Ўзбекистон республикаси ахборот технологиялари ва коммуникацияларини ривожлантириш вазирлиги муҳаммад ал-хоразмий номидаги

 
 
 
 
 
 
 
 
 
 
9-ШЎЪБА. 
 
ТЕЛЕ ВА КИНО 
МАҲСУЛОТЛАРИНИ ИШЛАБ 
ЧИҚАРИШНИНГ ЗАМОНАВИЙ 
МУАММОЛАРИ 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

---

247 
PERSONAL IDENTIFICATION BASED ON THE ANALYSIS OF SPEECH SIGNALS 
 
N.Mirzaev (TUIT, docent), 
Z.Norova (TUIT, student) 
 
Currently actively developing biometric technology - the methods and technical means of 
obtaining and using biometric data of a person in order to identify him [1]. The purpose of these 
technologies is the development of automated systems for human identification based on biometric 
characteristics: fingerprints, characteristics of voice and speech pattern of the iris, face images. 
This is due to the fact that at the moment the main obstacle for further development of information 
environments, various virtual services, etc. is the problem of reliable identification (verification) 
of the user. It is expected that the use of such systems will significantly reduce the number of 
crimes associated with unauthorized access, including computer networks, the technology of 
human identification based on voice and speech (voice signals) does not require physical contact 
with the device, unobtrusive, natural, and potentially could have sufficient reliability and speed. 
Task identification based on speech signals are divided into three groups: search the speech signal 
in large databases, control of access to and control of identity of source voice signal. They differ 
according to the requirements provided by the recognition systems, and solutions, and thus 
represent separate classes. Different requirements for errors first and second kind for these classes 
[1, 2]. The error of the first kind is a situation where the object of the specified class is not 
recognized (ignored) by the system. The second type of error occurs when the object of a given 
class is the object of another class. It should also be noted the difference between the concepts of 
verification and recognition (identification). In the task of verification of an unknown object 
declares that it belongs to some known class. The system confirms or refutes this statement. During 
the recognition (identification) is required to classify an object of unknown class to one of the 
known or to issue a conclusion about what this object does not belong to the known classes. All 
problems identification based on the analysis of speech signals is Central to the choice of the 
speech parameters, to ensure effective disclosure of the features of the speech signal for each of 
the investigated person. Many informative speech settings, allowing most effectively to reflect the 
characteristics of voice of each person: mean value of fundamental frequency, short-term spectrum 
of the speech signal, change in the intensity of the speech signal in time, the formant characteristics 
of the speech signal, etc. 
Description of the speech signal in this work is seen as a problem of approximation using 
a set of n numbers. Each of these numbers is determined in General by the following statement: 
 

T w
i
m
i
m
dwdt
t
w
S
t
w
Tw
X
0 0
,
)
,
(
)
,
(
1

where orthogonal function decomposition with the number of short - term spectrum of the speech 
signal; the maximum frequency of the spectrum; T - duration of the speech signal. The specific 
form of the operator (1) depends on the method of its implementation. The process of 
reconciliation speech delivered to the individual is reduced to the comparison of the obtained 
values with corresponding reference values.