JP3522421B2 - Speaker recognition system and speaker recognition method - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speaker recognition system and a speaker recognition method for speaker recognition.

[0002]

2. Description of the Related Art Conventionally, when a user uses an application such as deposit / withdrawal or balance inquiry in a bank, a personal identification number (PIN) or the like has been used to confirm that the user is the authorized user. I'm trying to let the user enter it. Further, the computer confirms the identity of the user by allowing the user to enter a personal identification code string similar to a personal identification number, called a password. However, such confirmation by inputting the personal identification number or personal identification character string can be stolen without difficulty as long as another person knows the personal identification number or personal identification character string. Moreover, many PINs and PINs use the date of birth or anniversary of the person who registered the PIN (person), the telephone number, or the spelling of his / her name. It's not difficult.

In order to avoid such drawbacks of the personal identification number and the personal identification character string, in recent years, attention has been paid to so-called speaker recognition, which is to judge whether or not the person is the person by voice. This speaker recognition is a feature amount (feature pattern) of a voice uttered by a speaker.
Determines whether the person is who he or she is by checking if it matches the voice standard pattern of this speaker registered in advance.
(Recognize). That is, the similarity between the feature amount (feature pattern) extracted from the speaker's voice and the voice standard pattern of this speaker is calculated, and it is determined whether or not the person is the person based on the degree of similarity. Since it utilizes the physical characteristics, it is difficult for other people to imitate the voice as compared with the personal identification number or the personal identification character string, and thus it is possible to more effectively prevent the theft of another person.

[0004]

However, such a speaker recognition system is conventionally generally installed at, for example, a bank window, and therefore, in order to perform speaker recognition, the user must There was a problem that I had to go to a bank counter etc. each time.

According to the present invention, a user can recognize a speaker without going to a bank counter or the like, and can use applications such as deposit / withdrawal and balance inquiry provided by a bank. The present invention aims to provide a simple speaker recognition system and a speaker recognition method.

[0006]

In order to achieve the above object, the invention according to claim 1 has at least one terminal and an intermediate device.
Central device is installed to send and receive information,
Is a voice input means for inputting the voice of the speaker and converting it into a voice signal.
And a feature extraction means for extracting a feature amount of a voice signal, and a speaker
Voice features and voice features as speaker recognition information
And a similarity calculation means for calculating the similarity of
In addition, the central unit has a speaker recognition that manages speaker recognition information.
Based on the similarity between the management means and the terminal similarity calculation means.
A judgment means for judging the talker is provided.
At the end, speaker recognition information used for similarity calculation
Is transferred from the central unit to the terminal, and the terminal similarity
The similarity calculated by the output means is transferred from the terminal to the central unit.
It is characterized by being adapted to .

[0007]

According to the second aspect of the invention, at least one terminal and the central unit are provided so as to be able to transmit and receive information, and the terminal inputs the voice of the speaker into a voice signal. Means, feature extracting means for extracting a feature amount of a voice signal, similarity calculating means for calculating a degree of similarity between the feature amount of the voice of the speaker and the voice feature amount as the speaker recognition information,
First determination means for determining the speaker based on the similarity from the similarity calculation means is provided, and the central device is
Second determination means for determining the speaker based on the similarity from the terminal similarity calculation means is provided, and the similarity calculated by the terminal similarity calculation means is the first determination of the terminal. Means provided or from the terminal to a second of the central unit
It is characterized in that it is transferred to the determination means of.

[0009] The invention of claim 3, wherein the claim 1 or
Further, in the speaker recognition system according to claim 2 , further, predetermined information is transferred from the central device to the terminal, and the feature extracting means provided in the terminal is based on the information provided from the central device. The feature is that the input voice is converted into a feature amount.

Further, in the invention according to claim 4, in the speaker recognition system according to claim 1 , change or modification of the speaker recognition information managed by the speaker recognition management means is decided. It is characterized in that it is regarded only by the information from the terminal that has been accessed.

Further, an invention according to claim 5, wherein, in claim 1 Symbol <br/> placing the speaker recognition system, one speaker recognition information from the central unit, at the same time only to only one terminal Is characterized in that it cannot be supplied.

[0012]

Further, in the invention according to claim 6 , at least one terminal and the central unit are provided so as to be capable of transmitting and receiving information, and when the voice of the speaker is input to the terminal, the voice signal is output. Is extracted, and the similarity between the feature amount of the voice of the speaker and the voice feature amount as the speaker recognition information is calculated. At this time, it is used in the similarity calculation in the terminal. The speaker recognition information is transferred from the central device to the terminal, the similarity calculated by the terminal is transferred from the terminal to the central device, and the central device determines the speaker based on the transferred similarity. The feature is to do.

Further, according to the invention of claim 7, at least one terminal and a central unit are provided so that information can be transmitted and received. When the voice of the speaker is input to the terminal, the voice signal is transmitted. Is extracted, and the similarity between the feature amount of the voice of the speaker and the voice feature amount as the speaker recognition information is calculated, and the similarity calculated by the terminal is
It is used for the determination of the speaker at the terminal, or is transferred from the terminal to the central device and used for the determination of the speaker at the central device, and the terminal performs processing only in the terminal.
Or to send and receive information to and from external devices.
Flip and is characterized by a Rukoto to change the threshold of the determination of the speaker.

