JP2000181490A - Method and device for user collation - Google Patents

Abstract

PROBLEM TO BE SOLVED: To enable only a registered regular user to use a device. SOLUTION: An identifier input part 1 receives the individual identifier that a user shows, reads in a symbol sequence corresponding to each registered individual identifier in a storage part 6 and sends the symbol sequence to an index input part 2. The index input part 2 selects the voice parameter corresponding to the index that the user has shown this time from the symbol sequence received from the identifier input part 1 and passes it to a collation part 4. The user performs vocal input corresponding to the index that the user has shown. A vocal input part 3 finds a voice parameter from the voice of the user which has been inputted this time and sends the voice parameter to the collation part 4. The collation part 4 calculates the similarity between the voice parameter of the current input voice of the user found by the voice input part 3 and the voice parameter which has been passed from the index input part 2 and registered, and a decision part 5 judges the authenticity of the user.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for confirming a user in an apparatus for limiting the user to a certain person, and using a voice parameter registered in advance to determine whether the user is a registered person. The present invention relates to a user collation method and a user collation device for determining whether or not there is a sound at the time of operation.

[0002]

2. Description of the Related Art Conventionally, a secret symbol string (password) for user verification has been widely used as a method of determining the validity of a user of an apparatus for limiting users, but a more reliable method has been used. There is a method in which judgment based on a user's voice is used together. However, if the judgment based on voice is based only on the identity of the voice, another person secretly records the voice of the authorized user (the person) with a tape recorder, and uses the reproduced sound to
It was possible to break through the collation system.

[0003]

Therefore, the reliability of the collation system is significantly reduced. According to the present invention, in order to increase the reliability of matching, by registering voices of a plurality of symbol strings, assigning an index to each symbol string, and inputting the index and the symbol string corresponding to the index at the time of matching. And to improve the reliability of the collation system.

[0004]

In order to solve the above-mentioned problems, contents corresponding to at least one of registered items which is a symbol string accompanied by an index and a personal identifier registered in advance for each user are referred to. First to specify the reference method for
And the second information to be designated based on the first information are input by the user,
By comparing the second information with the third information specified corresponding to the registration item based on the first information, whether the input of the second information is performed by a valid user Means for judging whether or not it is used is used. Hereinafter, this means will be specifically described.

FIG. 1 is a diagram showing the principle of the present invention. The identifier input unit 1 receives a personal identifier of a user.
The index input unit 2 receives an index of a registered symbol string input by a user. Voice input unit 3
Is for receiving a user's voice input and for converting a voice input by the user into a voice parameter used by the matching unit 4 described later. The matching unit 4 calculates the similarity between the input speech parameter and the speech parameter of the selected registered symbol string or the speaker model. The determination unit 5 determines whether or not the input voice is the identity of the personal identifier, and outputs a verification result. The storage unit 6 stores templates and the like.

First, when a personal identifier is indicated by a user, the identifier input unit 1 receives the personal identifier, and also stores a voice parameter (template) of a symbol string corresponding to each identifier registered in advance in the storage unit 6 or a template. Speaker model (e.g., HiddenMarkov Mode
From 1), the one corresponding to the identifier input this time is read and sent to the index input unit 2. For example, the storage unit 6 in FIG. 1 stores five voice parameters “Aomori, Tokyo, Osaka, Shimane, Miyazaki” for the identifier input this time as “1, 2, 3, 4, 5”. Is registered as a template together with each index (index of a registered symbol string).

[0007] Subsequently, when the user indicates any (or a plurality of) indexes of the registered symbol string, the index input unit 2 displays the index from the symbol string received from the identifier input unit 1 from the user this time. The voice parameter or the speaker model corresponding to the index thus selected is selected and passed to the matching unit 4.

[0008] Finally, the user performs a voice input corresponding to the index indicated by the user from the voice input unit 3 and performs digital conversion of the voice data. The voice input unit 3 obtains voice parameters from the user's voice data after the digital conversion, and sends them to the matching unit 4. The matching unit 4 determines the similarity between the voice parameter of the currently input voice obtained by the voice input unit 3 and the registered voice parameter passed from the index input unit 2 or the input speech parameter. Probability (likelihood) generated from the model
Is calculated.

