JP2000172296A - Speaker collating device and electronic mail system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an electronic mail system capable of coping with pretense as a sender by a third party by providing the system with an effective means to decide whether or not the sender of the electronic mail is the person in question. SOLUTION: This device is provided with a speaker collation processing part 12 for judging whether or not a sender is the person in question, by which a speech feature pattern in which a sender spoke his/her password is added to the electronic mail in addition to a conventional processing to receive and transmit an electronic mail, and the receiver of the electronic mail compares the speech feature pattern with a reference use speech feature pattern of the sender in question received beforehand, and is further made to a system provided with prevention measures to misrepresentation by duplication and change in speech with the elapse of time, and the system is enabled to resist to steal and duplication of the password by a third party and process collation of the sender in question by using the speech feature pattern of a personal living body feature.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speaker verification apparatus for analyzing a voice feature pattern of a sender attached to an electronic mail and determining whether or not the voice characteristic pattern belongs to the sender, and a speaker verification apparatus for the speaker verification apparatus. The present invention relates to an electronic mail system using a device.

[0002]

2. Description of the Related Art At present, electronic mail is widely used as a communication means between individuals and companies, and further demand is expected to increase in the future due to the progress of electronic commerce and the like.

Generally, in an electronic mail, a sender's name and a mail address are written in a header, and a receiver can see the mail to know from whom.

[0004]

By the way, even if the sender's name is described in the e-mail as described above, is it really the person who is notated, or if a third party intrudes into the terminal and the person's name is used? The receiving side did not have any effective means to determine whether or not the e-mail was sent.

Here, as a method of collating a sender, a method of adding a password known only to the sender and the recipient to the mail to check for spoofing is considered. If the password is stolen, the spoofing is detected. Can not do.

The present invention has been made in consideration of the above-mentioned conventional problems, and a speaker verification device capable of determining whether or not a sender is the person described in an e-mail even if a third party knows the contents of the password, and a spoofing method. The purpose is to provide an email system that can be prevented.

[0007]

SUMMARY OF THE INVENTION In order to achieve the above object, the present invention relates to a speaker verification apparatus for verifying a speaker's voice, which refers to a reference voice feature pattern of a sender previously received. The similarity calculated by the similarity calculator, which stores the received voice feature pattern, the stored reference pattern, and the similarity in the use pattern storage unit, and is the person whose mail sender is indicated using the voice password. It is assumed that the speaker verification device is configured to verify whether or not the speaker verification is performed.

According to the present invention, it is possible to reliably determine whether or not the sender of the mail is the stated person by collation, and to completely prevent spoofing in the electronic mail.

[0009]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS According to the first aspect of the present invention, there is provided an electronic mail receiving unit for receiving and storing an electronic mail,
A speaker verification control unit that controls the speaker verification process, a reference pattern storage unit that stores a reference voice feature pattern of the caller that has been received in the past, and a reference that stores the received voice feature pattern. And a similarity calculation unit for calculating the similarity, and a speaker verification device for verifying whether or not the sender of the mail is the written person using a voice password. This is to verify the identity of an e-mail using a voice feature pattern, which is one of the following.An e-mail with a voice feature pattern attached by a sender using a voice password is received by some means and stored in advance. Compared with the voice feature pattern for reference of a certain sender, the process of determining whether the sender is the person described in the email is performed, and even if a third party knows the contents of the password, Since it is not possible to produce a similar speech feature pattern, an effect that it is possible to prevent spoofing.

According to a second aspect of the present invention, in the speaker verification apparatus according to the first aspect, the received voice feature pattern is a voice feature pattern for personal reference or a voice feature pattern previously received. If it completely matches, it is equipped with a spoofing prevention processing unit that judges that it is a spoofing by duplication, and the voice feature pattern for personal reference or the voice feature pattern that completely matches the previously received voice feature pattern When the feature pattern is received, it has an effect of performing a spoofing prevention process of determining that it is a copy.

According to a third aspect of the present invention, in the speaker verification device according to the first or second aspect, when the received voice feature pattern can be determined not to be spoofing, it is used for reference. It has a pattern and a time-dependent change processing unit that updates the pattern as needed, and has the effect that the matching accuracy can be kept constant even if the voice characteristics of the speaker change with time.

According to a fourth aspect of the present invention, there is provided an electronic mail transmitting section for performing a transmission process by attaching a voice feature pattern to an electronic mail, a voice input section for recording a password voice of a mail sender, and a recording section. A voice feature pattern calculation unit that calculates a voice feature pattern from the obtained voice signal, a speaker verification control unit that controls the speaker verification process, and a voice feature pattern for personal reference transmitted to the receiver in the past. A reference pattern storage unit to be stored, and a similarity calculation unit that calculates the similarity between the voice feature pattern to be transmitted and the reference pattern, and attaches and transmits it only when the user can be authenticated. And a sender of the email having
This is an electronic mail system in which the speaker verification device is composed of an e-mail receiving side and is formed by both the transmitting side and the receiving side. On the transmitting side, a voice feature calculated by recording a voice signal in which the caller utters a password is calculated. The pattern is attached to the e-mail, and the receiving side compares the attached voice feature pattern with the reference pattern to determine whether or not the sender is the person himself / herself, and has the effect of preventing spoofing.

