JP2000067004A - Identity verification method, device using the method, and recording medium recording identity verification device control program - Google Patents

Abstract

(57) [Summary] [Problem] To reduce erroneous determination. SOLUTION: A similarity R between an input voice V1 and a voice model V0 recorded in advance corresponding to the principal ID is obtained,
In the identity verification method for determining whether or not the speaker M who uttered the input voice is the identity, the similarity R is determined by the threshold R
th, but when the difference ΔR from the threshold is relatively small, the date of birth registered in advance corresponding to the principal ID,
The speaker M is inquired of personal information PDi such as a home telephone number, and the answer to the question Q is input by voice. The speech content is identified from the input voice V2 of the answer by voice recognition technology. Identify yourself as correct as an answer. Further, the speaker determination based on the similarity may be performed again using the input voice V2 of the answer, and the identity of the speaker may be confirmed from the results of both.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention is used for confirming a person by voice in, for example, banking or shopping service using a telephone, and uses an input voice and a voice model registered in advance in correspondence with a personal ID. The present invention relates to a personal identification method for determining whether or not the speaker who uttered the input voice is an individual by determining the similarity of the personal identification, a device thereof, and a recording medium storing a personal identification device control program.

[0002]

2. Description of the Related Art In a conventional personal identification technique using voice, a similarity between an input voice and a voice model registered in advance corresponding to a personal ID is obtained, and the input voice is determined. Judge whether the speaker who uttered is himself, but the voice changes depending on the physical condition, and the input voice is deformed due to ambient noise and line distortion, etc., so it is really himself but not himself Erroneous judgment, or erroneous judgment that the person is the person who is not the person himself.

[0003]

According to a first aspect of the present invention, there is provided a method of obtaining a similarity between an input voice and a voice model recorded in advance corresponding to a principal ID, and uttering the input voice. The present invention relates to an identity verification method for determining whether or not a person is an identity. In particular, in the case where the similarity exceeds the threshold value, but the difference from the threshold value is relatively small, the date of birth and home telephone registered in advance in correspondence with the principal ID are particularly preferable. Ask the speaker for personal information such as a number, ask the speaker to input the answer to the question, identify the utterance content from the input voice of the above answer by voice recognition technology, and determine whether the answer to the above question is correct, Identify yourself.

(2) According to a second aspect of the present invention, in addition to the method of the first aspect, the speaker determination based on the similarity is performed again by using the input speech of the answer, and the identity is determined based on the result of both. Check. (3) The third aspect of the present invention relates to an identity verification device including first and second identity verification units and a system controller.

[0005] The first personal identification unit obtains a similarity R between the input voice and a voice model registered in advance corresponding to the personal ID, compares the similarity R with a threshold value Rth, and when R <Rth. Is determined that the speaker who uttered the input voice is another person, and R ≧ Rth
In the case of, ΔR = R−Rth and its allowable minimum value ΔRmi
n, and if ΔR ≧ ΔRmin, it is determined that the person is the person, or a confidence measure α = ΔR / Rth and its allowable minimum value α
and α is determined as the person if α ≧ αmin.

[0006] The second identity verification unit is adapted to execute the above-mentioned ΔR <ΔRmi
If n or α <αmin, ask for personal information such as the date of birth and home phone number registered in advance in correspondence with the principal ID, The utterance content is identified from the input speech of the answer by the speech recognition technology, and the speaker is determined based on whether or not the answer is correct as the answer to the question.

The system controller decodes and executes a control program stored in a built-in memory (recording medium) to control the operations of the first and second identification units. (4) In the invention according to claim 4, in the above (3), in addition to performing the speaker determination from the input voice of the answer, the first identity checker may determine the answer of the answer. The speaker determination based on the similarity is performed again using the input voice, and the identity of the speaker is confirmed from both determination results.

(5) According to a fifth aspect of the present invention, a speaker who utters the input voice by finding the similarity between the input voice and a voice model recorded in advance in correspondence with the user ID. The present invention relates to a recording medium on which a control program for controlling a personal identification device for determining whether or not to use a system controller by a system controller is recorded. If the similarity exceeds the threshold value but the difference from the threshold value is relatively small, the control program will call the home date of birth and home telephone registered in advance corresponding to the personal ID. The speaker is asked to ask personal information such as a number, the content of the utterance is identified from the answer to the question, it is determined whether the answer is correct as the answer to the question, and the speaker is determined from the result.

