355% page by themselves.
356
357
358
359
360
361% *** PDF, URL AND HYPERLINK PACKAGES ***
362%
363%\usepackage{url}
364% url.sty was written by Donald Arseneau. It provides better support for
365% handling and breaking URLs. url.sty is already installed on most LaTeX
366% systems. The latest version can be obtained at:
367% http://www.ctan.org/tex-archive/macros/latex/contrib/misc/
368% Read the url.sty source comments for usage information. Basically,
369% \url{my_url_here}.
370
371
372
373
374
375% *** Do not adjust lengths that control margins, column widths, etc. ***
376% *** Do not use packages that alter fonts (such as pslatex).         ***
377% There should be no need to do such things with IEEEtran.cls V1.6 and later.
378% (Unless specifically asked to do so by the journal or conference you plan
379% to submit to, of course. )
380
381
382% correct bad hyphenation here
383\hyphenation{op-tical net-works semi-conduc-tor}
384
385
386\begin{document}
387%
388% paper title
389% can use linebreaks \\ within to get better formatting as desired
390\title{Speech Recognition by Biometric voice Deconstruction}
391%
392%
393% author names and IEEE memberships
394% note positions of commas and nonbreaking spaces ( ~ ) LaTeX will not break
395% a structure at a ~ so this keeps an author's name from being broken across
396% two lines.
397% use \thanks{} to gain access to the first footnote area
398% a separate \thanks must be used for each paragraph as LaTeX2e's \thanks
399% was not built to handle multiple paragraphs
400%
401
402\author{Nargiz Nauryzbaikyzy,~\IEEEmembership{student, Suleyman Demirel University (SDU)}
403       % <-this % stops a space
404\thanks{O. Baimuratov is with the Department
405of Computer Engineering, 
406
407Suleyman Demirel University, Kaskelen,
408Kazakhstan, 040900 
409
410KZ e-mail: olimzhon.baimuratov@sdu.edu.kz}
411% <-this % stops a space
412\thanks{Revised: November 14, 2017.}}
413
414% note the % following the last \IEEEmembership and also \thanks - 
415% these prevent an unwanted space from occurring between the last author name
416% and the end of the author line. i.e., if you had this:
417% 
418% \author{....lastname \thanks{...} \thanks{...} }
419%                     ^------------^------------^----Do not want these spaces!
420%
421% a space would be appended to the last name and could cause every name on that
422% line to be shifted left slightly. This is one of those "LaTeX things". For
423% instance, "\textbf{A} \textbf{B}" will typeset as "A B" not "AB". To get
424% "AB" then you have to do: "\textbf{A}\textbf{B}"
425% \thanks is no different in this regard, so shield the last } of each \thanks
426% that ends a line with a % and do not let a space in before the next \thanks.
427% Spaces after \IEEEmembership other than the last one are OK (and needed) as
428% you are supposed to have spaces between the names. For what it is worth,
429% this is a minor point as most people would not even notice if the said evil
430% space somehow managed to creep in.
431
432
433
434% The paper headers
435\markboth{Research, November~2017}%
436{Shell \MakeLowercase{\textit{et al.}}: Bare Demo of IEEEtran.cls for Journals}
437% The only time the second header will appear is for the odd numbered pages
438% after the title page when using the twoside option.
439% 
440% *** Note that you probably will NOT want to include the author's ***
441% *** name in the headers of peer review papers.                   ***
442% You can use \ifCLASSOPTIONpeerreview for conditional compilation here if
443% you desire.
444
445
446
447
448% If you want to put a publisher's ID mark on the page you can do it like
449% this:
450%\IEEEpubid{0000--0000/00\$00.00~\copyright~2007 IEEE}
451% Remember, if you use this you must call \IEEEpubidadjcol in the second
452% column for its text to clear the IEEEpubid mark.