[0009] If the similarity or likelihood calculated by the collating unit 4 exceeds a predetermined threshold, the judging unit 5 recognizes that the user is the person of the personal identifier, and if the similarity or the likelihood is lower than the predetermined threshold. If so, the user determines that the user is an impostor and outputs the result.

Although the above description has been made on the assumption that the user inputs voice, the same effect may be expected by inputting sound using some sound generating means instead of voice. For this reason, it is assumed that the voice parameters and the voice input unit include those capable of responding to these sound inputs. Similarly, the term voice, as used below, shall have a broader meaning of sound.

[0011]

DESCRIPTION OF THE PREFERRED EMBODIMENTS In the following description, since the act of voice input by a user is included, the user will be expressed as a speaker as necessary.

FIG. 2 is a diagram for explaining Embodiment 1 of the present invention. A personal identifier input unit 11, which is one element of the identifier input unit 1, receives a personal identifier of a user (a user of a system or an apparatus). A template reading unit 12A, which is also an element of the identifier input unit 1, reads a speaker template corresponding to the identifier input this time from a registered speaker template registered in the storage unit 6 in advance, and stores a template arrangement described later. This is sent to the changing unit 22A.

An order input unit 21A, which is an element of the index input unit 2, receives the order of input of a symbol string. The template rearranging unit 22A, which is also an element of the index input unit 2, rearranges the templates in accordance with the input order of the symbol strings.

A symbol string dividing unit 41 which is an element of the matching unit 4
A divides the voice parameters of the input voice obtained by the voice input unit 3 into symbol strings. The symbol string matching unit 42A, which is also an element of the matching unit 4, includes a symbol string dividing unit 41.
The voice parameter of the symbol string divided by A is calculated for each symbol with the template rearranged by the template rearranging unit 22A, and the overall similarity is passed to the determination unit 5 described later.

The judging section 5 judges whether or not the user is himself / herself according to a threshold value. The storage unit 6 stores a template.

Hereinafter, the present embodiment will be described using a specific example. It is assumed that each registered speaker has registered a name (text data) and a personal identification number (voice, voice parameter) at the time of registration. The registered contents are stored in the storage unit 6. Here, "Hayakawa: 6732, Yamada: 054
3, Suzuki: 7332 "is registered.

(1) First, a user inputs information “personal identifier” for specifying an individual to the identifier input section 1 to inform the system of who the user is. The form of the input may be such that the member number or name is read using a numeric keypad or a card, or may be recognized by voice recognition.

(2) The personal identifier input section 11 receives the personal identifier of the system user. In this example, "Hayakawa (or the corresponding identifier number or the like)" is input. The template reading unit 12A reads a speaker template (here, a voice parameter of a personal identification number) corresponding to the input identifier from a pre-registered registrant template, and sends it to the template reordering unit 22A (personal identifier). No need to send).

(3) Next, the user inputs the input order of the registered personal identification numbers “input order” to the index input unit 2. In this example, "4213" is input, that is, the first character of the password is the first character, the second character of the password is second, the first character of the password is third, The fourth indicates that the third character of the personal identification number is to be input.

(4) The order input section 21A receives the order of input of the symbol string. The template reordering unit 22A
The security code templates are rearranged according to the order in which the symbol strings are input. In this example, the arrangement of the characters (numbers) that was “6732” is rearranged in the order of “4213” to “2
763 ".

(5) Finally, the user utters the number exchanged according to the order in which his / her personal identification number was input. In this example, the order “4213” was entered, so “673”
The personal identification number of the character string "2" is rearranged to "2763" and voice input is performed.

