KR20010069650A - Method of Recognizing Vocabulary Comprised by Number and Simultaneously Identifying Speaker and System Therefor - Google Patents

Abstract

PURPOSE: A system for recognizing word comprising numbers and certifying a speaker simultaneously and a method thereof are provided to enhance the speech recognition ratio and combine voice recognition with speaker recognition. CONSTITUTION: A memory stores a number stream transmitted by a communication network and related with certification and an individual character data corresponding to the number stream. A voice recognizer containing a voice model database selects a number stream similar to the speaking characteristics of the word without regard to the voice of the speaker. A speaker recognizer compares the individual character data and the speaker character data by extracting them from the number stream stored in the memory.

Description

Method and Recognizing Vocabulary Comprised by Number and Simultaneously Identifying Speaker and System Therefor}

Field of invention

The present invention relates to a method for authenticating a speaker while simultaneously recognizing a string of numbers. More specifically, the present invention relates to a method for selecting a plurality of speech recognition target candidates from a string of numbers based on numbers, and determining the voiceprint information of the selected speech recognition target candidate to select one string string having the highest similarity. will be.

Background of the Invention

The present invention relates to a method of voice recognition of a string of numbers and through which the speaker is authenticated. Numeric speech recognition is a very useful speech recognition technique in connection with the use of numbers used in various forms such as social security numbers and account numbers.

Korean numerals, however, are all composed of single syllables (eg, zero, one, two, third, etc.), and include words with similar phonemes, such as work, seven, five, and nine. For this reason, it acted as a cause of lowering the accuracy of recognition in speech recognition of numeric strings. Therefore, in the technology related to the current speech recognition, the practical use of the strings is delayed despite various usefulness.

Also, even if one number is likely to be recognized, the probability that all consecutive numbers are correctly recognized is very low. For example, assuming that the recognition rate for one number is 95%, the probability of accurately recognizing ten consecutive strings is only about 60%. As the number increases, the recognition probability becomes lower.

The speaker authentication method of the present invention may be used to recognize an account number, social security number, password, etc. in a phone-banking service, or may be used for entering a social security number or password when a civil service of a public institution is processed using an automatic response service. have.

These fields of application all require the ability of user authentication to verify user identity. With current technology, user authentication can be performed with more than 98% accuracy in the telephone network by using a speaker identification algorithm using voice.

The present invention proposes a method and system for authenticating a speaker according to a number string by combining the above two elements, namely, a speech recognition method and a speaker authentication method.

An object of the present invention is to provide an authentication method of the present invention that is related to the authentication of the speaker, and confirms the identity of the user from a string of numbers.

Another object of the present invention is to provide an authentication method of the present invention for identifying the user's identity through a string of numbers, such as a social security number, which is basic information used in the social life of an individual.

Another object of the present invention is to provide an authentication method of the present invention that confirms the identity of a user through a number string consisting of numbers and unique voice information of the speaker related to the authentication of the speaker.

It is still another object of the present invention to provide a method for authenticating a speaker by the string of numbers of the present invention, which represents a significant degree of recognition by adding voice text information of the speaker to the string speech recognition indicating a low recognition rate.

Another object of the present invention can be used to recognize the account number, social security number, password, etc. in the phone-banking service, or can be used in a variety of industries, such as civil affairs of public institutions, even in this case to ensure a high recognition rate An authentication method of the present invention is provided.

Another object of the present invention is to use the authentication method of the present invention more reliable in the matters related to the authentication that the user is a real identity by using a string of numbers indicating the identity of the individual and the voiceprint that is the unique identification information of the speaker, such as a social security number. To provide.

The above and other objects of the present invention can be achieved by the present invention described in detail below.

1 is a block diagram schematically illustrating an authentication system constructed in accordance with the present invention.

2 is a flowchart illustrating an operation process performed according to an embodiment of the present invention.

* Brief description of the major symbols in the drawings *

100: terminal 200: authentication server

210: speech model database 230: numeric string database

250: Gate Database 270: Speaker Recognizer

290: Speech Recognizer 300: Operator Server

Summary of the Invention

The present invention relates to a method of selecting a plurality of speech recognition target candidates from a string of numbers based on numbers, and determining the voiceprint information of the selected speech recognition target candidate to select one string string having the highest similarity.