[0015]

[0016]

FIG. 1 is a diagram showing an example of the configuration of a general speaker recognition system. Referring to FIG. 1, this speaker recognition system is for confirming the person himself / herself in a bank or the like by speaker recognition, and is a voice input means (for example, a microphone) for inputting a voice of a user. 1, a specifying means (for example, a keyboard) 2 for allowing a user to input predetermined specification information, and a voice that detects only a voice part of a speaker from a signal input from the voice input means 1 as a voice section. A section detection unit 3, a feature extraction unit 4 for extracting a feature amount (feature pattern) from a voice signal in the voice section detected by the voice section detection unit 3, and a speaker voice standard prior to speaker recognition. Registering unit 6 which pre-registers a typical feature amount (feature pattern) as a standard pattern in the speaker recognition information storage unit 5, and a feature amount of the voice of the user (speaker).
(Characteristic pattern) is collated with the standard pattern registered in the speaker recognition information storage unit 5, and the speaker recognition unit 7 that recognizes the speaker based on the similarity and the registration that registers the standard pattern It has a switching unit (for example, a switch) 8 that switches between a mode and a recognition mode for speaker recognition.

Here, the feature extraction unit 4 may convert the voice signal as a feature amount (feature pattern) into a spectrum or an LPC cepstrum, and the type of the feature amount is not particularly limited. Not something to do. It should be noted that in order to convert to a spectrum, F to conversion of feature quantity
FT is preferably used, and LPC analysis or the like is preferably used for conversion into LPC cepstrum.

When the standard pattern is registered (in the registration mode), the registration unit 6 uses the feature quantity (feature pattern) extracted by the feature extraction unit 4 based on the voice uttered by a speaker as the standard pattern. When registering in the speaker recognition information storage unit 5, as shown in FIG. 2, the designation information input from the designation means 2 by this speaker (for example, the name and birth date of this speaker, or this speaker). It is possible to register the standard pattern in the speaker recognition information storage unit 5 in association with the personal identification number (No. In other words, the speaker recognition information storage unit 5
The speaker recognition information necessary for speaker recognition is registered in the speaker recognition section, and the speaker recognition information storage unit 5
The speaker recognition information of a plurality of speakers (for example, users A, B, C, D, ...) Can be registered in.

Further, as a standard pattern of voices registered in the speaker recognition information storage unit 5, each user (speaker) is made to speak a word in advance in accordance with the usage pattern of the speaker recognition system. Alternatively, each user may freely utter a desired word.

Also, the speaker recognition unit 7 determines the characteristic pattern of the current speaker's voice as the speaker recognition information, as described in, for example, "Digital Speech Processing" by Furui (Tokai Publishing Co., Ltd.). A speaker that determines which one of the standard patterns of the plurality of speakers registered in the storage unit 5 is most similar, and identifies one speaker from the plurality of registered speakers. The identification pattern may be used, or a standard pattern corresponding to the current speaker is extracted from the standard patterns of a plurality of speakers registered in the speaker recognition information storage unit 5, and the standard pattern and the current pattern Speaker verification that matches the speaker's characteristic pattern and determines whether the current speaker is the regular speaker or not based on whether the similarity is higher or lower than a predetermined reference value (threshold) It may be of a system.

Further, the speaker recognizing unit 7 determines that the standard pattern of the voice registered in the speaker recognizing information storage unit 5 is for each user.
When the (speaker) has spoken a word in advance, the corresponding recognition can be performed, and the standard pattern of the voice registered in the speaker recognition information storage unit 5 can be used. Is a voice in which a desired word is freely uttered for each user, recognition corresponding to this can be performed. However, when each user (speaker) utters a predetermined word to perform speaker recognition, a similar criterion (threshold) can be set to a constant value for each speaker. However, when a desired word is uttered for each user to perform speaker recognition, a similar determination standard (threshold value) can be made different for each speaker.

In the following, for convenience of explanation, the speaker recognition system causes each user (speaker) to speak a predetermined word (specific word), and the speaker recognition unit 7
It is assumed that speaker recognition is performed by speaker verification. When the speaker recognition unit 7 performs speaker recognition by the speaker verification method, the user (speaker) is designated by the specifying unit 2 at the time of speaker recognition.
It is necessary to enter the same specified information as the specified information entered in the registration mode. As a result, in the speaker recognition unit 7,
A standard pattern corresponding to the current speaker can be extracted from the standard patterns of a plurality of speakers registered in the speaker recognition information storage unit 5, and the characteristics of the standard pattern and the voice of the current speaker can be extracted. The pattern can be matched.

When a user (for example, D) uses the speaker recognition system having such a configuration for the first time, this user (speaker)
First, D needs to register his own voice as a standard pattern. Therefore, the user D operates the switching unit (for example, the switch) 8 to connect the feature extraction unit 4 to the registration unit 6 and set the registration mode.

Next, the user (speaker) D inputs predetermined designation information, for example, (user D) from the designation means 2. Further, at this time, the user utters a predetermined specific word. This voice is input from the voice input means 1, converted into a feature amount (feature pattern) by the voice section detection unit 3 and the feature extraction unit 4, and given to the registration unit 6 as a standard pattern of the voice of this speaker. .

As a result, the registration unit 6 registers the standard pattern of the voice of the user (speaker) D in the speaker recognition information storage unit 5 in association with the designation information input from the designation unit 2. . For example, in the past, a plurality of users (different users) A, B, and C have registered their own voices as standard patterns in the speaker recognition information storage unit 5, and the current user D is as described above. When the user's own voice is registered as a standard pattern, the standard pattern is stored (registered) in the speaker recognition information storage unit 5 as shown in FIG.