453
454
455
456% use for special paper notices
457%\IEEEspecialpapernotice{(Invited Paper)}
458
459
460
461
462% make the title area
463\maketitle 
464
465
466
467
468
469
470\begin{abstract}{ person identification, especially in critical environments, has always been a subject of great interest. However, it has gained a new dimension in a world threatened by a new kind of social networks that uses to broadcast its message. In this new scenario, classical identification methods (such as fingerprints or face recognition) have been forcedly replaced by alternative biometric characteristics such as voice, as sometimes this is the only feature available. The paper hypothesizes that a gender-dependent characterization of speakers combined with the use of a set of features derived from the components, resulting from the deconstruction of the voice into its glottal source and vocal tract estimates, will enhance recognition rates when compared to classical approaches. A general description about the main hypothesis and the methodology followed to extract the gender-dependent extended biometric parameters is given. Experimental validation is carried out both on a highly controlled acoustic condition database, and on a mobile phone network recorded under non-controlled acoustic conditions. Although voice recognition is mostly secure, it still has flaws.  To aid its acceptance, this biometric system can be combined with more traditional security features to provide an additional layer of security.  These can include using other biometrics, or security mechanisms such as RSA, PINs or a combination of several different mechanisms.  Through further development, voice recognition can be one of the most successful and largest applications of biometrics in the future to come.}
471%\boldmath
472
473\end{abstract}
474
475% IEEEtran.cls defaults to using nonbold math in the Abstract.
476% This preserves the distinction between vectors and scalars. However,
477% if the journal you are submitting to favors bold math in the abstract,
478% then you can use LaTeX's standard command \boldmath at the very start
479% of the abstract to achieve this. Many IEEE journals frown on math
480% in the abstract anyway.
481
482% Note that keywords are not normally used for peerreview papers.
483\begin{IEEEkeywords}
484voice biometry, voice processing, , speaker recognition, speaker characterization, source-tract separation.
485\end{IEEEkeywords}
486
487
488
489
490
491
492% For peer review papers, you can put extra information on the cover
493% page as needed:
494% \ifCLASSOPTIONpeerreview
495% \begin{center} \bfseries EDICS Category: 3-BBND \end{center}
496% \fi
497%
498% For peerreview papers, this IEEEtran command inserts a page break and
499% creates the second title. It will be ignored for other modes.
500\IEEEpeerreviewmaketitle
501
502
503
504\section{Introduction}{Historically, speech signal analysis and processing has attracted wide attention, especially by its multiple applications. For instance, automatic speaker recognition (ASR) or speech synthesis (SS) have been active research ares at least since 70s. More recently, voice has captured again researchers’ attention thanks to its usefulness in order to assess early vocal pathologies, and neurodegenerative and mental disorders among others. Progress achieved thanks to these new applications have allowed for a better understanding of the mechanism of voice production, which have led to an improvement in speaker characterization.
505
506Not only speaker characterization is a key aspect for speaker recognition but it is also a challenging task for different reasons. First of all, voice involves both physiological and behavioral aspects. Regarding physiological characteristics, voices from different individuals differ due to differences in the voice production organs. Regarding behavioral features, voice communication is influenced by socio-cultural and emotional aspects (vocabulary selection, accent, intonation style, etc.). Additionally, voice presents high intra-speaker variability not only due to emotional or temporal health issues, but also due to changes that occur in the voices as a direct result of aging or neurological deterioration.
507
508From a biological point of view, like in any other important motor activity, speech communication requires the interaction of neurophysiologic systems, the motor system, and the sensory system.
509
510The motor system plays an important role in speech production, particularly, the vocal tract which can be roughly divided into three different areas: infraglottic, glottal (where the vocal folds are located) and supraglottic, depending on the location or functionality of the different organs and muscles.
511
512In voiced speech production, i.e., when vocal fold vibration exists, the joint effects of the subglottal and supraglottal air pressure difference, the laryngeal muscle tension and the elasticity of the vocal folds, cause an opening and closure of the vocal folds,which produces a glottal source. The glottal source can be defined as the sound pressure pattern that is produced in the supraglottic cavities immediately after the vocal folds, which is related to the Liljencrants-Fant excitation, in its ideal form (Fant et al.,1985).This glottal source is then modulated by the supraglottic cavities to produce different categories of sounds/phonetic classes. There-fore, the vocal tract, which includes the speech production organs above the vocal folds, can be regarded as a filter that alters the frequency content of the glottal source due to its resonances (also known as formants: energy amplification) and anti resonances (energy attenuation). This circumstance allows the estimation of the vocal tract shape from the spectral shape of the voice signal.The different categories of speech in which voiced and unvoiced sounds are classified depend on the manner and point of articulation, which are characterized by the supraglottic cavities. It must be noted that whereas the point and manner of articulation is liable to be imitated, the glottal source, as is linked to vocal fold vibration pattern, is almost impossible to be forged.