The speaker authentication method according to the present invention relates to speech recognition and authentication, and comprises a process of registering a sequence of numbers consisting of numbers, and a process of authenticating speech recognition and a speaker of a spoken sequence of numbers based on the registered sequence of numbers, The registration process is related to the authentication of the speaker's resident registration number, account number, or password transmitted through the communication network, and inputs a string of numbers through the terminal to the authentication server; Then, the voiceprint information of the speaker transmitted through the communication network is extracted based on the numeric string input through the terminal, and the numerical string database and the voiceprint information database are created and stored in the memory corresponding to the numeric string input by the speaker in the step. doing; The voice recognition of the numeric string and the authentication process of the speaker, the voice recognizer extracts the feature parameters of the speech from the numeric string voice transmitted through the communication network from a plurality of voice model database including the speech recognizer Selecting candidate strings of recognition candidates; After selecting the same number strings by comparing the candidate strings selected by the speech recognizer with the number strings stored in the number string database and stored in the memory by the speaker through the registration process, the selected number strings are selected. Calling the voiceprint information corresponding to the numeric string from the voiceprint database and comparing the communication network with voiceprint information extracted from the transmitted numeric string voice, the speaker recognizer selects one string string having the highest similarity and outputs it as a recognition result; Calculating a threshold for the selected sequence of numbers to determine whether the value is higher than a set value; And if the threshold calculated in the step is higher than a preset value, authenticating that the speaker is a registered person; Consists of steps.

According to another aspect of the present invention, the authentication server is associated with authentication of an individual's social security number, account number, or password transmitted through a communication network, and includes a numeric string consisting of numbers and personal voiceprint information generated corresponding to the numeric string, respectively. A memory for storing correspondingly, and a voice information database which is basic information for recognizing a corresponding string of numbers from a voice spoken by the speaker, and the pronunciation of the speaker uttered according to the utterance characteristic of the word regardless of the speaker's voice. A voice recognizer for selecting a plurality of similar numeric strings from the voice information database, and the voiceprint information of the individual extracted from the numeric string stored in the memory corresponding to the plurality of numeric strings selected by the voice recognizer from the voice information database; Two voices are extracted by extracting the speaker's voice information. Comparing the beams, and if the result value is higher than the preset threshold comprises a speaker recognizer for outputting the authentication result value.

In addition, the authentication system according to another aspect of the present invention relates to an authentication system for connecting the terminal and the authentication server through a communication network, and the speaker system through the numeric string transmitted from the terminal, the authentication server, Memory associated with authentication of an individual's social security number, account number, or password, and the like, comprising a number string consisting of numbers, and a memory corresponding to each person's voiceprint information generated corresponding to the number string, and a voice spoken by the speaker. A voice information database, which is basic information for recognizing a corresponding number string, and selecting from the voice information database a plurality of strings similar to the pronunciation of the speaker according to the utterance of the word regardless of the speaker's voice A voice recognizer, and the voice recognizer The voiceprint information of the individual extracted from the numeric string stored in the memory corresponding to the plurality of strings selected from the base and the voiceprint information of the speaker are extracted, and the two voiceprint information are compared, and the result is greater than the preset threshold. And a speaker recognizer that outputs a result of authentication if it is high, and the terminal is connected to an authentication such as a resident registration number, an account number, or a password of an individual transmitted to the authentication server through the communication network. Means for inputting, and means for receiving and transmitting the voice of the individual to the authentication server.

Detailed description of the invention

Hereinafter, with reference to the accompanying drawings will be described in detail the contents of the present invention. The server used in the present specification includes all devices connected to a communication network to provide a specific service to a user, and are intended to be limited to a computer that provides information or a peripheral device to process a user's command in a computer communication network. It is not.

1 is a block diagram schematically illustrating a system implementing the present invention. In the present invention composed of the terminal 100, the authentication server 200, and the service server 300, the service server 300 confirms the user authentication through the authentication process of the present invention, and provides a corresponding service Means a device for.

The terminal 100 includes an individual or a device that receives a service provided by the service server 300 through an authentication process, and the terminal also changes as the communication network 400 changes. For example, when the communication network uses a public telephone switching network, a general landline telephone corresponds to a terminal, and when the individual uses the Internet, the terminal corresponds to a personal computer.

The terminal basically includes a device for inputting a string of numbers consisting of a combination of numbers necessary for the speaker to authenticate and a transmission device for delivering the speaker's voice. As an example, the device for inputting the string of numbers corresponds to a keypad including a number key, and the transmission device corresponds to a modem including a microphone for receiving a voice. However, the present invention is not limited thereto, and any conventional one may be used within the technical scope of the present invention.

Authentication server 300 is a device for implementing the present invention, the engine 290 (hereinafter referred to as 'voice recognizer') for recognizing the speaker's voice and the engine 270 for extracting and verifying the voiceprint from the speaker's voice ( It is configured to include 'speaker recognition' below.

The speech recognizer 290 selects a target candidate of the string of numbers used for speech recognition of the string of numbers from the speech model database 210 of the recognizer 290, and the speaker recognizer 270 is the speech recognizer 290. Speaker authentication is performed by extracting the speaker's feature parameters from the selected candidate string.