In this way, when the standard pattern of the voice is stored in the speaker recognition information storage section 5, the user D
With this speaker recognition system, the speaker recognition for the user D can be performed. That is, this user D can use this system to determine whether or not the user who is currently using is the user D himself / herself.

Specifically, when the user D subsequently uses this system, the user D operates the switching unit 8 to connect the feature extracting unit 4 to the speaker recognizing unit 7, and to use this system. Set to recognition mode.

Next, the user D inputs predetermined designation information, for example, (user D) from the designation means 2. Further, at this time, the user D utters a predetermined specific word. This voice is input from the voice input means 1, and the voice section detection unit 3 and the feature extraction unit 4 input a feature amount (feature pattern).
And is given to the speaker recognition unit 7.

As a result, the speaker recognizing unit 7 causes the specifying unit 2
The standard pattern registered in association with the designated information (user D) input from is extracted from the speaker recognition information storage unit 5 and the standard pattern is compared with the feature pattern from the feature extraction unit 4. The similarity is calculated, and it is determined whether the similarity is higher or lower than a predetermined reference value. As a result, when it is determined that the degree of similarity is low, it is determined that the user is not the regular speaker himself D, and the use by this user is rejected. On the other hand, when it is determined that the degree of similarity is high, it is determined that the user is the regular speaker himself D, and the use is permitted by the user. That is, the user is permitted to use the application (for example, processing such as deposit / withdrawal and balance inquiry).

By the way, a speaker recognition system as shown in FIG. 1 is generally installed at a bank teller, etc.
Therefore, there is a problem that the user has to go to a bank counter or the like each time to perform speaker recognition.

The present invention avoids such a problem and allows the user to recognize the speaker at his / her home or at the user's office, etc. , Applications such as balance inquiry) are intended to be available.

FIG. 3 is a diagram showing a first configuration example of the speaker recognition system according to the present invention. In this first configuration example, the speaker recognition system includes at least one terminal 31-1 to 31-3.
1-n and the central unit 32 are provided so that information can be transmitted and received (for example, wired or wireless communication means 33-1).
~ 33-n are provided so that they can communicate with each other). Here, it is assumed that the terminals 31-1 to 31-n have the same configuration for convenience of description.

In the configuration example of FIG. 3, in the configuration example of FIG. 1, the voice input means 1 (or, further, the voice section detection unit 3), the feature extraction unit 4, and the speaker recognition unit 7 are provided on the terminal side. Further, the speaker recognition management means 10 is provided in the central device 32. Here, the speaker recognition management means 10 includes the speaker recognition information storage unit 5 and the registration unit 6 of FIG.
It also has the function of, and can also have the function of managing and controlling the entire speaker recognition. Further, the function of the switching unit 8 in FIG. 1 may be provided to the terminal side, or the central device 32.
Although it may be provided on the side, in the following, for convenience, it is assumed that the function of the switching unit 8 is provided on the terminal side.

FIG. 4 is a diagram showing a specific example of the speaker recognition system of FIG. Note that, in FIG. 4, only one terminal 31-1 is shown for simplicity, but other terminals 31-2 are shown.
.. 31-n have the same configuration as the terminal 31-1. Referring to FIG. 4, the terminal 31-1 is provided with a voice input unit 1, a designation unit 2, a voice section detection unit 3, a feature extraction unit 4, a speaker recognition unit 7, and a switching unit 8, and further, the terminal 31. -1, means for communicating information with the central unit 32
A transmission / reception interface unit 34-1 for transmitting / receiving via (for example, a telephone line or wireless) 33-1 is provided. Further, the central device 32 is provided with a speaker recognition information storage unit 5 and a registration unit 6, and the central device 32 further includes
A transmission / reception interface unit 35 for transmitting / receiving information to / from each of the terminals 31-1 to 31-n is provided.

Here, as each of the terminals 31-1 to 31-n, for example, a personal computer (a personal computer having a microphone, a voice capturing function such as A / D conversion) can be used. The users of -1 to 31-n can own their own terminals, for example, at home or in the office. More specifically, an existing personal computer (personal computer having a personal computer communication function) can be used for each of the terminals 31-1 to 31-n. In this case, in the configuration example of FIG. Transmission / reception interface unit 34-1
Is realized, for example, as a modem built in a personal computer, the voice input means 1 is realized by a microphone provided in the personal computer, the designating means 2 is realized by a console of the personal computer, and the voice section detection unit 3 is provided. The feature extraction unit 4 and the speaker recognition unit 7 are realized as software installed in a personal computer, for example, voice section detection software, feature extraction software, and speaker recognition software.

Further, in the configuration example of FIG. 4, the central unit 3
An exchange, for example, can be used as the second transmission / reception interface 35. The registration unit 6 is realized as registration software installed in the central device, and the speaker recognition information storage unit 5 can use a memory provided in the central device 32.

Further, in the configuration example of FIG. 4, the switching unit 8 provided on the terminal side can be configured as, for example, a switch operated by a user on the terminal side.

Here, the speaker recognition information storage unit 5 stores, as the speaker recognition information, a standard pattern in association with the designated information, for example, as shown in FIG. When the speaker recognition is performed on the side, if an instruction to that effect is transferred from the terminal to the central device 32, the central device 32 side recognizes the speaker recognition stored in the speaker recognition information storage unit 5. The information is read out and transmitted to the terminal. As a result, the speaker recognition unit 7 on the terminal side
Compares the feature pattern of a speaker's voice with the standard pattern of the speaker recognition information transmitted from the central unit 32, obtains the similarity between the feature pattern of the speaker and the standard pattern, and Person recognition can be performed. More specifically, for example, when the terminal side recognizes the speaker by the speaker verification method, the central unit 32 uses the standard pattern corresponding to the designation information from the designation unit 2 of the terminal for the speaker recognition. It can be transmitted to the terminal as information.

