JP2002221990A - Personal authentication device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To make authentication with high accuracy by a system for authenticating individuals by speech. SOLUTION: The higher accuracy than by the prior art is obtained by making combination use of speaker characteristic recognition and recognition of utterance contents and the accuracy of the speaker characteristic recognition and the accuracy of content recognition is improved by adopting an 'incantation' as utterance contents. The systems thus has an effect of making it possible to obtain the high individual authentication accuracy not possible heretofore.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an apparatus for authenticating an individual used in a security system of a bank or a personal computer.

[0002]

2. Description of the Related Art A conventional personal authentication technique will be described below.

In recent years, with the advancement of information-related technology, the computerization of industry and life using networks, computers, and the like has been remarkable. Under such circumstances, the demand for security technology for the purpose of information security is rapidly increasing. In particular, the personal authentication technology is a cryptographic technology using a person as a key, and is the most basic and increasingly necessary security technology.

Until now, personal identification numbers, signatures, fingerprints, irises (irises) and the like have been studied and commercialized. Recently, is utilized to interface or the like of the personal computer high accuracy authentication technique for error rate of 10 ^-6 fingerprint is established is about to be commercialized, also authentication by the iris 10 ^-4
It is said that the accuracy of the error rate can be obtained, and is used in some banks. Among them, voice authentication technology has been studied for more than 10 years by NTT, Tokyo Institute of Technology, etc., but it is said that the accuracy is 95-98% even in the text-dependent type. The accuracy is extremely low.

[0005] However, voice is a natural interface used by humans for communication, and it is strongly desired that the way of use be opened because of its simplicity. However, personal authentication by voice could not be realized without achieving unprecedented performance.

[0006]

As described above, personal authentication by voice has been studied, but its accuracy is 9%.
The performance was much worse than that of fingerprints and irises, at 5 to 98%. However, voice is a natural interface used by humans for communication, and it is strongly desired that the way of use be opened because of its simplicity, and realizing unprecedented performance has been a major issue. .

[0007]

In order to solve this problem, the invention according to claim 1 comprises a speaker recognizing unit, a speech content recognizing unit, a recognition result integrating unit, and a speaker registering unit. A personal authentication device configured to perform both recognition and recognition of utterance contents and perform personal authentication by integrating the results thereof.

As a result, since the recognition result of the speaker recognition unit and the recognition result of the utterance content can be integrated, personal authentication with higher accuracy than the personal authentication technology conventionally realized only with speaker recognition is realized. It becomes possible.

The invention according to claim 2 is characterized in that the authentication language is generated irrespective of the meaning of the language based on the acoustic information and the similarity between the contents.

[0010] Thereby, first, a high speaker recognition accuracy is obtained because of the acoustic information, and a high content recognition accuracy is obtained because of the similarity between the contents. High personal authentication accuracy can be obtained.

[0011]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The invention according to claim 1 of the present invention is a personal authentication apparatus characterized by comprising a speaker recognizing section, a speech content recognizing section, a recognition result integrating section, and a speaker registering section. Further, by adopting the invention described in claim 2 of the present invention as utterance contents, there is an effect that it is possible to obtain exceptional authentication accuracy more than before. In the following description, the utterance content having the three characteristics described in claim 2 of the present invention will be temporarily referred to as a “spell” (which indicates a recognition language).

An embodiment of the present invention will be described below.

FIG. 1 is a functional block diagram of the personal authentication device according to the present embodiment.

In FIG. 1, reference numeral 100 denotes a microphone, 101 denotes voice input means, 102 denotes a keyboard, 103 denotes key input means, 104 denotes a speaker registration unit, 105 denotes a personality storage unit,
06 is a speaker recognition unit, 107 is a registration number storage unit, 108 is an utterance content recognition unit, 109 is a spell storage unit, 110 is a recognition result integration unit, 111 is an authentication result, and 112 is a switch. The personal authentication algorithm will be described below with reference to FIG.

First, a method of registering a user to be authenticated (referred to as an authenticator) will be described.

At this time, the switch 112 is set to the speaker registration means 1
From 04, the speaker recognition means 106 and the utterance content recognition means 108
Switch to lead to. This switching is performed by, for example, a human (authenticator).