In addition, the authentication server 300 is related to the use of the service, such as the resident registration number, account number, or password of the individual that the speaker is used for authentication, and stores and manages a string of numbers in the database 230, and It includes a voiceprint database 250 that contains the feature parameters of the speaker according to the number string, that is, the voiceprint information.

The reference numeral 400 denotes a communication network, which is for providing communication between the terminal 100 and the authentication server 200, and may be any conventional one within the scope of not impairing the technical idea of the present invention. Do. For example, when the present invention is used in connection with phone-banking, the communication network will be a public switched telephone network, and when used for internet banking, it will be an internet network.

2 is a flowchart illustrating a speaker authentication process of a string of numbers according to an embodiment of the present invention. First, the authentication server 200 is related to the use of the service, such as an individual's social security number, account number, or password, and receives a string of numbers from the terminal 100 (S100). Although described in detail later, the numeric string is stored in a specific memory in correspondence with the voiceprint information corresponding to the numeric string, which is managed by the numeric string database 230 and the voiceprint database 250.

At this time, any communication method may be used between the server 200 and the terminal 100. For example, a wired internet network based on TCP / IP, a wireless internet network based on WAP, a wireless network using a mobile phone, or a public telephone switching network used in a phone-banking service may be used.

The authentication server 200 extracts the speaker's voice feature, that is, a feature parameter corresponding to the voiceprint, based on the number string transmitted to the server 200 (S200). The speaker pronounces the number string transmitted to the server a plurality of times, the authentication server 200 extracts the feature parameters of the speaker.

The feature parameter may represent the same voice feature according to the same letter or number regardless of the speaker, but depending on the speaker, the feature parameter may be different even in the same word due to differences in oral structure, pronunciation characteristics, and tone. The feature parameter extracted in this step (S200) refers to voice information that appears differently according to the speaker (hereinafter, referred to as 'gender information'), and is used to extract information corresponding to the voiceprint.

The information on the voiceprint of the extracted numeric string and the corresponding numeric string, that is, the text information of the number input by the speaker in the step S100 may correspond to a specific memory of the server, that is, the numeric string database 230 and the voiceprint database 250. )

Up to this step, the voice information of the speaker corresponding to the number string is registered for the next authentication process. The voice information of the speaker selected in the above steps and the number string information corresponding to the voice information are the voice of the number string in the subsequent authentication process. It is used to perform the recognition and thus the speaker authentication process.

The speaker speaks a string of numbers such as a resident registration number, an account number, and a password registered in the step S100 for authentication by voice (S300). The voice of the spoken person is inputted to the terminal, transmitted to the authentication server through a communication network, processed by the authentication server, and the next step is performed.

Next, the voice recognizer 290 of the authentication server 200 extracts the speaker's feature parameters based on the spoken number string (S400). The feature parameter of the voice extracted in this step is not the glottal information extracted in the above step, but refers to a feature that appears uniquely according to a word or a number regardless of the speaker. Using the feature parameter, the speech recognizer selects a plurality of strings similar to the spoken pronunciation. This increases the probability of the number string corresponding to the pronunciation pronounced among a plurality of strings in preparation for the deterioration of the accuracy of the speech recognizer, and then uses a single number by using the glottal information among the number strings selected in the process of recognizing the speaker. To select a column. In addition, when multiple users register with the same password, all of them are selected by the voice recognizer, and then a single string is selected by using voice information in the process of recognizing the speaker.

For example, a speaker named 'Hong Gil Dong' may register '123456' as a password, or 'Yi Sun Shin' may register the same password '123456'.

According to the operation of the speech recognizer 290, a plurality of candidate candidate strings to be recognized are selected from the speech model database 210, and a speaker authentication process is performed on the selected speech model.

The speech sequence selected by the speech recognizer 290 has corresponding feature parameters of the speaker registered in the step S200, that is, information on the voiceprint. Using this, the speaker recognizer extracts the voiceprint information of the speaker from the speaker's voice transmitted to the authentication server in operation S400 with the operation of the voice recognizer 290, and the voiceprint corresponding to the number string selected by the voice recognizer 290. The information is called from the gated database, and the two gated information, that is, the extracted voiced gate information and the gated information selected from the gated database, are selected to select a single string having the highest similarity.

The number string selected in the step S500 is not directly output as a result of the speaker authentication, but the recognition result may be lowered by a plurality of factors even though the speaker authentication process is performed.

Accordingly, the speaker recognizer 270 determines whether the selected one string is higher than the finally set threshold value, and recognizes the selected one string as the authentication result value when the speaker recognizer 270 is higher than the preset threshold value.

