|Article title||DEVELOPMENT OF THE METHOD OF FEATURE VECTORS CHARACTERISTICS IN SPEAKER RECOGNITION PROCESS|
|Section||SECTION I. INFORMATION SECURITY|
|Month, Year||07, 2015 @en|
|Abstract||This paper presents the method allowing use the reconstructed components of state vectors for the identifying users of highly reliable computing systems by coefficients for systems of biometric authentication. The existing identifiable characteristics extraction from the speech signals methods such as method of identifiable feature extraction from speech signals: mel-frequency cepstral coefficient method, a method the cepstral coefficients of a linear prediction are described. The objective of this paper is to present the method allowing to use the reconstructed components of state vectors as the identifying coefficients for systems of biometric authentication of users of highly reliable computing systems. Shortcomings of authentication systems applying spectral characteristics in time-frequency localization (area) are emphasized. The quality of these systems is in direct relationship to speech sample volume of the estimated selection, external noise has a very significant effect on these systems. To solve this problem we developed a method that takes into account the non-stationary and nonlinear dynamics of inner speech production apparatus based on the components of the state of the reconstructed model of the speech process. The numerical experiment was conducted and its results are presented with use of the described model. The object of the research is the speech signals; the samples of words and phonemes from pronunciation database recorded by one speaker were used as an experimental material. For the solution of the task next lines of the experiment were determined: experimental evidence of coefficient consistency obtained from reconstructed state –vector component for [o] and [i] phonemes; experimental evidence of possible use of state vectors obtained from reconstruction coefficient. Numerical experiment evidences were summarized: the obtained state vectors represents individual characteristics of a speaker; near live environment testing (experimenting) indicates the possibility of further practical use; no-need for spectral characteristics application in time frequency localization for identifying feature extraction allows to decrease some negative impact of the environment and avoid pre filtration requirement. Some suggestions for further use of obtained state vectors for authentification systems: clusterization to find average state vector, with the use k-average method, affinity measure identification rectangular metric.|
|Keywords||Speaker recognition; voice identification; personality state vector; mfcc; mel frequiency cepstral coefficients; lpcc linear prediction cepstral coefficients; components of vector-state model of the speech process.|
|References||1. Ivanov A.I. Neyrosetevye tekhnologii biometricheskoy autentifikatsii pol'zovateley otkrytykh sistem: dis. … d-ra tekhn. nauk [Neural network technology biometric user authentication for open systems. Dr. eng. sc. diss.]. Moscow, 2002, 383 p.
2. Klein D.V. «Foiling the cracker»: A survey of and improvements to, password security, UNIX Security II: USENIX Workshop Proceedings. Berkeley, 1990, pp. 5-14.
3. Afanas'ev A.A. Autentifikatsiya. Teoriya i praktika obespecheniya bezopasnogo dostupa k informatsionnym resursam: Uchebnoe posobie dlya vuzov [Authentication. Theory and practice of secure access to information resources: the textbook for high schools]. Moscow: Goryachaya liniya-Telekom, 2009, Ch. 11.
4. Kharkevich A.A. Spektry i analiz [Spectra and analysis]. Moscow: Gosudarstvennoe izd-vo tekhniko-teoreticheskoy literatury, 1953, 213 p.
5. Zakovryashin A.S., Malinin P.V., Lependin A.A. Primenenie raspredeleniy mel-chastotnykh kepstral'nykh koeffitsientov dlya golosovoy identifikatsii lichnosti [The use of the distributions of Mel-frequency cepstral coefficients for voice identification], Upravlenie, vychislitel'naya tekhnika i informatika [Management, Computer Sciences and Informatics], 2007, No. 5, pp. 156-160.
6. Repalov C.A. Razrabotka matematicheskikh modeley i robastnykh algoritmov identifikatsii diktorov po ikh rechi: dis. … kand. fiz.-mat. nauk [Development of mathematical models and robust algorithms to identify speakers by their speech. Cand. phys. and math. sci. diss.]. Rostov-on-Don, 2003, 146 p.