In such a speaker recognition system, the user is installed in the user's home or office in order to register the standard pattern (further change or update the standard pattern) and recognize the speaker. A central device installed at a bank teller, for example, by operating a terminal
The standard pattern registration operation and the speaker recognition operation can be performed on (for example, the speaker recognition device unit) in the same manner as described above.

For example, when registering a standard pattern, the user uses his / her terminal, for example, the switching unit 8 of 31-1.
And the feature extraction unit 4 operates the transmission / reception interface 34.
Set to switch to connect directly to -1. Next, when this user inputs predetermined designation information from the designation means 2, for example, (user D), this designation information is transmitted to the communication means 33-
1 to the central unit 32. Also, at this time,
The user D utters a predetermined specific word. This voice is input from the voice input unit 1, is output as a voice signal from the voice section detection unit 3, is converted into a feature amount by the feature extraction unit 4, and is transmitted to the central device 32 via the communication unit 33-1. To be done. As a result, the registration unit 6 of the central device 32 can register the transmitted feature amount (feature pattern) signal as a standard pattern in the speaker recognition information storage unit 5 in association with the transmitted designation information. it can.

In addition, in this speaker recognition system, when performing speaker recognition, the user has his own terminal, for example, 3
The switching unit 8 of 1-1 is switched to the speaker recognition unit 7 side. Next, this user uses the specifying unit 2 to specify predetermined information,
For example, when (user D) is input, this designation information is transmitted to the central device 32 via the communication means 33-1 and, as a result, the central device 32 uses this designation information as speaker recognition information. A standard pattern corresponding to the information, for example, the standard pattern of the user D is read from the speaker recognition information storage unit 5 and transmitted to the terminal. Then, the user D of the terminal
Utters a predetermined specific word. This voice is input from the voice input unit 1, is output as a voice signal from the voice section detection unit 3, is converted into a feature amount (feature pattern) by the feature extraction unit 4, and is provided to the speaker recognition unit 7.

As a result, the speaker recognition unit 7 of the terminal collates the speaker recognition information (that is, the standard pattern in the above example) transmitted from the central unit 32 with the characteristic pattern from the characteristic extraction unit 4, The similarity is calculated, it is determined whether this similarity is higher or lower than a predetermined reference value (threshold value) (that is, it is determined whether the user is a regular user), and this determination result is Transmit to central unit 32. In the central unit 32,
Based on the determination result from the terminal, it is determined whether or not to permit the application.

That is, when the result of the determination that the degree of similarity is low is transmitted, the central unit 32 determines that the user is not the authorized speaker himself D, and rejects the use by this user. On the other hand, when the determination result that the degree of similarity is high is transmitted, it is determined that the user is the regular speaker himself D, and the application by the user (for example, deposit / withdrawal,
Processing such as balance inquiry) is permitted.

Thus, in this speaker recognition system,
Using a terminal (for example, a personal computer) installed at the user's home or office, etc., the speaker is recognized, and the judgment result is a central device (for example, a speaker recognition device unit) installed at a bank or the like. To the central device,
After it is confirmed that the person is the person based on the result of this determination, applications such as deposit / withdrawal and balance inquiry can be used. That is, the user can use an application of a bank or the like by causing a terminal installed at home, a company, or the like to recognize a speaker without going to the bank or the like each time.

Further, in this configuration example, an existing personal computer (personal computer having a personal computer communication function) can be used as the terminal on the user side.

Further, in this configuration example, since the determination result from the speaker recognition section 7 is transmitted to the central unit 32 via the communication means (for example, telephone line or wireless), the communication means (telephone line (Or wireless or the like) or the communication environment is somewhat poor, the determination result signal is not easily affected, and therefore, the central device 32 determines whether the determination result signal is transmitted based on the transmitted determination result.
It is possible to correctly determine whether or not to allow the user to use the application. Further, the determination result signal has a very small amount of data, and the transmission time can be significantly shortened.

Further, in this configuration example, since the voice section detecting section 3, the feature extracting section 4, and the speaker recognizing section 7 are provided on the terminal side, the user conforms to the characteristics of his own voice. As described above, it is possible to manage the characteristics of the voice section detection unit 3, the characteristics of the feature extraction unit 4, and the like. For example, it is possible to adjust the sensitivity of voice section detection (threshold of voice volume) and the like according to the volume and sound quality of one's own voice.

In general, the speaker recognition unit 7 has a function of calculating the similarity between the characteristic pattern and the standard pattern and a function of determining whether the calculated similarity is higher or lower than a threshold value. However, these functions can be configured by one block, but can also be configured by separate blocks (software) as the similarity calculation unit and the determination unit.

In this case, for example, as shown in FIG. 5, the similarity calculation section 60 is provided on the terminal 31-1 side and the determination section 62 is provided on the central device 32 side to calculate the similarity degree of the terminal. The similarity between the characteristic pattern and the standard pattern calculated by the unit 60 is transmitted to the determination unit 62 of the central apparatus via the communication unit 33, and the determination unit 62 of the central apparatus is configured to determine the speaker. You can also Also in this case, as in the configuration examples of FIGS. 3 and 4, the speaker recognition information used in the similarity calculation in the terminal can be transferred from the central device to the terminal.