The spell stored in advance in the spell storage means 109 has three features: (1) the content itself has no meaning; and (2) it contains sufficient acoustic information to identify an individual. , And also robust enough (environment, personal condition)
, The recognition accuracy of the individual is extremely high, and (3) a group of contents (a sufficiently large group of predetermined groups) is provided.

This spell has an extremely low maximum similarity between contents, and the recognition rate becomes extremely high, regardless of who utters it). Remember any spells. An example of a list of spells is shown in Figure 2 below. As shown in FIG. 2, each spell is numbered. Also, when used worldwide, it stores spells of phonemes of each country. The similarity between contents refers to the similarity between registered character strings.

Then, the authenticator registers the learned spell.
First, the spell number is input using the keyboard 102. The key input unit 103 sends the input number to the speaker registration unit 104. In addition, the certifier uses the microphone 10
Say 0. The voice input unit 101 sends the input voice signal to the speaker registration unit 104. First, the speaker registration unit 104 sends the spell number transmitted from the key input unit 103 to the utterance content recognition unit 108.

At this time, the utterance content recognition unit 108 recognizes that the mode is the registration mode from the state of the switch 112, and stores the transmitted number in the registration number storage unit 107. Next, speaker registration section 104 sends the speech signal sent from speech input section 101 to speaker recognition section 106. The speaker recognition unit 106 recognizes that the mode is the registration mode from the state of the switch 112, analyzes the sent voice signal, creates a personality pattern (speaker phoneme model) of the authenticator, and stores the personality storage unit 105. To store. The personality pattern is a parameter that represents the characteristics of an individual used for speaker identification.
For example, the speaker phoneme model of Document A below corresponds to this.

Various methods have been studied as methods for creating this personality pattern. Reference A (Matsui et al. “Text-Specific Speaker Recognition”, IEICE Transactions D-II,
Vol. J79-D-II, no. 5, pp. 647-
656, May 1996), based on a speaker hidden Markov model, using learning data of each speaker and a hidden Markov model of an unspecified speaker. A speaker hidden Markov model is trained to generate individuality parameters.

Note that the personality pattern is a parameter representing the characteristics of an individual used for speaker identification, and corresponds to, for example, the "speaker phoneme model" in the above-mentioned document.

Thus, the registration of the certifier has been completed. In addition,
The verifier does not need to remember the spell number.

Next, the function at the time of authentication will be described in detail.

At the time of authentication, the switch 112 is set in advance so that the voice input unit 101 makes the speaker recognizing unit 106 and the utterance content recognizing unit 10
9 so that it is connected.

First, a person to be authenticated (called a subject) speaks a spell toward the microphone 100. (of course,
If the subject is a certifier and wants to be recognized as a certifier, he utters the spell that he remembers. The voice input unit 101 converts the input voice into a speaker recognizing unit 106 and an utterance content recognizing unit 1.
Send to 08. The speaker recognition unit 106 recognizes that the switch is in the state at the time of authentication, analyzes the transmitted voice signal, and extracts the individuality pattern of the subject.

Then, the personality pattern is collated with the personality pattern of the certifier stored in the personality storage unit 105 to determine a degree indicating whether or not the subject is a certifier. Send to 110. The specific collation method is disclosed in the above (Document A). Further, the utterance content recognition unit 108 compares the transmitted voice signal with the pattern stored in the spell storage unit 109.

As a specific matching method, various methods such as the HMM, the model speech method, the Compats method, and the CLM method have been studied. As an example, as described in the above-mentioned literature, a speaker phonemic model is connected, a model of the text is created, a likelihood when an input voice is given to the model of the text is calculated, and the speaker and the text are calculated. Make a decision.

Further, regarding the CLM method, a study group material: Document B (Hiraoka et al., “Speech Recognition Apparatus for Unspecified Speakers / Minor Words Using Word Spotting Technique”, IEICE, Speech Workshop Group Material, SP88- 18, 1988 6
Mon).

For the model speech method, refer to the document of the study group: Document C (Miyata et al. “Unrecognized speaker speech recognition by the model speech method in units of CV, VC and VCV”, IEICE, Speech Research Group. Material, SP91-83, December 1991). Then, the matching result (“how close to which spell” and “registration number storage unit 107
To the recognition result integration unit 110).