513
514Speaker recognition systems have historically used different features in order to cover the variability present in voice. Taking into account the different nature of the features use for speaker recognition, we can classify feature extraction modules in two categories: high level features and low-level features. 
515
516High level features reflects behavioral characteristics of speakers, such as prosody (pitch, duration, and energy), phonetic information, pronunciation, emotion, stress, idiolect word usage, conversational patterns, or other acoustic events. These differences in the speaking habits result from the manner in which people have learned to use their speech mechanism; but at the same time, the socio-linguistic context, the education and the socio-economic environment play an important role in these differences. The main drawback, as reported in different studies of this kind of systems, is its necessity for more information for both training and testing phases if compared to low-level feature systems and are also easily forged. Another problem is that speaker recognition results tend to be corrupted by the errors in detecting phonetic events. On the other hand, these high-level features are less sensitive to noise and channel mismatch than the low-level ones. 
517
518Biometric and spectral levels can be considered as low-level features. Biometric level refers to the use of specific characteristics. In the speaker’s production of voice difficult to impost as they are related to physiological or/and behavioral aspects. On the other hand, spectral level has been extensively used in speaker recognition systems for feature extraction. Typical methods in this spectral level are as follows: Short-time spectrum(no matter if we use the exact representation or its approximation by filter banks), predictor coefficients [based on a linear model of speech production: formant frequencies and bandwidth defined as the resonance frequencies of the vocal tract: or even the formant trajectories. 
519
520Moreover, the front-end of a speaker recognition system seems to have been relegated to a secondary plane when compared to the research interest in classification and normalization methods. 
521
522As a result, improvements in speaker recognition have been tightly linked to improvements in classification and normalization methods rather than in the use of more accurate parameters to precisely model a speaker. 
523
524The hypotheses sustaining the proposed methodology are the following: male and female voices exhibit not only acoustic-phonetic differences but physiological variations as well, therefore a gender-dependent parameterization which also takes advantage of the voice production model, by incorporating voice-source and tract features, will provide a more precise characterization of speakers that will help us to increase overall recognition rates of speaker recognition systems. According to a simplified speech production model, voice can be regarded as the result of filtering an excitation signal with the transfer function of the vocal tract and the lip-radiation model. The methodology followed in this study relies on model inversion to obtain the following voicing speech estimates: glottal source (more related to phonation/physiology) and vocal tract (more related with acoustics/behavior). This methodology has been adapted to properly work on running speech, solving problems found on early works. Information extracted from these estimates in a gender-dependent basis may be used as a complement to classical parameters in classifying speaker patterns.The paper is organized as follows: after a brief introduction to voice production, the voice deconstruction algorithm based on inverse filtering and adapted to work on running speech is presented. “Feature Vector Composition” focuses on the frequency-domain parameterization performed on the glottal source estimate (GSE) and vocal tract estimate (VTE) obtained following this methodology. Section “Speaker Recognition” describes the experimental framework defined to validate the proposed GDEB parameterization from different points of view as follows: the speaker recognition system implemented (including not only the front-end but the classification method), the different defined scenarios in which the proposed is evaluated, and the metrics used to measure the performance of the system.}
525
526
527\normalsize \textbf{Voice Deconstruction}
528
529As we have already established, both glottal source and vocal tract systems are involved in speech production processes. It may be expected that glottal information will be more influenced by the speaker’s phonation habits, while the description of the vocal tract will be more conditioned by the phonetic structure of the message. On its turn, the power spectral density of the glottal source is strongly conditioned by the biomechanics of the vocal folds. Thus, both vocal tract and glottal information seem to be relevant when characterizing a speaker. In order to establish the influence of vocal and glottal information in speaker recognition applications, it would be useful to break down the speech signal in a GSE and a VTE.