Alternatively, as shown in FIG. 6, a similarity calculation section 60 and a determination section 61 are provided on the terminal 31-1 side,
Further, on the side of the central device 32, separately from the determination unit 61 of the terminal,
The determination unit 62 is provided, and the similarity between the characteristic pattern calculated by the similarity calculation unit 60 of the terminal and the standard pattern is given to the determination unit 61 of the terminal 31-1 as the case may be, and the speaker on the terminal side. Or the determination unit 62 of the central device 32.
It can be configured to select whether or not to perform the speaker determination on the side of the central device. That is, in the configuration example of FIG. 6, in the terminal 31-1, the similarity calculation unit 60 and the determination unit 61 configure a first speaker recognition unit, and on the side of the central device 32, the similarity of the terminal is determined. The degree calculation unit 60 and the determination unit 62 of the central device constitute a second speaker recognition unit. Whether the first speaker recognition unit performs speaker recognition,
It is possible to select whether or not to perform speaker recognition in the second speaker recognition unit.

Also in the configuration example of FIG. 6, the speaker recognition information (standard pattern, etc.) is collectively managed only by the speaker recognition management means 10 of the central unit 32, and is processed only in the terminal. Also, although it can be transferred from the central device 32, the speaker recognition information different from the speaker recognition information (standard pattern etc.) set in the central device 32 is used.
If a standard pattern (such as a standard pattern) is also prepared in the terminal and speaker recognition is performed only within the terminal, the standard pattern is not transmitted from the central unit, and the speaker recognition information (standard Pattern) can also be used.

By the way, in general, a feature amount (feature pattern) of a voice uttered by a regular user for speaker recognition and a feature amount (standard pattern) of a voice registered in advance by the user himself / herself. Since there is a time lag between them, it is rare that the feature pattern perfectly matches the standard pattern even with the same speaker's voice, and normally,
Somewhat different. Therefore, it is necessary for the speaker recognition unit 7 of the terminal to set an appropriate threshold value for the determination of the similarity between the characteristic pattern and the standard pattern.

However, if the threshold value for this determination is set high, the accuracy of speaker recognition can be increased.
On the other hand, despite being the voice of the legitimate user,
This increases the probability that the user is judged not to be the legitimate user, making it difficult for the user to use. On the other hand, if the threshold value for this judgment is set low, the probability of erroneously recognizing the voice of another person other than the legitimate user as the voice of the legitimate user is high, and the information of the legitimate user is Increased risk of being stolen and misused by others. Therefore, it may be difficult to set the threshold value to a constant and appropriate value for speaker recognition.

The present invention is intended to further provide a speaker recognition system and a speaker recognition method which can satisfactorily solve such conflicting problems. In order to solve such problems, the present invention is provided. Then, the threshold value for determination, that is, the recognition accuracy is variably set depending on the case so that the speaker recognition is performed.

More specifically, this terminal (for example, a personal computer)
If you want to use the speaker recognition function installed in the terminal only inside this terminal (for example, if you use it for applications only inside the terminal such as starting up this terminal), for example, communication line etc. Since it is less likely to be stolen by others through the, the threshold for judgment is set low (decreasing recognition accuracy), and the sound of the legitimate user is judged to be the sound of the user himself. Increase the probability of

When the speaker recognition function installed in this terminal (for example, a personal computer) is used in association with or in cooperation with another device (for example, another terminal or central device) (for example, external communication) In such a case), there is a risk of being stolen by another person via a communication line, etc., so the threshold for judgment is set high (to increase the recognition accuracy), so that the legitimate user can Information from being stolen by others.

Note that such a function of the present invention (a function of changing the recognition accuracy) is as follows in the configuration example of FIGS.
For example, a plurality of threshold values are prepared in one terminal, for example, 31-1, and depending on the processing of this terminal (for example,
(Depending on whether this terminal processes only in this terminal or transmits / receives information to / from an external device),
This can be realized by the speaker recognition unit 7 selecting and using the optimum one from the plurality of threshold values.

Alternatively, in the configuration example of FIG. 3 and FIG.
Instead of preparing all the threshold values in the terminal, for example, for the threshold value for processing only in the terminal, prepare it in this terminal and process it with an external device. The threshold for use may be transmitted from the speaker recognition management means 10 of the central device, for example, when the terminal is communicatively connected to an external device (for example, the central device).

For example, when speaker recognition is performed at this terminal, the speaker recognition management means 10 of the central unit is connected to this terminal.
Therefore, when transferring the standard pattern for the user of this terminal, it is possible to transfer the threshold value for processing with the central device of this terminal together with this standard pattern as additional information. .

In this case, the speaker, for example user D,
For example, if the user (user D) is input from the terminal 31-1, it is transmitted to the speaker recognition management means 10 of the central apparatus via a communication means such as a telephone line or a radio, and thereby,
The speaker recognition management means 10 of the central apparatus returns the voice standard pattern corresponding to the speaker and the threshold value of the similarity degree as accompanying information to the terminal 31-1, and the terminal 31-1 speaks. When the person recognizing unit 7 determines whether or not the person is the person, the threshold value sent from the management unit 10 of the central apparatus 32 can be used. This allows the central unit 3
The recognition accuracy of the terminal 31-1 can be controlled by the second speaker recognition management means 10.