According to a fifth aspect of the present invention, in the electronic mail system according to the fourth aspect, the receiving side performs a fraud prevention process of the speaker verification device of the second aspect, and It has the effect of preventing spoofing.

According to a sixth aspect of the present invention, in the electronic mail system according to the fourth or fifth aspect, the receiving side performs the time-dependent change processing according to the third aspect,
This has an effect that the receiving side can cope with a temporal change of the voice feature pattern.

Hereinafter, embodiments of the present invention will be described with reference to the drawings. (Embodiment 1) FIG. 1 is a configuration block diagram of a speaker authentication device and an electronic mail device according to Embodiment 1 of the present invention.

As shown, an e-mail receiving side (speaker verification device) 1 includes an e-mail reception unit 11 for receiving and storing an e-mail, and a speaker verification control unit 12 for controlling a speaker verification process. And a reference pattern storage unit 13 for storing a previously received caller's reference voice feature pattern, and a similarity calculation unit for calculating the similarity between the received voice feature pattern and the stored reference pattern. 14, an anti-spoofing processing unit 15 for determining whether or not the audio feature pattern is a spoof by duplication, and a temporal change handling unit 16 corresponding to temporal changes in voice utterance.

The transmitting side 2 of the e-mail includes an e-mail transmitting section 21 for performing a transmitting process by attaching a voice feature pattern to the e-mail, a voice input section 22 for recording a password voice of the mail sender, and a recorded voice. A voice feature pattern calculation unit 23 that calculates a voice feature pattern from a voice signal, a speaker verification control unit 24 that controls the speaker verification process, and a reference that stores a voice feature pattern for personal reference transmitted in the past. And a similarity calculation unit 26 that calculates the similarity between the voice feature pattern to be transmitted and the reference pattern.

The processing of the speaker verification device and the electronic mail system according to the first embodiment configured as described above will be described.

(1) Process on the Calling Side The process of attaching the voice feature pattern of the caller referred to by the speaker verification device 1 will be described with reference to the flowchart of FIG.

In the electronic mail system, the sender performs this processing. Step 1: When the sender of the e-mail performs transmission processing after creating the text, the e-mail system requests the utterance of a password attached to the e-mail.

Step 2: The sender utters the same password as previously transmitted to the receiver, and the voice signal is recorded. Recording parameters are from 8 to 22 KH depending on the balance between matching accuracy and processing volume.
z, the quantization precision is set to 8.16 bits.

Step 3: The recorded audio signal is divided into frames, subjected to acoustic analysis, and converted into an audio feature pattern such as a time series of cepstrum / Δcepstrum coefficients. Although it is conceivable to directly attach and transmit an audio signal, it is not practical because the data amount is very large. The recorded audio signal is discarded immediately after analysis.

Step 4: The calculated voice feature pattern is compared with a reference voice feature pattern that has been transmitted to the receiver in advance, and if it cannot be determined that the voice feature pattern matches to a certain extent, the flow returns to step 2 to re-utter the password. By this processing, the possibility that the receiving person is incorrectly rejected on the receiving side due to utterance error or noise contamination can be theoretically reduced to zero. However, storing the voice feature pattern for the user's reference on the starting side has a risk of spoofing due to duplication, so that spoofing prevention processing is effective. When a probabilistic model such as an HMM is used for the reference pattern, there is no danger of duplication because the configurations of the test pattern and the reference pattern are completely different.

Step 5: Along with the sender name of the mail,
Send the e-mail with the voice feature pattern of the password attached.

(2) Processing on the Receiving Side The following describes a voice feature pattern matching process in the speaker matching device according to the first embodiment with reference to the flowchart in FIG.

In the electronic mail system, this processing is performed on the receiver side. Step 1: The receiving side 1 receives the electronic mail transmitted to itself, and confirms the sender name of the mail.

Step 2: If the voice characteristic pattern of the sender is attached to the e-mail, it is determined by the receiver whether or not the sender is to be verified. This timing may be before or after viewing the body of the email. If the sender is checked before looking at the e-mail, the malicious virus sent by the impostor can be prevented from being developed.

Step 3: If it is determined that collation of the caller is necessary, the speaker collation processing is started.

The attached voice feature pattern is developed and compared with the voice feature pattern of the password of the sender who has received it in advance. The similarity between patterns is
The time series patterns having the same parameter are calculated by the DP matching method or the like, and when the reference side is a stochastic model such as the HMM, the pattern generation probability is calculated.

Step 4: If the calculated similarity is larger than the threshold value, that is, if the degree of matching with the person is high, it is determined that the sender described in the mail is genuine.

According to the second aspect of the present invention, the following spoofing prevention processing is performed in order to prevent spoofing caused by copying of the voice feature pattern on the transmitting side.