(6) The recording medium according to claim 6, wherein (5)
The control program makes the speaker determination based on the answer to the above-mentioned question, performs the speaker determination based on the similarity again using the answer to the above-described question, and performs the final speaker determination from the result of both. .

[0010]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS (Embodiment 1) An embodiment of the present invention will be described by taking a member verification service on a telephone network as an example. In the present invention, as shown in FIG. 1, first, an information ID ′ relating to a personal ID is obtained from a user (speaker) M and input to the personal ID specifying unit 1. In the principal ID specifying unit 1, the information ID 'is stored in the ID list (ID file) 1a.
Is confirmed to be registered in the voice model storage unit 2, and the user ID is specified. The voice model storage unit 2 stores a voice model V0 corresponding to the personal ID.
(Stored in advance) is input to the speaker recognition unit 3.

In the speaker recognition unit 3, the input voice V1 and the person I
The similarity R with the model V0 of the voice corresponding to D is obtained, and the threshold Rth indicating the range of the variation of the similarity R that can be regarded as the voice of the person is read from the threshold storage unit 4, and the similarity R is read. If the similarity R is larger than the threshold Rth, it is determined that the person is GO (person) and the threshold R
If it is smaller than th, it is determined to be NG (another person). G
In the case of O, the result determination unit 5 determines whether the amount ΔR = R−Rth in which the similarity is separated from the threshold is larger than the allowable minimum value ΔRmin, or the confidence measure α = ΔR / Rth is equal to the allowable minimum value. Identify whether it is greater than αmin, GO if it is greater
(The person), otherwise, it is judged as NG (another person).

[0012] The NG information is input from the result determination unit 5 to the personal information storage unit 6. In the personal information storage unit 6, the personal ID is acquired from the personal ID identification unit 1, and the personal information PDi (previously stored) corresponding to the personal ID is extracted and input to the personal information inquiry unit 7. The personal information interrogation section 7 asks for personal information PDi such as a birth date and a home telephone number registered in advance in correspondence with the personal ID, and asks the user to input the answer by voice to obtain correct personal information. PDi is input to the voice recognition unit 8. The speech recognition unit 8 determines the utterance content from the input speech V2 of the answer to the question,
In other words, if it matches with PDi, if it matches, it is determined as GO (person), and if it does not match, it is determined as NG (other person). If necessary, return to the personal information storage unit 6 to obtain another personal information PD.
The question is repeated about (i + 1), and a more reliable judgment result is issued.

The above-mentioned personal ID specifying unit 1 and ID list 1
a, a voice model storage unit 2, a speaker recognition unit 3, a threshold storage unit 4, and a result determination unit 5 constitute a first personal identification unit 100, a personal information storage unit 6, a personal information inquiry unit 7, and a voice recognition unit. The unit 8 constitutes a second identification unit 200. The system controller 9 has a built-in memory (recording medium) 9a
Of the first and second identification units. (Embodiment 2) In Embodiment 2, as shown in FIG.
In the same manner as described above, the second confirmation unit 200 asks for personal information, asks the user to input the answer to the above question by voice, determines the utterance content from the input voice V2 of the above answer by voice recognition technology, and confirms the identity. In addition, the input voice V2 of the answer is input to the speaker recognition unit 3, and the speaker determination based on the similarity is performed again as before, and the determination and the second identification unit 20 are performed.
The overall judgment unit 10 confirms the identity of the person from both the judgment with 0.

The speaker determination may be performed again by the speaker recognizing unit 3 based on whether or not the similarity R is larger than the threshold value Rth.
A determination result of whether or not R> ΔRmin or α> αmin may be added.

[0015]

According to the present invention, even when the conventional identity verification technology based on the similarity of voice cannot be sure of the judgment, the personal information is questioned and the answer is determined using the voice recognition technology to determine whether or not the answer is correct. Since identification and identification are performed, erroneous determination can be reduced as compared with the related art. According to the experiment, in the member verification service on the telephone network shown in the embodiment, when the conventional method is compared with the method of the present invention of FIG. 1, the rate of falsely recognizing that the user is another person is half. I was able to.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention.