7. Ermilov A.V. Metody, algoritmy i programmy resheniya zadach identifikatsii yazyka i diktora: dis. … kand. fiz.-mat. nauk [Methods, algorithms and programs for solving tasks of identification of language and speaker. Cand. phys. and math. sci. diss.]. Moscow, 2014, 135 p.
8. Sorokin B.N. Teoriya recheobrazovaniya [The theory of speech formation]. Moscow: Radio i svyaz', 1985, 312 p.
9. Turchak L.I., Plotnikov P.V. Osnovy chislennykh metodov: uchebnoe posobie [Fundamentals of numerical methods: a tutorial]. 2nd ed. Moscow: Fizmatlit, 2003, 304 p.
10. Yakushev D.V. Uvelichenie dlitel'nosti vremennogo okna rekonstruktsii matematicheskoy modeli khaoticheskoy dinamicheskoy sistemy s pomoshch'yu drobno-ratsional'nogo interpolirovaniya [Increasing the duration of the time window of the reconstruction of mathematical models of chaotic dynamical systems via rational interpolation], Sovremennye metody i
sredstva obrabotki prostranstvenno-vremennykh signalov: Sbornik statey 6-y Vserossiyskoy nauchno tekhnicheskoy konferentsii [Modern methods and means of the processing of spatiotemporal signals: collected papers of the 6th all-Russian scientific technical conference]. Penza, 2008, pp. 54.
11. Rabiner L., Juang B.-H. Fundamentals of speech recognition: Prentice-Hall, Inc., 1993, 507 p.
12. Kopytov V.V., Yakushev D.V., Ivanov I.I., Semenyak I.A. Metody otsenki parametra kubicheskogo otobrazheniya po khaoticheskim vremennym ryadam s shumom [Methods of assessment of parameter cubic map from chaotic time series with noise], Vestnik SevKavGTI: Sb. nauchnykh trudov prepodavateley i aspirantov [Bulletin of Sevkavgtu": Collection of scientific works of teachers and graduate students]. Stavropol': SevKavGSh, 2012, Issue XIII, 224 p.
13. Anishchenko B.C., Astakhov V.V., Vadivasova T.E., Neyman A.B., Strelkova G.I. Nelineynye effekty v khaoticheskikh i stokhasticheskikh sistemakh [Nonlinear effects in chaotic and stochastic systems]. Moscow-Izhevsk: Institut komp'yuternykh issledovaniy, 2003, 544 p.
14. Takens F. Detecting Strange Attractors in Turbulence, in Dynamical Systems and Turbulence: Lecture Notes in Mathematics, Vol. 898 [ed. by D. Rang and L.S. Young]. Springer, Berlin, Heidelberg, 1980, 366 p.
15. Bezruchko B.P., Smirnov D.A. Matematicheskoe modelirovanie i khaoticheskie vremennye ryady [Mathematical modeling and chaotic time series]. Saratov: GosUNTs «Kolledzh», 2005, 320 p.
16. Malinetskiy G.G., Potapov A.B. Sovremennye problemy nelineynoy dinamiki [Modern problems of nonlinear dynamics]. 2nd ed. Moscow: Editorial URSS, 2002, 360 p.
17. Fonem [The phoneme]. Available at: http://ru.wikipedia.org/wiki/Fonema (accessed: 15 October 2015).
18. Bortovye informatsionnye sistemy: kurs lektsiy [On-Board information systems: lectures]. Available at: http://window.edu.ru/catalog/pdf2txt/082/59082/29039 (accessed: 15 October 2015).
19. Kopytov V.V., Yakushev D.V. Metody kodirovaniya rechevykh signalov s pomoshch'yu rekonstruirovannoy modeli rechevogo protsessa [Methods of coding of speech signals with the help of the reconstructed model of speech process], Izvestiya YuFU. Tekhnicheskie nauki [Izvestiya SFedU. Engineering Sciences], 2013, No. 1 (138), pp. 37-44.