(6) The voice input unit 3 calculates the voice parameters of the voice input (input voice). The symbol string division unit 41A divides the speech parameters of the input speech obtained by the speech input unit 3 into word strings (numerical strings). The symbol string collating unit 42A calculates the distance or similarity of each of the voice parameters of the symbol string divided by the symbol string dividing unit 41A and the template rearranged by the template rearranging unit 22A, and determines the value. Hand over to Part 5. The judging unit 5 regards the user as an individual if the distance or likelihood passed from the symbol string collating unit 42A in the collating unit 4 exceeds a certain criterion. As there are
Output the result.

FIG. 3 is a diagram for explaining Embodiment 2 of the present invention. A personal identifier input unit 11, which is one element of the identifier input unit 1, receives a personal identifier of a user. A number template reading unit 12B, which is also an element of the identifier input unit 1, reads a number template corresponding to the speaker of the input identifier from the number templates of the registered speakers registered in advance, and selects a number template to be described later. Hand over to section 23.

The operation rule input unit 21B, which is an element of the index input unit 2, receives an operation rule for a numeric string. Similarly, an operation unit 22B, which is an element of the index input unit 2, operates a numeric string which is a personal identification number in accordance with the input operation rule, and passes it to a numeric template selection unit 23 described later. The number template selection unit 23 selects a number template according to the calculation result of the calculation unit 22B. Reference numeral 41B denotes a numeral string dividing unit that divides the voice parameters of the input voice obtained by the voice input unit 3 into digit digits. Reference numeral 42B denotes a numeric string collating unit, which calculates the similarity of the voice parameters of the numeric string divided by 41B with the numeric string template passed from the numeric template selecting unit 23 for each numeral, and determines the overall similarity in a later-described determination unit. Pass to 5. The determination unit 5 determines whether or not the user is a person according to the threshold.

The present embodiment will be described below using a specific example. It is assumed that each registered speaker has registered a name, a password (text data) and numerical data from 0 to 10, 20, and 30 (voice, voice parameters) at the time of registration. In this example, the operation is assumed to be a combination of addition and subtraction. However, assuming that the operation result is not negative, the addition / subtraction result of the four-digit number of each digit from 0 to 9 is in the range of 0 to 36. It is going to be. If it is assumed that the calculation result is negative, it is necessary to register the negative voice. The registered contents are stored in the storage unit 6. Here, the registered contents are as follows: "Hayakawa is the voice data and the password No. 6732, Yamada is the voice data and the password No. 0543, and Suzuki is the voice data and the password No. 73.
32 "is registered.

(1) First, a user inputs a personal identifier to inform the system of who he or she is. In this example, it is assumed that Hayakawa has been input.

(2) The personal identifier input section 11 receives a personal identifier of the user. In this example, the relationship “speaker (user) = Hayakawa” is obtained from the input result. The number template reading unit 12B reads 0 to 0 corresponding to the speaker of the input identifier from a database of registered speakers registered in advance.
The numeric template and the personal identification number of 10, 20, and 30 are read and passed to the numeric template selection unit 23. Here, the password (text data) of "Hayakawa" and 0-10,2
Numeric templates (voice parameters) of 0 and 30 are read.

(3) Next, the user inputs an operation rule. In this example, it is assumed that "++-" (plus plus minus) is input.

(4) The operation rule input section 21B receives a number operation rule. In this example, "++-"
To receive. Arithmetic unit 22B calculates the personal identification number according to the input arithmetic rules, and passes the arithmetic result to numeric template selecting unit 23, which will be described later. Here, since the input operation rule is "++-", "6 + 7 + 3-2 = 14" is passed. The number template selection unit 23 selects a number template according to the calculation result. Since the calculation result is "14", "10" and "4" are arranged.

(5) Finally, the user calculates and utters the personal identification number in accordance with the operation rule input by the user. In this example, the password is “6732” and the input calculation rule is “++
"-", So say "14".

(6) The digit string dividing section 41B
The voice parameters of the input voice determined by the above are divided for each number. The number string collating unit 42B calculates the similarity of each of the voice parameters of the number string divided by the number string dividing unit 41B with the number string template passed from the number string template selecting unit 23 for each number, and determines the overall similarity. Hand over to Part 5. The determination unit 5 determines whether or not the user is the user according to the threshold.