FIG. 2 is a block diagram showing another embodiment of the present invention.

──────────────────────────────────────────────────続 き Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI Theme coat ゛ (Reference) H04M 3/42 P G06F 1/00 370E

Claims (6)

[Claims] 1. An identification method for determining whether or not a speaker who has uttered the input voice is an identity by determining a similarity between the input voice and a voice model recorded in advance corresponding to the identity. In the above, when the similarity exceeds the threshold value, but the difference from the threshold value is relatively small, when the birth date and home telephone number registered in advance corresponding to the personal ID Ask the speaker for personal information, ask the speaker to answer the question,
An identity verification method characterized by identifying the utterance content from the input voice of the answer by voice recognition technology, and confirming the identity of the answer as correct to the question. 2. A personal identification method for determining whether or not a speaker who has uttered an input voice is an individual by obtaining a similarity between an input voice and a model of a voice recorded in advance corresponding to a personal ID. If the similarity exceeds the threshold value, but the difference from the threshold value is relatively small, an individual such as a birth date or a home phone number registered in advance corresponding to the personal ID. Ask the speaker for information and ask them to speak the answer,
The voice recognition technology identifies the utterance content from the input voice of the answer, and performs the speaker determination based on whether the answer is correct as the answer to the question, and performs the speaker determination based on the similarity again using the input voice of the answer. An identity verification method, wherein the identity of the user is verified based on the results of the two. 3. An identification device comprising first and second identification units and a system controller, wherein the first identification unit includes an input voice and a voice registered in advance in correspondence with a personal ID. Finding the similarity R with the model of
If R <Rth, then the speaker who uttered the input speech is judged to be another person. If R ≧ Rth, then ΔR = R−Rth and its allowable minimum value ΔRmin are further compared. If ΔR ≧ ΔRmin, the user is determined to be a person, or the confidence measure α = ΔR / Rth is compared with its allowable minimum value αmin, and if α ≧ αmin, the person is determined to be a person. In the case of ΔR <ΔRmin, or in the case of α <αmin, the user is asked for personal information such as the date of birth and home telephone number registered in advance corresponding to the personal ID, and the answer to the question is spoken. Input, identify the utterance content from the input speech of the answer by speech recognition technology, and determine whether the speaker is correct as the answer to the question, and determine the speaker. Decrypt and execute the stored control program Identification device and controls the first, the operation of the second identification portion. 4. The method according to claim 3, wherein the second identification unit uses the input speech of the answer in addition to the speaker identification performed from the input speech of the answer. An identity verification device, wherein the speaker determination based on the similarity is performed again, and the identity is confirmed based on both determination results. 5. An identification device for determining whether or not a speaker who has uttered the input voice is a true person by obtaining a similarity between the input voice and a voice model recorded in advance corresponding to the personal ID. Is a recording medium on which a control program for controlling the similarity by a system controller is recorded, wherein the similarity exceeds the threshold, but the difference from the threshold is relatively small. Ask the speaker to ask personal information such as the date of birth and home phone number registered in advance in correspondence with the principal ID, identify the utterance content from the answer to the question, and correct the answer to the above question. A recording medium storing a personal identification device control program, characterized in that it is determined whether or not the speaker is a speaker, and a speaker is determined from the result. 6. An identity verification device for determining the similarity between an input voice and a voice model recorded in advance corresponding to a personal ID and determining whether or not the speaker who utters the input voice is the identity. Is a recording medium on which a control program for controlling the similarity by a system controller is recorded, wherein the similarity exceeds the threshold, but the difference from the threshold is relatively small. Ask the speaker to ask personal information such as the date of birth and home phone number registered in advance in correspondence with the principal ID, identify the utterance content from the answer to the question, and correct the answer to the above question. Is determined, and the speaker determination is performed based on the result, and the speaker determination based on the similarity is performed again using the answer to the question, and the final speaker determination is performed based on the results of the two. It is characterized by Recording medium that records the identity verification device control program that.