As described above, in the configuration examples of FIGS. 3 and 4, the threshold value is changed and used depending on the case (for example, in the case of processing only in the terminal, the threshold value of the similarity is Using a lower threshold and using a higher value as the threshold of similarity in the case of processing between external devices), for example, even if the same personal computer is used as a terminal, In the case of recognizing only in the personal computer, the speaker recognition can be realized by the gradual determination set in the personal computer and the strict determination when connecting to another device.

The function of changing the recognition accuracy is shown in FIG.
In the configuration example of, for example, the threshold value used in the determination unit 61 of the terminal is prepared in the terminal, and the threshold value used in the determination unit 62 of the central device 32 is in the central device 32. It can be realized by preparing.

That is, in the configuration shown in FIG. 6, when the user of the terminal intends to perform speaker recognition for an application only in this terminal, the similarity calculated by the similarity calculator 60 is calculated. An instruction is given to this terminal so that the determination section 61 in this terminal can make the determination.

In this case, when the user of this terminal generates a voice for speaker recognition, this voice feature amount (feature pattern) is converted into a standard pattern (for example, in this terminal) by the similarity calculation unit 60 of this terminal. Is calculated, and the similarity is given to the determination unit 61 of the speaker recognition unit 7 of this terminal. The determination unit 61 can perform speaker recognition by determining whether the similarity is high or low with respect to a predetermined threshold prepared in advance in this terminal.

On the other hand, when the user of the terminal wants to perform speaker recognition in order to use the application of the central device 32, the terminal device sends a predetermined message to the central device 32 (speaker recognition management means 10). Give instructions. As a result, the central device 32 sends, for example, a standard pattern to this terminal according to the instruction from the terminal. Next, when the user of this terminal generates a voice for speaker recognition, this voice feature
The (feature pattern) is a standard pattern (for example, a standard pattern transmitted from the central device) in the similarity calculation unit 60 of this terminal.
Is calculated, and this similarity is transmitted to the central device 32 and given to the determination unit 62 of the central device 32. The determination unit 62 of the central apparatus 32 determines whether or not the person is the person based on whether the similarity is high or low with respect to a predetermined threshold prepared in advance in the central apparatus 32.

As described above, in the configuration example of FIG. 6, when the application in the terminal is used, the determination unit 61 in the terminal is used.
In the case of using the application of the central device, the determination unit 62 of the central device recognizes the speaker.
(Determination) can be performed, and the determination criterion (threshold value) in the determination unit 61 of the terminal and the determination criterion (threshold value) in the determination unit 62 of the central device 32 are set to the terminal and the central device 32. Since they can be set independently (for example, the judgment unit 61 can set the judgment standard loosely (the threshold value is low) and the judgment unit 62 can set the judgment standard severely (the threshold value is high),
The recognition accuracy can be changed depending on the case. That is, similar to the configuration examples of FIGS. 3 and 4, for example, even when the same personal computer is used as a terminal, when only the personal computer is to be recognized, the gentle determination set in the personal computer is used. When connected to another device, it is possible to realize speaker recognition with strict judgment.

As described above, according to the present invention, for example, even when the same personal computer is used as a terminal, when only the personal computer is recognized, a gentle judgment set in the personal computer is used, and another personal computer is used. When connected to the device, it is possible to perform speaker recognition with strict judgment. That is, even if the same personal computer is used as a terminal, it is possible to realize speaker recognition in which the recognition accuracy can be changed depending on whether the personal computer is connected to another device or connected.

In each of the above configuration examples, the threshold value may be stored together with the speaker recognition information. For example, in the configuration example of FIGS. 3 and 4, for example, in the speaker recognition information storage unit 5 of the central device 32, as shown in FIG. May be stored. At this time, for example, the speaker recognition unit 7 on the terminal side determines the characteristic pattern of the voice of a speaker as
When the similarity between the characteristic pattern of this speaker and the standard pattern is obtained by collating with the standard pattern read from the speaker recognition information storage unit 5 of the central unit 32 and transmitted, this similarity is, for example, the above-mentioned standard. It is determined whether or not this speaker is the true person based on the accompanying information read from the speaker recognition information storage unit 5 of the central unit 32 and transmitted together with the pattern, that is, whether the speaker is higher or lower than the threshold value. You can

In the configuration examples of FIGS. 3 to 5, for example, the threshold value stored in the speaker recognition information storage unit 5 as accompanying information as shown in FIG. 7 is registered as shown in FIG. A different threshold value can be set for each standard pattern that is set, or the same (constant) threshold value can be set for all registered standard patterns. When the same (fixed) threshold is used for all registered standard patterns, only one threshold is stored in the speaker recognition information storage unit 5 (one memory). It can be stored and used in common.
Similarly, in the configuration example of FIG. 6 as well, for example, as threshold values set on the terminal side and threshold values set on the central device side, for each registered standard pattern. You can set different thresholds,
It is also possible to set the same (constant) threshold value for all registered standard patterns.

In addition, in each of the above-mentioned configuration examples, the user can use a terminal other than his / her own terminal, but on the other hand, this increases the possibility that another person will use the terminal without his / her knowledge. It will also happen. In speaker recognition,
Especially important for determining whether or not the person is the speaker recognition information (especially the standard pattern of voice), and if this is maliciously rewritten, it becomes impossible for the person to use it thereafter. Alternatively, the information of the person concerned may be misused by another person.

Therefore, it is possible to configure the speaker recognition system so that the speaker recognition information is changed and corrected only with the information from the determined terminal. For example, the home terminal or the like may be designated, and the standard voice pattern may be rewritten or updated only from the home terminal of the user. As a result, it is possible to prevent others from changing or modifying (rewriting) the speaker recognition information.