Step 5: The comparison with the voice feature pattern received in the past from the same sender is continuously performed.

Step 6: If any one of the past voice feature patterns completely matches, it is determined that it is a copy by a third party. Generally, even if the same speaker utters a voice feature pattern, it cannot be completely matched (the distance between the patterns becomes zero). Therefore, there is no danger of erroneously rejecting the person by this processing.

If the voice feature pattern is scrambled on the transmission side using a conversion formula generated based on the contents of the mail text (character array pattern etc.), and the same conversion formula is generated and restored on the receiving side, , And become more fraudulent. Even if a third party duplicates the voice feature pattern and attaches it to an e-mail of other content, the original pattern is not restored because the conversion formula for descrambling is different, and it can be determined to be spoofing at the matching stage.

In the voice collating apparatus according to the third aspect, when it is determined that the sender is not spoofing, the following aging process is performed.

Step 7: The received voice feature pattern is replaced with a reference pattern. Next time, speaker verification will be performed using this pattern. As a result, even if the utterance of the speaker changes with time, it is possible to suppress a decrease in the matching accuracy.

When a probabilistic model such as an HMM is used as the reference pattern, the model is modified by the MAP estimation method using the latest voice feature pattern to cope with a temporal change.

(3) Processing on the sending / receiving side of the electronic mail system The electronic mail system according to the fourth aspect has the following features.
3 is an e-mail system configured by both transmission and reception, which performs the processing on the sender side and performs the processing on the reception side shown in FIG. 3 on the receiving side.

An electronic mail system according to a fifth aspect performs the spoofing prevention process according to the second aspect shown in FIG. 3 on the receiving side of the electronic mail system according to the fourth or sixth aspect.

In the electronic mail system according to the sixth aspect, the receiving side of the electronic mail system according to the fourth or fifth aspect performs the aging process shown in FIG.

[0041]

As is apparent from the above description, according to the present invention, it is possible to verify the caller's speaker using the voice password attached to the e-mail, and the caller's judgment can be surely made. it can.

Further, according to the present invention, it is possible to more robustly verify the speaker of an e-mail message against a spoofing by a third party, and it is difficult to reduce the accuracy of the utterance of the speaker even with time. The ability to verify e-mail speakers, to configure a high-security e-mail system using a voice password, to configure an e-mail system that is more robust against spoofing by a third party, There is an effect that an e-mail system in which the accuracy does not easily decrease even with the lapse of time can be configured.

[Brief description of the drawings]

FIG. 1 is a configuration block diagram of a speaker verification device and an electronic mail device according to a first embodiment of the present invention.

FIG. 2 is a flowchart of a process on a transmitting side in the electronic mail device.

FIG. 3 is a flowchart of processing in the speaker verification device on the receiving side.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Receiving side (speaker verification device) 2 Transmission side 11 E-mail receiving unit 12 Speaker verification control unit 13 Reference pattern storage unit 14 Similarity calculation unit 15 Spoof prevention processing unit 16 Temporal change correspondence processing unit 21 E-mail transmission unit Reference Signs List 22 voice input unit 23 voice feature pattern calculation unit 24 speaker verification control unit 25 reference pattern storage unit 26 similarity calculation unit

Claims (6)

[Claims] An e-mail receiving unit that receives and stores an e-mail; a speaker verification control unit that controls a speaker verification process;
A reference pattern storage unit that stores a reference voice feature pattern of the caller received in the past, and a similarity calculation unit that calculates a similarity between the received voice feature pattern and the stored reference pattern. A speaker verification device for verifying, using a voice password, whether or not the sender of the mail is the stated person. If the received voice feature pattern completely matches the voice feature pattern for personal reference or the voice feature pattern previously received, a spoofing prevention processing unit that determines that the voice feature pattern is a fake by duplication is provided. The speaker verification device according to claim 1, further comprising: 3. A time-dependent change processing unit which, when it is determined that the received voice feature pattern is not spoofed, updates the reference pattern and the time-dependent change using the pattern if necessary. 2. The speaker verification device according to item 2. 4. An electronic mail transmitting unit for performing a transmission process by attaching a voice characteristic pattern to an electronic mail, a voice input unit for recording a password voice of a mail sender, and calculating a voice characteristic pattern from the recorded voice signal. A voice feature pattern calculating unit for controlling the speaker matching process, a reference pattern storing unit for storing a voice feature pattern for personal reference transmitted to the receiver in the past, and a transmitting unit. A sender of an e-mail having a similarity calculator for calculating a similarity between a voice feature pattern to be tried and a reference pattern, and having a function of attaching and transmitting the same only when the user can be authenticated; Item 1. An electronic mail system wherein the speaker verification device according to item 1 comprises a receiving side of the electronic mail, and is established by both the transmitting side and the receiving side. 5. The electronic mail system according to claim 4, wherein the receiving side performs a fraud prevention process of the speaker verification device according to claim 2. 6. The e-mail system according to claim 4, wherein the receiving side performs the time-dependent change processing according to claim 3.