FIG. 4 is a diagram for explaining Embodiment 3 of the present invention. A personal identifier input unit 11, which is one element of the identifier input unit 1, receives a personal identifier of a user. Template reading unit 1 which is also an element of identifier input unit 1
2A reads the template of the speaker corresponding to the input identifier from the template of the registered speaker registered in advance, and sends the template to the template deforming unit 22C described later.
The transformation rule number input unit 21C, which is an element of the index input unit 2, receives information on transformation rules of an input symbol string from among a plurality of transformation rule candidates, and passes the information to a template transformation unit 22C described later. The template transformation unit 22C, which is also an element of the index input unit 2,
The symbol string template is transformed according to the inputted information on the transformation rule. The collation unit 4 includes a voice parameter of the input voice obtained by the voice input unit 3 and a template transformation unit 22.
The similarity with the template deformed in C is calculated, and the overall similarity is passed to the determination unit 5 described later. The determination unit 5 determines whether or not the user is the user according to the threshold.

The present embodiment will be described below using a specific example. At the time of registration, it is assumed that each registered speaker has registered a name, a password (text data), and a 0-9 numeral template (voice, voice parameters). Further, it is assumed that the user has been notified of the number (index) corresponding to the conversion rule in advance by the user having a table list of the conversion rule for the personal identification number or displaying the table by the system. Here, the contents of the conversion table are as follows: "1: add 1 to each digit of the password, 2: subtract 1 from each digit of the password, 3: remove the third digit of the password, 4: add 1 to the password. Swap the third and third digits. "
Further, it is assumed that the registered contents regarding the registrant are the same as those in the example of the second embodiment of the present invention. Although the conversion table is not shown in FIG. 4, it is assumed here that the conversion table is provided in the template conversion unit.

(1) First, a user inputs a personal identifier to inform the system of who he is. In this example, it is assumed that “Hayakawa (6732)” has been input.

(2) The personal identifier input section 11 receives the personal identifier of the system user. In this example, “speaker = Hayakawa” from the input. The template reading unit 12A
The speaker's password (text data) and template (speech parameter) corresponding to the input identifier are read from the database of the registered speaker's password and template registered in advance, and sent to the index input unit 2.

(3) Next, the user selects a conversion rule from the conversion table and inputs its number (index). In this example, it is assumed that the user inputs "No. 1 is subtracted from each digit of the password".

(4) The transformation rule number input section 21C receives information on transformation rules of an input symbol string from among a plurality of transformation rule candidates, and passes the information to a template transformation section 22C described later. In this example, it is assumed that the transformation rule “2” has been received. The template deforming unit 22C also deforms the numeric string template by referring to the conversion table on the system side in accordance with the information on the input deformation rule. In this example, since 1 is subtracted from each digit of the password “6732”, “562”
1 ".

(5) Finally, since the user has already selected and input "2" from the transformation rule table, he pronounces (5621) one less than each digit of the personal identification number (voice input), and outputs a voice. Enter 3

(6) The collating unit 4 includes the voice parameter of the input voice obtained by the voice input unit 3 and the template deforming unit 22
The similarity with the template deformed in C is calculated, and the overall similarity is passed to the determination unit 5. The determination unit 5 determines whether or not the user is the user according to the threshold.

FIG. 5 is a diagram for explaining Embodiment 4 of the present invention. A personal identifier input unit 11, which is one element of the identifier input unit 1, receives a personal identifier of a user. Template reading unit 1 which is also an element of identifier input unit 1
2A reads the template of the speaker corresponding to the input identifier from the template of the registered speaker registered in advance, and sends it to the index input unit 2 described later. External information monitoring unit 21D which is one element of index input unit 2
Adopts different information, such as time and place, depending on the collation time, and passes it to a template selection unit 22D described later. The template selection unit 22D, which is also an element of the index input unit 2, selects a symbol string template according to the external information sent from the external information monitoring unit 21D. The voice input unit 3 calculates voice parameters of the voice of the secret symbol string that has been deformed and input according to a deformation rule determined by external information, and passes the calculated voice parameters to the collating unit 4 described below. The matching unit 4 calculates the similarity between the voice parameter passed from the voice input unit 3 and the template deformed by the template selecting unit 22D,
The overall similarity is passed to a determination unit 5 described later. The determination unit 5 determines whether or not the user is a person according to the threshold.