Further, in each of the above configuration examples, it is possible to manage the speaker recognition information in the speaker recognition information storage unit 5 by a flag. For example, in order to distinguish the currently used speaker recognition information (standard pattern) from the unused speaker recognition information (standard pattern), the speaker recognition information (standard pattern) is set to “0”. Alternatively, a flag of "1" can be set to manage this. In this case, this flag management allows one speaker recognition information (standard pattern) to be supplied to only one place at a time, and the speaker recognition information (standard pattern) in use is used.
Can also be made unavailable to others. This prevents others from using it when the person is using it, and when the person uses it when another person is using it, this person uses someone's own voice standard pattern. You can understand that and take measures quickly.

Further, for example, in the configuration examples of FIGS.
Since the voice section detection unit 3, the feature extraction unit 4, and the similarity calculation unit 60 are provided on the terminal side, the user can adjust the characteristics of the voice section detection unit 3 so as to match the characteristics of his or her voice. Although it has the advantage that the characteristics of the feature extraction unit 4 can be managed, on the other hand, it is possible to perform processing such as voice section detection, feature extraction, similarity calculation, and speaker determination (threshold determination) processing. However, because the terminal side and the central device side are distributed and sometimes distributed, the administrator of the central device side does not recognize the authorized user as an authorized user, or When a so-called erroneous recognition occurs, such as recognizing a user as a legitimate user, the cause may not be collectively managed only by the central device side. For example, it is not possible for the central device alone to determine whether this is due to the detection of the voice section of the terminal, feature extraction, similarity calculation, or the speaker determination (threshold value) of the central device. is there.

The cause of erroneous recognition is, for example, that the beginning or end of the word generated by the speaker is weak, the voice section is not normally detected, and this part is missing in the characteristic pattern. In some cases, the speaker may not be aware of this, and the same result may be obtained no matter how many times he or she repeats it.

In order to avoid such a problem, for example, in each of the above configuration examples, for example, the central device 32 (for example, the speaker recognition management means 10) provides predetermined information to a terminal, for example, 31-1, and the terminal 31 The -1 voice section detection unit 3 and the feature extraction section 4 can also perform voice section detection and feature amount conversion based on the information provided from the central device 32.

For example, from the central device 32 side, for example, sensitivity instruction information for voice section detection can be provided as information.

In this case, on the terminal side, when the user initially utters a voice and this voice is detected in the voice section,
For example, data of about 0.5 seconds is added before and after it is determined to be a voice section, and the voice signal in the voice section including this data is stored as it is in a file (not shown) on the terminal side. In this way, after the voice signal of the user is stored in the file on the terminal side, this is converted into a feature amount (feature pattern), the similarity calculation unit 60 calculates the similarity, and the central device determines The section 62 determines the speaker depending on whether the similarity is higher or lower than the threshold value. As a result, when it is determined that the user is not the user himself, the central device 32 side, for example, the sensitivity of voice section detection stored as accompanying information in the speaker recognition information storage unit 5 (voice volume The threshold value, etc.) is increased, and this is provided (transmitted) to the terminal side as information, and the terminal side is made to perform voice section detection again for the voice signal stored in the file. In addition, the sensitivity of voice section detection stored as accompanying information in the speaker recognition information storage unit 5 is lowered, and this is transferred to the terminal side.
It is provided (transmitted) as information, and the terminal section is made to perform the voice section detection again for the voice signal stored in the file.

In this way, the speaker recognition is performed for each of the case where the sensitivity of the voice section detection is increased and the case where the sensitivity is decreased, and if it is determined that the correct speaker is detected in either case,
This user can be recognized as the correct speaker. In this way, by improving the sensitivity of voice section detection, for example, when the voice of the speaker is small, the situation where the voice section is not correctly detected (the situation of causing omission of detection) has been improved. Also, by lowering the sensitivity of voice section detection, for example, when noise occurs before and after the voice of the speaker, the voice section detection unit may detect a voice longer than it actually is. I was able to improve it.

In the above-mentioned example, as the information provided from the central device 32 side to the terminal side, the sensitivity of voice section detection is taken as an example, but the sampling frequency for feature amount conversion can also be provided to the terminal side as information. In this case, on the terminal side, the sampling frequency for feature amount conversion can be changed by the information from the central unit 32. Further, various information other than the above can be provided from the central device 32 side to the terminal side.

As described above, by providing the terminal with the predetermined information from the central unit, the central unit manages the voice section detection for speaker recognition, the feature extraction, the speaker recognition, etc. Can be controlled.

In the above description of each structural example, the case where the speaker recognition is performed has been described, but new registration, change and update of the standard pattern can be similarly performed from the terminal side. If the central device has a function for automatically updating the standard pattern according to the data used for speaker recognition, the central device automatically performs the operation without any operation from the terminal side. The standard pattern can be updated.

Further, in each of the above-described configuration examples, the degree of similarity between the characteristic pattern and the standard pattern is regarded as the degree of similarity between them, but it may be regarded as the degree of difference between them. When the similarity is regarded as the degree of dissimilarity, the method of determination is opposite to that when it is regarded as the degree of similarity, and when the similarity (dissimilarity) is higher than the threshold value, it is determined that the speaker is not the person himself. , If the degree of similarity (degree of difference) is lower than the threshold value, it is determined that the speaker is the person himself.