530
531Early implementations used to deconstruct the voice signals required frame-based pitch-synchronous processing of the glottal source by phonation cycles for the precise estimation of the voice components. However, this requirement is difficult to be met with sounds of dynamic nature not to mention when facing continuous speech. To solve this problem, the theory of inverse filtering via linear prediction, applied to Fant’s production model has been used for the reconstruction of the GSE and VTE. More specifically, predictive structures based on the Itakura–Saito Partial Correlation algorithm (PARCOR) have been conveniently modified in order to model and invert the voice production system, providing a highly efficient algorithmic structure known as paired lattice.
532 
533\textbf F\textbf i\textbf g\textbf u\textbf r\textbf e \textbf 1, provides the block diagram of the voice deconstruction algorithm used to separate the vocal tract and glottal estimates of voice from continuous speech. The proposed algorithm not only allows for a simultaneous estimation of both voice components, but also guarantees that they are orthogonal in terms of correlation. A brief description of the main blocks involved in the process is given below: 	
534
5351)Radiation Compensation Block: a first-order prediction lattice has been implemented 		to compensate lip-radiation effects.
536
5372)Inverse Filtering Block: a k-order filtering process is applied to remove the vocal tract information from the radiation-compensated speech. This process can be implemented using a prediction error lattice.
538
5393)Joint-Process Estimation Block (JPE): The residual is used as the reference signal in an Adaptive lattice-ladder filter used for joint-process estimation on the radiation-compensated speech sl(n). Through this process, a GSE and a VTE are extracted which can be considered fully uncorrelated.
540
541\textbf F\textbf i\textbf g\textbf u\textbf r\textbf e \textbf 2, depicts the VTE and GSE obtained from a female speech utterance of vowel/a/applying the described algorithm.
542
543\includegraphics[width=3.4in,height=1.8in,clip,keepaspectratio]{1}
544
545\tiny \textbf {Figure 1 | Separation algorithm using first-order prediction lattice and including a lip-radiation compensation stage.}
546
547\includegraphics[width=4in,height=2in,clip,keepaspectratio]{2}
548
549\tiny \textbf {Figure 2 | Vocal tract(middle) and Glottal source(lower) estimates for a female sustained vowel/a/utterance(upper).}
550
551
552\normalsize \textbf F\textbf i\textbf g\textbf u\textbf r\textbf e \textbf 3, represents the power spectral density of the GSE, evaluated over a temporal window which includes multiple glottal cycles. This figure clearly shows a peak and trough patterns, agreeing with previous works in the area that the glottal source do not present a flat spectrum. 
553
554\begin{IEEEbiography}[{\includegraphics[width=1.4in,height=1.45in,clip,keepaspectratio]{3}}]  The present algorithm needs to be tuned according to the kind of recordings it is supposed to deal
555with, i.e., we need to find the best parameters in the inverse filtering block that provide the best second-order decoupling in the estimated signals. Different values have been tested for the two  specific parameters used in the inverse filtering block, namely the order of the filter and the forgetting factor, which according to “helps it deal better with statistical variations when operating in non-stationary environments.” As the algorithm needs to be tuned accordingly to the kind of recordings, this means that no universal configuration can be provided, instead, when applied, it is necessary to define a development set which collects the variability of data under test and which helps in tuning the meta-parameters of the algorithms used. In other words, the selection of these parameters is done by carrying out a deep search from a number of pre-selected parameters and selecting those for which the recognition system provides better results. There have been previous attempts to use the glottal source for speaker recognition purposes; however, these approaches differ from our approach not only in the separation algorithm but also in the features derived from the source and tract estimations. According to their results, this modification provides better estimation of the first formants, especially the first one, thus decreasing the amount of formant ripple in the estimated 
556glottal flow. However, this improvement is more relevant when applied to high pitch frequency voices and when the vocal tract can be well modeled using an all-pole envelope, which is not always possible.