The form of storing the spell in the spell storage section 109 has various structures depending on the collation method of the utterance content recognition section 108. For example, when a speech signal is statistically analyzed and recognized by a time-series model of parameters selected as in (Reference B), the speech signal is stored in the form of a time-series pattern representing a spell.

When a speech unit is used as a unit of recognition as in (Reference C), the data is stored in the form of text. Then, the recognition result integration unit 110 makes a comprehensive determination based on the results sent from the speaker recognition unit 106 and the utterance content recognition unit 108, and determines whether or not the subject is the authenticator as the authentication result 111. Output.

There are various methods for comprehensive judgment. As an example, as an example, the value S1 of “how close to the authenticator (similarity)” and the number stored in the registration number storage unit 107 are The value S2 of "how close the spell was to the spell of (similarity)" and the value of "similarity of the closest spell other than the number stored in the registration No. storage unit 107" are S3 and are expressed by the following (Equation 1). There is a method of determining.

[0034]

(Equation 1)

With the above configuration, high personal authentication accuracy can be obtained.

Note that the constants a, b, and t are the teacher data (whether the output is the speaker recognition unit 106 or the output of the speech content recognition unit 108 obtained by uttering various persons as subjects) whether or not they are authenticators. Is obtained by learning (finding a boundary surface (straight line) that maximizes the authentication accuracy) based on a database in which the subject is recorded as an authenticator or not. An example of the learning method will be described below.
First, teacher data is shown in the following (Equation 2).

[0037]

(Equation 2)

First, constants a and b are determined. These are determined so as to minimize the following cost function (Equation 3).

[0039]

(Equation 3)

Therefore, the constants a and b can be obtained by solving a linear equation with two variables a and b obtained by partially differentiating the cost function by a and b as simultaneous equations. . This makes it possible to determine the slope of the boundary between the set that is the authenticator and the set that is not the authenticator.

Next, learning of the constant t is performed. In the present embodiment, as an example, two target false authentication rates (the rate at which a person who is not an authenticator is mistaken for an authenticator (false error rate), It is determined by a successive approximation algorithm that reduces both while reflecting the required standard of the rate (error rejection rate) in which the user is not an authenticator. The procedure of the algorithm is shown below. (1) Set t = 0 as an initial value. (This is (Equation 2)
Depends on the setting of Tn. (2) The judgment of (Equation 1) is performed for each sample of the teacher data, and an error tolerance rate P and an error rejection rate Q are calculated for the teacher data. (3) Calculate the value R of the following (Equation 4).

[0042]

(Equation 4)

(4) Using the predetermined fixed minute amount x, two cases of t + x and t−x are described in (2) above.
The same processing as (3) is performed, and each value (R +, R−
Call). (5) Termination when R <R + and R <R−, t at that time
Use

When R> R + and R <R−, t = t +
Go to (2) as x.

When R <R + and R> R−, t = t−
Go to (2) as x.

R> R + and R> R− and R + <R−
At the time, go to (2) as t = t + x.

R> R + and R> R− and R +> R−
Then, go to (2) as t = tx.

The constants a, b, and t can be learned by the above algorithm. In addition, learning can also be performed using a neural network learning algorithm, an algorithm such as a maximum likelihood descent method, or a genetic algorithm. Further, a method of adaptively gradually changing the variation X in the above algorithm is also effective.

Here, an example of a method of creating a “spell group”, which is an important feature of the present invention, will be described. First, the definition of a spell is shown below. (1) The content itself has no meaning. (2) It contains acoustic information sufficient to identify an individual, has sufficient robustness (environment, state of the individual), and has extremely high individual recognition accuracy. (3) Content group (sufficiently large group determined in advance)
Has an extremely low maximum value of the similarity between contents, and the recognition rate becomes extremely high, no matter who utters.

Hereinafter, a method of creating a spell group will be described.
Note that the units of phonemes are various such as phonemes, semi-syllables, and syllables, and there are even more diverse ones including foreign languages. I do.

First, in order to satisfy the above feature (2),
Set the minimum kana that makes up the spell and use random numbers to
A sufficient number of spell candidate groups (referred to as “population A”) are created. Although research has been conducted on which phonemes and frequencies are particularly included in personality, a method that considers phoneme balance is used here as an example.