The present embodiment will be described below using a specific example. This embodiment assumes speaker verification on a telephone line. It is assumed that the system can be accessed from three telephone numbers (numbers are distinguished by 1, 2, and 3 respectively). It is assumed that each registered speaker has registered a name (text data) and a password (voice, voice parameter) corresponding to each telephone number at the time of registration. Here, "1: 925-8314:
Aomori, 2: 925-8315: Tokyo, 3: 9
25-8316: Osaka "has been registered.

(1) First, a user selects a telephone number for accessing a plurality of systems, makes a call, and inputs a personal identifier by voice recognition or a push phone. In this example, enter Hayakawa.

(2) The personal identifier input section 11 receives the personal identifier of the system user. In this example, “speaker = Hayakawa” is received. The template reading unit 12 </ b> A reads a template of a speaker corresponding to the input identifier from a database of a plurality of templates (corresponding to telephone numbers) of registered speakers registered in advance, and stores the template in the index input unit 2. send. In this example, the template “1: Aomori, 2: Tokyo, 3: Osaka” of “Hayakawa” is read.

(3) The external information monitoring unit 21D takes in different information such as time and place depending on the collation time, and passes it to the template selection unit 22D described later. This time on the phone line,
A password (index) corresponding to the called phone number because we are trying to determine a password depending on where we have accessed from multiple access points
“2” is sent to the template selection unit 22D described later. The template selection unit 22D selects a symbol string template according to the external information sent from the external information monitoring unit 21D. In this example, since “2” is sent among the numbers 1-3, the “Tokyo” template of “2” is selected.

(4) Finally, since the user has already called the second access point, he pronounces the password "Tokyo" corresponding to the second.

(5) The voice input unit 3 calculates a voice parameter of the input voice and passes it to a collating unit 4 described later. Collation unit 4
Calculates the similarity between the voice parameter passed from the voice input unit 3 and the template selected by the template selection unit 22D, and passes the overall similarity to the determination unit 5. Judgment unit 5
Determines whether or not the user is a person according to the threshold value.

FIG. 6 is a view for explaining Embodiment 5 of the present invention. A personal identifier input unit 11, which is one element of the identifier input unit 1, receives a personal identifier of a system user. A multiple template reading unit 12C, which is also an element of the identifier input unit 1, reads a plurality of symbol string templates of the speaker corresponding to the input identifier from the registered speaker templates registered in advance, and reads a template described later. This is passed to the selection unit 22D. The template number input unit 21E, which is an element of the index input unit 2, receives a template number and passes it to a template selection unit 22D described later. A template selection unit 22D which is also an element of the index input unit 2.
Is for receiving a template number from the template number input section 21E and selecting a template of a symbol string corresponding to the number. The matching unit 4 calculates the similarity between the voice parameter of the input voice obtained by the voice input unit 3 and the template selected by the template selection unit 22D,
The overall similarity is passed to the determination unit 5 described later. Judgment unit 5
Determines whether or not the user is a person according to the threshold value.

The present embodiment will be described below using a specific example. It is assumed that each registered speaker has registered a password (voice, voice parameter) paired with a name, a password (text data), and an index at the time of registration. Here, "Hayakawa" means "1: Aomori, 2: Tokyo, 3: Osaka, 4: Shimane, 5:
"Miyazaki" and "Yamada" for "1: Apple, 2:
It is assumed that "Banana, 3: mandarin orange, 4: strawberry, 5: meron" is registered.

(1) First, the user inputs a personal identifier to inform the system of who he or she is. In this example, it is assumed that Hayakawa has been input.