557\end{IEEEbiography}
558
559\normalsize \textbf{Speaker Recognition}
560
561{From a practical point of view, progress in speaker characterization has a direct effect in speaker recognition systems. Indeed, an adequate way of verifying that an improvement in characterizing a speaker is achieved is by obtaining an improvement in the recognition rates of the speaker recognition system in which this characterization is used. The following sections are devoted to present the speaker recognition system implemented (including not only the front-end but the classification method), the different scenarios in which the system is evaluated, i.e., the different databases used in the experiments, and finally the metrics used to evaluate the performance of the system.}
562
563
564\normalsize \textbf{GMM–UBM Speaker Recognition System}
565
566{Although new modeling strategies have been proposed in recent years in order to improve recognition rates, the Gaussian Mixture Model (GMM) – Universal Background Model (UBM) probabilistic paradigm strategy is still considered the \textit{de facto} reference method in text-independent speaker recognition when the available amount of information for training purposes is limited. \textbf F\textbf i\textbf g\textbf u\textbf r\textbf e \textbf 4, provides a block diagram of the speaker recognition system implemented applying the GMM–UBM approach. In this section, we do not care about the feature extraction process which has been already presented. 
567
568In the set of experiments that has been carried out, we have used a standard mixture classifier with diagonal covariance matrix. Each speaker is represented by a GMM, \begin{eqnarray}\lambda(speaker-k),\end{eqnarray} which has been adapted from a gender-dependent UBM using the MAP algorithm in which only the distribution means have been adapted (part B in Figure 4). The UBM is also represented as a GMM, 
569\begin{eqnarray}\lambda(UBM),\end{eqnarray} which has been trained from the training set via the expectation maximization (EM) algorithm (part A in Figure 4). The number of Gaussians, as well as the relevance factor used in the MAP algorithm, depends on the specific experiment carried out. 
570
571The log-likelihood ratio (LLR) has been the score used to take a decision on whether a test audio-segment is likely to be spoken by a specific speaker, with claimed identity \begin{eqnarray}\lambda(speaker-i).\end{eqnarray}
572In other words, the set of feature vectors extracted from a test audio-segment is compared with the claimed speaker model giving a match score which measures their similarity (part C in Figure 4).}
573
574\begin{eqnarray}LLR(X,\lambda_ speaker-i) = \log P(X | \lambda_ UBM) \end{eqnarray}
575
576Additionally, the decision scores can be normalized using zero normalization and test normalization or by combining both of them. In the case of ZNorm, the normalized score is given by: 
577\begin{eqnarray}S_ZNorm = \frac {\log(P(X | \lambda_speaker-i)) - \mu_Z} {\sigma_Z} \end{eqnarray} where \begin{eqnarray} \mu_z\end{eqnarray} and \begin{eqnarray}\sigma_z \end{eqnarray}are the mean and standard deviation for the impostor distribution. To estimate these values, the target speaker model is tested against utterances from impostors; this results in a set of likelihood scores from which the impostor distribution is estimated. 
578
579Finally, a decision is made to either accept or reject the claimant according to the match scores, and a specific threshold (part D in Figure 4). 
580
581\includegraphics[width=4in,height=2in,clip,keepaspectratio]{4}
582
583\tiny \textbf {Figure 4 | Separation algorithm using first-order prediction lattice and including a lip-radiation compensation stage.}
584
585\normalsize \textbf{Test framework}
586
587{The main objective of the speaker recognition is to deter- mine whether a specific target is present
588or not in a given speech segment. We have defined two different scenarios which involve the use of two different databases: ALBAYZIN and MOBIO databases. The ALBAYZIN database allows us to define highly controlled acoustic scenario without channel variability,with gender and age variability. The use of the MOBIO speech corpus constitutes and additional challenge as it contains text- independent recordings acquired in non-controlled mobile environments. Additionally, as ALBAYZIN is a Spanish database and MOBIO is an English recorded database, we are testing our system against language variability.
589
590In each of the two proposed scenarios, the databases are split into three different subsets as follows:
591
5921) Background training set: used to learn the background parameters of the algorithm (UBM,subspaces, etc.) or for normalization purposes.