For example, robustness (environment, personal condition)
In consideration of the above, the five types of "vowel", "voiced consonant", "unvoiced consonant", "half vowel", and "dullary" are always included, and an arbitrary pseudonym is inserted between them.

As a method, first, the number of characters is determined by a random number (5 or more characters. Setting is made so as not to be too long in consideration of the number that can be easily remembered by ordinary people). Is determined by random numbers, and further, which kana of each type is to be used is determined by random numbers, and finally, the remaining locations of kana are extracted and determined by random numbers from the set of all kana. Is mentioned.

With this method, a population A consisting of an enormous number of elements can be created. These include acoustic information sufficient to identify an individual, and have the feature (2). At the time of creating the population A, it is effective to give common sense constraints by simple rules.

For example, "" does not come to the head ",""does not come after two", "" does not come to the end ", or""No vowels and half vowels do not come after""or""."
And so on.

Next, from the population A, (3)
Are automatically selected to create a new population (referred to as “population B”). First, an example of the standard pattern of each kana, the similarity of the kana, and the similarity between contents will be described.

First, a time series of acoustic parameters of each kana is obtained from a word group of many speakers labeled with kana (where the kana is described in time and which kana is added to the voice data). A standard pattern of each kana can be obtained by cutting out and obtaining an average and a variance from many sets of these (called kana data).

Then, the similarity obtained by matching the standard pattern of each kana with the above-mentioned many kana data is averaged (the relationship between the pattern and the data is changed and the matching is performed again to take the average of all). To determine the similarity between the pseudonyms. For example, the following table (Table 1) can be created. The above matching method is disclosed in (Reference B). ("Word matching" can be applied to "kana matching".)

[0059]

[Table 1]

Using (Table 1), the similarity between the pseudonyms can be obtained. Therefore, the similarity between contents is obtained using this table. As a matching method at this time, a method using DP (Dynamic Programming), partial spotting, or the like can be used. In the present embodiment, as the most basic example, 1
An algorithm is employed in which the total similarity of the similarities of each kana when the similarity is maximized in consideration of the displacement of the two kana.

For example, the following (FIG. 3) shows how to determine the degree of similarity between the contents of “Aiueo” and “Enkikkauio”. (Actually, the above two contents do not exist due to the characteristic (2) of the spell, but it should be noted that these contents are used to explain using (Table 1). The similarity is (Table 1) As shown in FIG. 3, in the case of the above two contents, the similarity between the contents is obtained by calculating the sum of the mutual similarities obtained using (Table 1) and has a value of "75". . In this way, the similarity between two contents can be obtained.

If the above algorithm is added with a method of calculating the similarity between contents by dividing the above value by the number of kana and normalizing the same, a similarity that does not depend on the number of kana characters can be obtained.

Further, since it is conceivable that the character strings are partially similar, a character string obtained by adding “「 ”before and after a short character string of two character strings to be compared (the number of characters of a long character string and Matching with the number of differences in the number of characters in a short character string) can obtain a more reliable similarity between contents.

For example, in the matching between "aiueo" and "gapiaiueaautoru", "aiueotsu", "aiueootsu", "hiaiueooh", and "ohaiou" What is necessary is to match "eye-eye", "key-eye-eye", and "key-eye-eye", and determine the maximum similarity as the similarity between contents. With this ingenuity, a high degree of similarity can be reliably obtained even when the areas are partially close. This concludes the description of how to determine the similarity between contents.

Therefore, a population B is created by automatically selecting each element of the population A using the above-mentioned similarity between contents. The automatic selection algorithm is described below. In the present embodiment, the algorithm is such that the population B is created by thinning out the elements of the population A to a predetermined number of elements. (0) A target value of the number of elements of the population B is determined in advance. (1) A number (order) is assigned to all elements of the population A. (2) Move n one by one from 1 to the number of elements of the population A, and perform (3). (3) Match the element of the number n with all the elements of the population A, obtain the maximum value of the similarity between the contents and the next largest value (quasi-maximum value), and calculate these as the overall similarity of the element of the number n And (4) The total similarity of all the elements of the population A is checked, and two elements having the maximum total similarity (the maximum value) are specified. (5) Of the two elements, the element with the larger semi-maximum value is deleted from the population A.

(6) When the number of elements of the population A reaches the target value specified in advance, this is set as the population B, and the processing ends. Otherwise, go to (1).