(2) The personal identifier input section 11 receives the personal identifier of the user. In this example, “speaker = Hayakawa” is received from the input. Multiple template reading unit 12C
Reads a plurality of symbol string templates of the speaker corresponding to the input identifier from a database of templates of registered speakers registered in advance, and passes the template to the template selecting unit 22D. In this example, the password column "Aomori, Tokyo, Osaka, Shimane, Miyazaki" of the registered speaker "Hayakawa" corresponds to the index "1, 2, 3, 4, 5".
Is read with

(3) Next, the user inputs a template number. In this example, it is assumed that "2" has been input.

(4) The template number input section 21E receives the template number and sends it to the template selection section 22D.
Pass to. In this example, "2" is received. Reference numeral 22 denotes a template selection unit, which receives a template number from 21;
Select the template of the symbol string corresponding to the number. In this example, the second “Tokyo” is selected from the templates 1 to 5.

(5) Finally, enter the password corresponding to the entered template number. In this example, since "2" is input, "Tokyo" is pronounced.

(6) The voice input unit 3 analyzes the voice in the same manner (format) as the template stored in the database, and passes the voice to the matching unit 4. The matching unit 4 calculates the similarity between the voice parameter of the input voice obtained by the voice input unit 3 and the template selected in 22D, and passes the overall similarity to the determination unit 5. The determination unit 5 determines whether or not the user is a person according to the threshold.

In each of the above embodiments, various inputs for pre-registration performed at the time of first use and various inputs performed at each time of use are the identifier input section, index input section, and voice input section in each of the above examples. Need not be provided independently for all of them, and a configuration commonly used is also possible. For example, it can be composed of one keyboard commonly used for inputting symbols (such as numbers) to the identifier input unit and the index input unit and one microphone for voice input to the voice input unit. If all inputs are made by voice, it is possible to configure with one microphone.

[0056]

As is apparent from the above description, according to the speaker verification method of the present invention, the input symbol string must be inputted together with the index of the symbol string, so that the input speech symbol string is heard by another person. Even if the index is not known, there is an effect that the user cannot impersonate himself and use the collation system.

[Brief description of the drawings]

FIG. 1 is a principle diagram of the present invention.

FIG. 2 is a diagram illustrating Embodiment 1 of the present invention.

FIG. 3 is a diagram illustrating a second embodiment of the present invention.

FIG. 4 is a diagram illustrating a third embodiment of the present invention.

FIG. 5 is a diagram illustrating a fourth embodiment of the present invention.

FIG. 6 is a diagram illustrating a fifth embodiment of the present invention.

[Explanation of symbols]

 1 is an identifier input unit, 2 is an index input unit, 3 is a voice input unit, 4 is a collation unit, 5 is a judgment unit, 6 is a storage unit, 11 is a personal identifier input unit, 12A is a template reading unit, and 12B is a numeric template. A reading unit, 12C is a plural template reading unit, 21A is a sequence input unit, 21B is an operation rule input unit, 21C is a transformation rule number input unit, 21D is an external information monitoring unit, 21E is a template number input unit, and 22A is a template rearrangement. , 22B is an operation unit, 22C is a template transformation unit, 22D is a template selection unit, 23 is a numeric template selection unit, 41A is a symbol string division unit, 41B is a numeric string division unit, 42A is a symbol string collation unit, and 42B is a number. Column collation part.

Claims (7)