593
5942) Development set: the data assigned to this set are split into two subsets: enrollment and test. The first one is used to create a model of each of the target speakers. The second one contains a list of audio samples that must be tested against all the target speakers. The data on this set are supposed to be used to tune meta-parameters of the algorithm.
595
5963) Evaluation set: the final evaluation performance is analyzed using this set, which has a similar structure as the development set. A score must be provided for each trial, for instance
597in the form of log-likelihood, representing how accurately the test segment is classified as
598containing, or not, speech for the target speaker against which is confronted.}
599
600\normalsize \textbf{Metrics}
601
602{In order to evaluate the performance of the systems, regardless the scenario, we will use the EER quality measure, and the Half Total Error Rate (HTER) which can be defined from a score threshold, \begin{eqnarray} \theta_dev,
603\end{eqnarray} obtained from the development set as follows: 
604
605\begin{eqnarray} \theta_dev = |FAR_dev (\theta) - FRR_dev(\theta)|
606\end{eqnarray} where FAR is the false acceptance rate and FRR is the false rejection rate. This score threshold, \begin{eqnarray} \theta_dev, \end{eqnarray} provides the EER operating point for the system on development:
607\begin{eqnarray} EER = \frac{FAR_dev(\theta_dev) + FRR_dev(\theta_dev)}{2}\end{eqnarray}
608
609This threshold is then used on the evaluation data set to obtain the HTER that can be defined as:
610
611\begin{eqnarray} HTER = \frac{FAR_eval(\theta_dev) + FRR_eval(\theta_dev)}{2}
612\end{eqnarray}
613
614However, as no cross-gender trials are going to be present in the tests, we can define a new metric which we call Half Equal Error Rate(HEER):
615
616\begin{eqnarray} HEER = \frac{EER_M + EER_F}{2}
617\end{eqnarray}
618
619As it must be noted, we distinguish between EER for male (EERm) and female (EERf) speakers. Obviously, when using a gender-independent parametrization, it is necessary to reach a compromise between both EER in order to minimize HEER. However, when using a gender-dependent parametrization, this compromise disappears and the objective is to minimize EERm and EERf independently. It is worth nothing that the threshold used will be different depending on the gender, since no normalization has been carried out on the scores in order to obtain a universal threshold.}
620
621\normalsize \textbf{Results, Hypothesis Testing}
622
623{\textbf F\textbf i\textbf g\textbf u\textbf r\textbf e \textbf 5, shows the ultimate configurations chosen for each gender, as well as the recognition rates
624obtained in each case in terms of EERm, EERf, and HEER. It must be noted that 40 \% of RR of EERm is achieved for male speakers, while 30 \% is achieved for female speakers.
625
626\includegraphics[width=4in,height=2in,clip,keepaspectratio]{5}
627
628\tiny \textbf {\textbf F\textbf i\textbf g\textbf u\textbf r\textbf e \textbf 5 | EERm, EERr, and HEER obtained on development set, comparing classical parameters and extended biometric parameters for the configurations providing the most successful results [RR: Relative Reduction]/[threshold](p-value)].}
629
630{DET curves corresponding to the results presented in Table 3 are depicted in \textbf F\textbf i\textbf g\textbf u\textbf r\textbf e \textbf 6 for both
631male and female speakers.}
632
633\includegraphics[width=3.3in,height=2in,clip,keepaspectratio]{6}
634
635\tiny \textbf {Figure 6 | Separation algorithm using first-order prediction lattice and including a lip-radiation compensation stage.}
636}
637
638
639
640
641
642
643
644
645
646% needemped in second column of first page if using \IEEEpubid
647%\IEEEpubidadjcol
648
649% An example of a floating figure using the graphicx package.
650% Note that \label must occur AFTER (or within) \caption.
651% For figures, \caption should occur after the \includegraphics.
652% Note that IEEEtran v1.7 and later has special internal code that
653% is designed to preserve the operation of \label within \caption
654% even when the captionsoff option is in effect. However, because
655% of issues like this, it may be the safest practice to put all your
656% \label just after \caption rather than within \caption{}.