By the above algorithm, a population which is not optimal but has relatively low similarity between contents and which has the characteristic (3) of the definition of the spell can be created. There are various other selection algorithms, such as an algorithm that increases the number of elements.

Finally, the human examines the elements (spell candidates) of the population B and deletes those that cannot be uttered or those that are not suitable as keys. The following cases are possible. (1) Those that are difficult to utter. (Such as "Iahechoropybich")
Excludes those who cannot produce stable utterances. (2) Those containing a meaningful character string. Except for words that have a problematic word.

The "spell group" is completed by the above operation. This concludes the description of how to create spells.

Using the spell group, a “numbered list of spell groups” is created and prepared for registration of the certifier,
In addition, a spell pattern to be stored in the spell storage unit 109 for authentication is created.

Here, a method of creating a pattern to be stored in the spell storage unit 109 will be described. In the case of using the unrecognized speaker's voice recognition technology disclosed in Document B, a large number of young and old speakers are required to utter spells of a group of spells, and the spells are recorded and voice sections are cut out. A spell storage unit 109 is prepared by creating a standard spell pattern using the voice database and storing the spell standard pattern. This algorithm is disclosed in reference B.

When the number of spells is large, a standard pattern of each phoneme (half syllable, syllable, etc.) may be created and used from the speech database of many speakers. This algorithm is disclosed in reference C. In this case, the spell only needs to be described in text, and the spell standard pattern is created by connecting the phoneme standard patterns during recognition. Also, when using HMM or other recognition methods,
A standard pattern in a form different from the above two is stored in the spell storage unit 109.

The whole description of the authentication method has been completed.

By transmitting the authentication result 111 to the actual key as described in the above description of the authentication method and opening the key when the result indicates that the subject is an authenticator, a room or a safe using voice can be used. Security system can be constructed. Also, by using this authentication result 111 in combination with other authentication (fingerprint, iris, face recognition, signature recognition, etc.),
Furthermore, a highly accurate security system can be realized.

If the algorithm of the personal authentication apparatus of the present invention is recorded as a program in a recording medium such as an HD or a ROM, highly accurate personal authentication using voice can be realized in a wide range of applications such as a personal computer and a mobile phone. Various security systems can be built.

FIG. 4 shows an example of a wireless communication device using a personal authentication device. 201 is a microphone, 202 is a communication unit, 2
03 is an antenna, 204 is a speaker, 205 is a keyboard, 206 is a control unit, 207 personal authentication device, 208 is a hand switch, 209 is a switch, and 210 is a power supply.

An embodiment of the wireless communication apparatus according to the present invention will be described.

First, a conventional call function conventionally used for a mobile phone or the like will be described. The voice input from the microphone 201 is encoded, modulated, and high-frequency converted by the communication unit 202 and transmitted from the antenna 203. The radio wave received by the antenna 203 is subjected to low-frequency conversion, demodulation, and decoding in the communication unit 202, and the sound is output from the speaker 204.

In addition, the communication section 202 has a function of detecting a ring tone (dual tone) and outputting a melody for notifying an incoming call from the speaker 204. The keyboard 205 inputs a telephone number input or power on / off, and the control unit 206 receives the input from the keyboard 205, encodes the destination number, and sends it to the communication unit 202. All of the above functions are realized by electricity supplied from the power supply 210.

Next, the function at the time of actual use will be described.
First, the hand switch 208 and the switch 209 are initially off. Here, the hand switch 208
When is pressed, power is supplied from the power supply 210 to the microphone 201, the keyboard 205, and the personal authentication device 207, and a mode is set in which personal authentication can be performed.

The personal authentication device 207 has the configuration and function of the contents described in the first embodiment, performs personal authentication there, and sends the authentication result to the switch 209. Switch 20
9 is ON only when the speaker recognizes as a certifier,
Electricity is supplied to the microphone 201, the communication unit 202, the speaker 204, the keyboard 205, and the control unit 206.

Thus, the device can be used as a normal wireless communication device. Therefore, when the authentication result sent from the personal authentication device 207 to the switch 209 is “not an authenticator”, the switch 209 remains off and remains in a state where it cannot be used as a wireless communication device. .