On the other hand, if it is lower than the threshold value, the process returns to step S300 for generating a string for authentication, which is a step for speaker authentication, to repeat steps S400 to S600, or a password registered by the speaker using the keypad of the terminal. It can also be configured to perform other steps, such as verification.

According to the present invention, the present invention shows a remarkable speech recognition rate in the numerical speech recognition that shows a low recognition rate in the prior art, the authentication method of the present invention combined with speech recognition and speaker recognition has an excellent effect without limiting the industrial use range in various forms Indicates.

Simple modifications and variations of the present invention can be readily used by those skilled in the art, and all such variations or modifications can be considered to be included within the scope of the present invention.

Claims (6)

Related to speech recognition and authentication, the process of registering a number sequence consisting of numbers, and the process of authenticating the speech recognition and the speaker of the spoken number string based on the registered number sequence, the registration process, Related to authentication of a speaker's resident registration number, account number, or password transmitted through a communication network, and inputs a string of numbers through a terminal to be transmitted to the authentication server; And Extracting voiceprint information of the speaker transmitted through the communication network based on the numeric string input through the terminal, and creating a numeric string database and a voiceprint database corresponding to the numeric string input by the speaker in the step and storing the same in the memory; Consists of steps, The voice recognition of the number string and the authentication process of the speaker, A speech recognizer extracts feature parameters of the speech from the speech string transmitted through the communication network and selects a plurality of candidate strings of recognition candidates from the speech model database included in the speech recognizer; After selecting the same number strings by comparing the candidate strings selected by the speech recognizer with the number strings stored in the number string database and stored in the memory by the speaker through the registration process, the selected number strings are selected. Calling the voiceprint information corresponding to the numeric string from the voiceprint database and comparing the voiceprint information extracted from the numeric string voice transmitted through the communication network, the speaker recognizer selects and outputs one numeric string having the highest similarity as a recognition result; Calculating a threshold for the selected sequence of numbers to determine whether the value is higher than a set value; And If the threshold calculated in the step is higher than a preset value, authenticating that the speaker is a registered person; Speaker authentication method comprising the steps. The method of claim 1, wherein when the calculating of the threshold is lower than the set value, the method returns to selecting the candidate strings for recognition. A memory associated with authentication of an individual's social security number, an account number, or a password transmitted through a communication network, and correspondingly storing a numeric string consisting of numbers and voiceprint information of the individual generated corresponding to the numeric string; It includes a voice model database that is a basic information for recognizing a corresponding sequence of numbers from the voice spoken by the speaker, wherein the speaker speaks according to the utterance characteristics of the word irrespective of the speaker's voice and recalls a plurality of similar sequence of numbers. A speech recognizer for selecting from a speech model database; And The voice recognizer extracts the voiceprint information of the person and the voiceprint information of the speaker extracted from the numeric string stored in the memory corresponding to the plurality of numeric strings selected from the voice model database, and compares the two voiceprint information. A speaker recognizer that outputs an authentication result when the result is higher than a preset threshold; Authentication server, characterized in that configured to include. The authentication apparatus of claim 3, wherein the speaker recognizer does not output an authentication result value when the result value is lower than a threshold value, and transmits a determination result to the voice recognizer to select a candidate candidate string again. server. An authentication system which connects a terminal and an authentication server through a communication network and performs speaker authentication through a numeric string transmitted from the terminal, wherein the authentication server includes: A memory associated with authentication such as a social security number, an account number, or a password of an individual transmitted through the communication network, and correspondingly storing a numeric string consisting of numbers and voiceprint information of the individual generated in correspondence with the numeric string; It includes a voice model database that is a basic information for recognizing a corresponding sequence of numbers from the voice spoken by the speaker, wherein the speaker speaks according to the utterance characteristics of the word irrespective of the speaker's voice and recalls a plurality of similar sequence of numbers. A speech recognizer for selecting from a speech model database; And The voice recognizer extracts the voiceprint information of the person and the voiceprint information of the speaker extracted from the numeric string stored in the memory corresponding to the plurality of numeric strings selected from the voice model database, and compares the two voiceprint information. And a speaker recognizer that outputs an authentication result when the result is higher than a preset threshold. The terminal, Means for inputting a string of numbers that is related to authentication, such as a resident registration number, an account number, or a password of a speaker transmitted through the communication network; And Means for receiving and transmitting a voice of the individual to the authentication server; Authentication system, characterized in that configured to include. The authentication apparatus of claim 5, wherein the speaker recognizer does not output an authentication result when the result value is lower than a threshold value, and transmits the determination result to the voice recognizer to select the candidate candidate string again. system.