657%
658% Reminder: the "draftcls" or "draftclsnofoot", not "draft", class
659% option should be used if it is desired that the figures are to be
660% displayed while in draft mode.
661%
662%\begin{figure}[!t]
663%\centering
664%\includegraphics[width=2.5in]{myfigure}
665% where an .eps filename suffix will be assumed under latex, 
666% and a .pdf suffix will be assumed for pdflatex; or what has been declared
667% via \DeclareGraphicsExtensions.
668%\caption{Simulation Results}
669%\label{fig_sim}
670%\end{figure}
671
672% Note that IEEE typically puts floats only at the top, even when this
673% results in a large percentage of a column being occupied by floats.
674
675
676% An example of a double column floating figure using two subfigures.
677% (The subfig.sty package must be loaded for this to work.)
678% The subfigure \label commands are set within each subfloat command, the
679% \label for the overall figure must come after \caption.
680% \hfil must be used as a separator to get equal spacing.
681% The subfigure.sty package works much the same way, except \subfigure is
682% used instead of \subfloat.
683%
684%\begin{figure*}[!t]
685%\centerline{\subfloat[Case I]\includegraphics[width=2.5in]{subfigcase1}%
686%\label{fig_first_case}}
687%\hfil
688%\subfloat[Case II]{\includegraphics[width=2.5in]{subfigcase2}%
689%\label{fig_second_case}}}
690%\caption{Simulation results}
691%\label{fig_sim}
692%\end{figure*}
693%
694% Note that often IEEE papers with subfigures do not employ subfigure
695% captions (using the optional argument to \subfloat), but instead will
696% reference/describe all of them (a), (b), etc., within the main caption.
697
698
699% An example of a floating table. Note that, for IEEE style tables, the 
700% \caption command should come BEFORE the table. Table text will default to
701% \footnotesize as IEEE normally uses this smaller font for tables.
702% The \label must come after \caption as always.
703%
704%\begin{table}[!t]
705%% increase table row spacing, adjust to taste
706%\renewcommand{\arraystretch}{1.3}
707% if using array.sty, it might be a good idea to tweak the value of
708% \extrarowheight as needed to properly center the text within the cells
709%\caption{An Example of a Table}
710%\label{table_example}
711%\centering
712%% Some packages, such as MDW tools, offer better commands for making tables
713%% than the plain LaTeX2e tabular which is used here.
714%\begin{tabular}{|c||c|}
715%\hline
716%One & Two\\
717%\hline
718%Three & Four\\
719%\hline
720%\end{tabular}
721%\end{table}
722
723
724% Note that IEEE does not put floats in the very first column - or typically
725% anywhere on the first page for that matter. Also, in-text middle ("here")
726% positioning is not used. Most IEEE journals use top floats exclusively.
727% Note that, LaTeX2e, unlike IEEE journals, places footnotes above bottom
728% floats. This can be corrected via the \fnbelowfloat command of the
729% stfloats package.
730
731
732
733
734
735\normalsize \textbf{Conclusion}
736
737{The main purpose and benefit of a voice recognition system is the amount of security that it provides. Although voice recognition is mostly secure, it still has flaws. To aid its acceptance, this biometric system can be combined with more traditional security features to provide an additional layer of security. These can include using other biometrics, or security mechanisms such as RSA, PINs or a combination of several different mechanisms. Through further development, voice recognition can be one of the most successful and largest applications of biometrics in the future to come. 
738
739Speaker recognition system was perfectly designed and implemented, and then it has been tested for unknown users. Simply the system extracts the features of the current user and compares it with the features already stored in database. According to the given deviation values the user will either be rejected or permitted.}
740
741
742
743
744
745
746
747
748% if have a single appendix:
749%\appendix[Proof of the Zonklar Equations]
750% or
751%\appendix  % for no appendix heading
752% do not use \section anymore after \appendix, only \section*
753% is possibly needed
754
755% use appendices with more than one appendix
756% then use \section to start each appendix
757% you must declare a \section before using any
758% \subsection or using \label (\appendices by itself
759% starts a section numbered zero.)
760%
761
762
763
764% use section* for acknowledgement
765
766
767
768% Can use something like this to put references on a page
769% by themselves when using endfloat and the captionsoff option.