The hand switch has a function of returning to the off state when the finger is released, so that wasteful electricity is not consumed. When the power switch on the keyboard 205 is turned off, the switch 209 is turned off, and will not operate unless the authenticator is authenticated next.

The registration of the authenticator is performed by using the microphone and the keyboard while turning on the hand switch. This function has been described in the first embodiment.
Also, in order to prevent unauthenticated persons from registering maliciously, by adding a function to the personal authentication device that once registered, you can not re-register unless you authenticate with the authenticator,
Further, the degree of security is improved.

The present invention has an effect that a wireless communication device which can be used only by a specific person by the personal authentication function by voice can be realized by the above embodiment.

FIG. 5 shows a security system using a personal authentication device. 301 is a microphone, 302 is a personal authentication device, 303 is a key, and 304 is a keyboard. Note that the above “key” is a key for enabling the system,
It shows keys that enable doors in rooms and safes, personal computers, bank terminals, car engines, etc., and can be applied to any of them.

Embodiments of the security system according to the present invention will be described below.

It is assumed that the key 303 is locked. First, the user speaks into the microphone 301. The personal authentication device 302 authenticates whether or not the user is an authenticator from the input voice, and outputs the authentication result to the key 303. Key 3
03 unlocks the key if the user is the authenticator.

Now, a system with a key can be used. The authentication method has been disclosed in the first embodiment. The key can be manually unlocked but cannot be unlocked, and can only be unlocked by voice authentication.

As described above, security for use of the system can be maintained. The registration of the certifier is performed in advance using the microphone 301, the keyboard 304, and the personal authentication device 302. The registration method has been described in the first embodiment. In addition, the security level can be further improved by allowing only the authenticator to re-register once registered.

It is also possible to perform the act of “locking” using a personal authentication device instead of manually. At this time, if the key is opened and closed with different utterance contents, a key that can be opened and closed only by the authenticator can be realized.

The present invention has an effect that a system in which only a specific person can use the personal authentication function by voice can be realized by the above embodiment.

[0093]

As described above, by using the features including the speaker recognition unit, the utterance content recognition unit, the recognition result integration unit, and the speaker registration unit, the recognition result of the speaker recognition unit and the recognition result of the utterance content are integrated. Therefore, it is possible to realize personal authentication with higher accuracy than the personal authentication technology which has conventionally been realized only by speaker recognition. Further, according to the feature of the uttered contents of the second aspect of the present invention, (( A higher speaker recognition accuracy can be obtained from the feature (2), and a higher content recognition accuracy can be obtained from the feature (3). By integrating them, a higher personal authentication accuracy can be obtained. According to the third feature of the present invention, highly accurate personal authentication can be realized, a secure security system using voice can be constructed, and a special effect which cannot be obtained conventionally can be obtained.

[Brief description of the drawings]

FIG. 1 is a functional block diagram of a personal authentication device according to an embodiment of the present invention.

FIG. 2 is a diagram showing an example of a spell list

FIG. 3 is a diagram showing an example of a method of calculating the similarity between contents;

FIG. 4 is a diagram showing a configuration of a wireless communication device using a personal authentication device.

FIG. 5 shows a security system using a personal authentication device.

[Explanation of symbols]

 Reference Signs List 101 voice input means (microphone) 102 keyboard 103 key input means 104 speaker registration unit 105 personality storage unit 106 speaker recognition unit 107 registration No storage unit 108 utterance content recognition unit 109 spell storage unit 110 recognition result integration unit 111 authentication result 112 switch

──────────────────────────────────────────────────続 き Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI Theme coat ゛ (Reference) H04L 9/32

Claims (3)

[Claims] 1. A speaker recognizing unit that analyzes a voice signal emitted by a speaker, extracts a personality pattern, and checks the personality pattern against a registered personality pattern of an authenticator. An utterance content recognizing unit that performs matching with a voice pattern of a registered authentication language and recognizes an uttered word; similarity as the authenticator obtained from the speaker recognizing unit and the generated content A personal authentication device, comprising: a recognition result integration unit that performs personal authentication of the speaker based on a similarity that is the registered voice pattern obtained from a recognition unit. 2. The personal authentication apparatus according to claim 1, wherein the authentication language is generated irrespective of the meaning of the language based on the acoustic information and the similarity between the contents. 3. A security system comprising the personal authentication device according to claim 1.