[Claims] 1. first information for designating a reference method for referring to content corresponding to at least one of registered items which is a symbol string with an individual identifier and an index registered in advance for each user; In response to the user inputting second information to be designated based on the first information, the second information corresponds to the registration item based on the first information. And comparing the third information specified with the third information to determine whether the input of the second information is performed by a valid user. 2. The method according to claim 1, wherein the second and third information is audio information. 3. A user verification device having an identifier input unit, an index input unit, a voice input unit, a conversion unit, a storage unit, a verification unit, and a determination unit. Each user inputs his or her personal identifier from the identifier input unit and a plurality of symbol strings from the voice input unit, and the plurality of symbol strings input by the user from the voice input unit are used as the registered voice parameters by the conversion unit. The converted speech parameters and the registration index corresponding to the symbol string are pre-registered in the storage unit together with the personal identifier. In each use, the user inputs the personal identifier from the identifier input unit and one or more The registered index is specified from the index input unit, and the symbol string corresponding to the specified index is input from the voice input unit, and a user inputs the symbol string to the voice input unit. The symbol string input from is converted into a voice parameter by the conversion unit, and the converted voice parameter and the corresponding registered voice parameter are compared by the matching unit for each specified index, and the result is also obtained. Wherein the determination unit outputs a verification result indicating whether or not the user is a valid user, based on whether or not both satisfy a predetermined standard. 4. A user collation device comprising an identifier input unit, an index input unit, a voice input unit, a conversion unit, a storage unit, a rearrangement unit, a collation unit, and a determination unit, Prior to the first use, each user inputs his or her personal identifier from the identifier input unit, inputs a plurality of symbol strings from the voice input unit, and one symbol string input by the user from the voice input unit is The conversion unit converts the registered voice parameter into a registered voice parameter, and the registered voice parameter and a registration index corresponding to the symbol string are pre-registered in the storage unit together with a personal identifier. In each use, the user inputs the personal identifier into the identifier input unit. And input from the index input unit the input order of each symbol constituting the symbol string,
The symbols are input from the voice input unit according to the specified input order, and the symbols input by the user from the voice input unit are converted into voice parameters by the conversion unit, and the registered voice parameters are rearranged. Unit is rearranged according to the input order, the converted voice parameters and the registered voice parameters rearranged by the rearrangement unit are compared by the collating unit, and based on the result, both are based on a certain standard. The user collation device according to claim 1, wherein the determination unit outputs a collation result indicating whether or not the user is a valid user depending on whether the user satisfies the condition. 5. A user collation device comprising an identifier input unit, an index input unit, a voice input unit, a conversion unit, a storage unit, a calculation unit, a collation unit, and a determination unit. Prior to the use of each, each user inputs a personal identifier from the identifier input unit, inputs one numeric string from the index input unit, inputs a set of fixed numbers from the voice input unit, and The result obtained by converting the set of one number string and the certain number inputted from the input unit into the registered voice parameter by the conversion unit is registered in the storage unit in advance. While inputting the personal identifier from the identifier input unit, inputting from the voice input unit the operation rule using each number of the numeral string and the result calculated using the operation rule, input from the voice input unit The conversion result is converted to a voice parameter by the conversion unit, the registered voice parameter of the number string and the result calculated by the calculation unit using the calculation rule are compared by the matching unit, based on the result The user collation device according to claim 1, wherein the judging unit outputs a collation result as to whether the user is a legitimate user based on whether or not both satisfy a predetermined standard. 6. A user matching device comprising an identifier input unit, an index input unit, a voice input unit, a conversion unit, a storage unit, a symbol string conversion unit, a comparison unit, and a determination unit. Prior to the first use, each user inputs his / her personal identifier from the identifier input unit and one symbol string from the voice input unit, and the user inputs one symbol string from the voice input unit. The result of the conversion of the symbol string into the registered voice parameter by the conversion unit and a plurality of symbol string conversion rules distinguished by the index are registered in the storage unit in advance, and each time the user uses his or her own A personal identifier is input from the identifier input unit, and one of the symbol string conversion rules is specified from the index input unit. The symbol string converted according to the conversion rule is input from the voice input unit, and the symbol string input by the user from the voice input unit is converted into a voice parameter by the conversion unit, and the voice parameter and the user specify The registered voice parameter converted by the symbol string conversion unit in accordance with the symbol string conversion rule is compared with the converted registered voice parameter by the matching unit, and the determination unit is used depending on whether or not both satisfy a predetermined criterion. A user collation device that outputs a collation result as to whether or not a user is a valid user. 7. The user verification device according to claim 3, wherein the user uses the communication device via a communication line.