770
771
772
773
774% trigger a \newpage just before the given reference
775% number - used to balance the columns on the last page
776% adjust value as needed - may need to be readjusted if
777% the document is modified later
778%\IEEEtriggeratref{8}
779% The "triggered" command can be changed if desired:
780%\IEEEtriggercmd{\enlargethispage{-5in}}
781
782% references section
783
784% can use a bibliography generated by BibTeX as a .bbl file
785% BibTeX documentation can be easily obtained at:
786% http://www.ctan.org/tex-archive/biblio/bibtex/contrib/doc/
787% The IEEEtran BibTeX style support page is at:
788% http://www.michaelshell.org/tex/ieeetran/bibtex/
789%\bibliographystyle{IEEEtran}
790% argument is your BibTeX string definitions and bibliography database(s)
791%\bibliography{IEEEabrv,../bib/paper}
792%
793% <OR> manually copy in the resultant .bbl file
794% set second argument of \begin to the number of references
795% (used to reserve space for the reference number labels box)
796\begin{thebibliography}{1}
797\bibitem{IEEEhowto:kopka}
798A.G.~Adami, R.~Michaescu, D.A.~Reynolds and J.J.~Godfrey.
799\relax Hong Kong, China: Modeling prosodic dynamics for speaker recognition, 2003.
800
801
802\bibitem{IEEEhowto:kopka}
803W.~Campbell, J.P.~Campbell, D.A.~Reynolds, `D.A.~Jones and T.R.~Leek.
804\relax High-level speaker verification with support vector machines, 2004.
805
806\bibitem{IEEEhowto:kopka}
807\relax Acoustics, Speech, and Signal Processing, 2004.
808
809\bibitem{IEEEhowto:kopka}
810B.~Sohal, S.~Kaur.
811\relax A HMM Integrated SVM Model for Hindi Speech Recognition, Journal of Science and Technology, 2016.
812
813\bibitem{IEEEhowto:kopka}
814L.M.~Fernandez, A.A.~Marquina.
815\relax Research: New methodology for the integration of biometric features in speaker recognition systems applied to security environments, 2014.
816
817\bibitem{IEEEhowto:kopka}
818National Science and Technology Council(NSTC).
819\relax Research: Speaker Recognition, 2006.
820
821\bibitem{IEEEhowto:kopka}
822N.~Singh.
823\relax Research: A study on speech and speaker recognition technology and its challenges, 2014.
824
825\bibitem{IEEEhowto:kopka}
826D.~Subhadeep, B.~Sujit, R.K.~Bhukya, R.K.~Das, B.C.~Haris.
827\relax Research: Speech biometric based attendance system, 2014.
828
829\bibitem{IEEEhowto:kopka}
830D.~Herath.
831\relax Research: Automated Speech Recognition For Children
832With Cleft Lip And Palate Issues, 2017.
833
834\bibitem{IEEEhowto:kopka}
835M.F.~Abdollah, M.N.~Kamarudin, Z.~Kamis.
836\relax Research: Biometric Voice Recognition in Security System, 2014.
837
838
839
840
841\end{thebibliography}
842
843
844
845% biography section
846% 
847% If you have an EPS/PDF photo (graphicx package needed) extra braces are
848% needed around the contents of the optional argument to biography to prevent
849% the LaTeX parser from getting confused when it sees the complicated
850% \includegraphics command within an optional argument. (You could create
851% your own custom macro containing the \includegraphics command to make things
852% simpler here.)
853%\begin{biography}[{\includegraphics[width=1in,height=1.25in,clip,keepaspectratio]{mshell}}]{Michael Shell}
854% or if you just want to reserve a space for a photo:
855
856
857
858% You can push biographies down or up by placing
859% a \vfill before or after them. The appropriate
860% use of \vfill depends on what kind of text is
861% on the last page and whether or not the columns
862% are being equalized.
863
864%\vfill
865
866% Can be used to pull up biographies so that the bottom of the last one
867% is flush with the other column.
868%\enlargethispage{-5in}
869
870
871
872% that's all folks
873\end{document}