Further, in the above-mentioned example, when the standard pattern is transferred from the central unit, the threshold value of the similarity is transferred together with the standard pattern, but they may be transferred separately (at different timings). .

Further, in each of the above-described configuration examples, the case where information is transmitted and received between one terminal and the central unit 32 has been described as an example. However, for example, one user uses two terminals.
In the case of having a table, the standard pattern registration function is given to the terminal 31-1, for example, and the speaker recognition function is given to the terminal 31-2, for example, and the terminal 31-1 registers information such as the standard pattern. And sends information such as the standard pattern being used from the terminal 31-1 to the terminal 31-2,
It can also be used for recognition at -2.

In each of the above configuration examples, the feature extraction unit 4 is provided after the voice section detection unit 3, but the feature extraction unit 4 is provided before the voice section detection unit 3 as necessary. It is also possible to configure so that it can be performed.

[0086]

As described above, according to the present invention, according to claim 1 or 請
According to the invention described in claim 7 , for example, a terminal installed in the user's home or office is used to perform speaker recognition, and an application such as deposit / withdrawal or balance inquiry of a bank is used. be able to. Also, for example,
Even if you use the same personal computer as a terminal, if you recognize only in that personal computer, the gentle judgment set in the personal computer, and when connecting to another device,
Speaker recognition can be realized with strict judgment.

[Brief description of drawings]

FIG. 1 is a diagram showing a configuration example of a general speaker recognition system.

FIG. 2 is a diagram showing a configuration example of a speaker recognition information storage unit.

FIG. 3 is a diagram showing a configuration example of a speaker recognition system according to the present invention.

FIG. 4 is a diagram showing a specific example of the speaker recognition system of FIG.

FIG. 5 is a diagram showing another configuration example of the speaker recognition system according to the present invention.

FIG. 6 is a diagram showing another configuration example of the speaker recognition system according to the present invention.

FIG. 7 is a diagram showing a configuration example of a speaker recognition information storage unit.

[Explanation of symbols]

1 Voice input means 2 instruction means 3 Voice section detector 4 Feature extraction unit 5 Speaker recognition information storage 6 registration department 7 Speaker recognition unit 8 switching unit 10 Speaker recognition management means 31 terminals 32 Central unit 33 Communication means 60 Similarity calculation section 61 Judgment unit 62 Judgment unit

Claims (7)

(57) [Claims] 1. At least one terminal and a central unit,
The terminal is provided so that information can be transmitted and received, and the terminal has a voice input means for inputting a voice of a speaker into a voice signal, a feature extracting means for extracting a feature amount of the voice signal, and a feature of the voice of the speaker. Amount and a similarity calculation means for calculating the similarity between the voice feature amount as the speaker recognition information, and the central device, and a speaker recognition management means for managing the speaker recognition information. , Determination means for determining the speaker based on the similarity from the similarity calculation means of the terminal, and the speaker recognition information used for the similarity calculation in the terminal is transferred from the central device to the terminal. The speaker recognition system, wherein the similarity calculated by the similarity calculation means of the terminal is transferred from the terminal to the central device. 2. At least one terminal and a central unit,
The terminal is provided so that information can be transmitted and received, and the terminal has a voice input means for inputting a voice of a speaker into a voice signal, a feature extracting means for extracting a feature amount of the voice signal, and a feature of the voice of the speaker. The similarity calculation means for calculating the similarity between the volume and the voice feature amount as the speaker recognition information, and the first determination means for determining the speaker based on the similarity from the similarity calculation means are provided. Also, the central device is provided with a second judging means for judging a speaker based on the similarity from the similarity calculating means of the terminal, and the second judging means calculates the speaker by the similarity calculating means of the terminal. The speaker recognition system, wherein the degree of similarity is given to the first determining means of the terminal or transferred from the terminal to the second determining means of the central device. 3. The speaker recognition system according to claim 1 or 2, wherein predetermined information is transferred from the central device to the terminal, and the feature extraction means provided in the terminal is a central device. A speaker recognition system characterized by converting an input voice into a feature amount based on information provided from a device. 4. The speaker recognition system according to claim 1, wherein the change or modification of the speaker recognition information managed by the speaker recognition management means is regarded as information from a determined terminal. Features speaker recognition system. 5. The method of claim 1 Symbol placement of speaker recognition system, one speaker recognition information from the central unit is configured such that it can be supplied simultaneously only to only one terminal A speaker recognition system characterized by the above. 6. At least one terminal and a central unit,
The terminal is capable of transmitting and receiving information, and when the voice of the speaker is input, the feature amount of the voice signal is extracted in the terminal, and the feature amount of the voice of the speaker and the speaker recognition information are used. It is designed to calculate the similarity with the voice feature amount of
At this time, the speaker recognition information used for similarity calculation in the terminal is transferred from the central device to the terminal, and the similarity calculated in the terminal is transferred from the terminal to the central device. A speaker recognition method characterized by making a speaker determination based on the transferred similarity. 7. At least one terminal and a central unit,
It is provided so that information can be sent and received.
When the voice of the speaker is input, the
Extracted as a feature amount of the speaker's voice and speaker recognition information
It is designed to calculate the similarity with the voice feature amount of
The similarity calculated at the terminal is the
Be used for centralized equipment or transferred from the terminal to a central unit.
It is sent to be used for the determination of the speaker in the central device, and the determination of the speaker is made according to whether the terminal processes only in the terminal or transmits / receives information to / from an external device. Recognition method characterized by changing the threshold of the speaker.