JP2001265385A - Speaker recognizing device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a speaker recognizing device capable of updating the registered voice at a proper timing and capable of securing the safety at the time of performing the updating of the registered voice. SOLUTION: The voice data of a speaker which are given via an input part 1 are compared with registered voice data becoming a reference which are stored in a voice data storage part 2 in a voice collating part 4 and a collation score indicating the degree of similarity is obtained in the part. When the collation score is larger than a threshold, this device judges that the voice data are of the person to move to a prescribe service. At this time, when, although the data are judged to be the voice data of the person, the collation score is low and the difference between the present data and the last data is large, a necessity of update judging part 7 judges that there is the necessity of the update to deliver the judged result to a voice registering and managing part 5. The part 5 updates the registered voice data to the voice data inputted this time on condition that the voice data are judged that there is the necessity of the update and, also, there is the agreement of the update from a speaker.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speaker recognition device.

[0002]

2. Description of the Related Art As one type of personal identification performed by a computer, specific identification information such as a personal identification number or a keyword that has been registered in advance is input, and if the identification information matches the registered information, the user is identified. Is generally performed. However, in such a system, if the designated information is known or found by another person, the designated information may be stolen and misused.

[0003] Therefore, physical characteristics (fingerprint, voiceprint, voice,
There is a technology for confirming the identity of a person using pupils and the like, and one of them is speaker recognition such as speaker verification and speaker identification. However, in the case of speaker recognition, one of the problems is to prevent “decrease in matching accuracy due to aging of the voice”, and in order to solve such a problem, updating the registered voice at an appropriate time ( Re-registration). Conventional techniques for performing such updating include the following.

[0004] In other words, there is an apparatus that automatically updates (re-registers) a registered voice when the input voice is determined to be the person (Japanese Patent Laid-Open No. 57-013493). According to the invention disclosed in this publication, the registered voice can be updated in accordance with the voice of the person, which changes slightly from moment to moment, so that there is an effect that it becomes strong against aging.

[0005] However, in the case of speaker recognition, it is actually difficult to make the accuracy 100% perfect, so that there is a possibility that another person may be mistakenly recognized as the person.
In such a case, the registered voice is updated based on the voice of the other person, and thereafter, not only the authorized user is not authenticated as the user, but also the authorized user (authorized person) at the time of voice update. There is a problem that the position of the person is taken over.

[0006] Even if the user uses the voice, if the voice is unavoidably updated with a voice different from the normal voice such as a cold voice, the user may not be correctly authenticated after the cold has returned to the normal voice. is there.

In order to solve such a problem, for example,
There is an invention in which an update process is actually performed after a confirmation is made directly to a legitimate user at the time of updating the registered voice (for example, by calling a connection destination of the legitimate user) (Japanese Patent Laid-Open No. 9-127)
975). However, in the invention disclosed in this publication, the processing is complicated because the confirmation means contacts the authorized user each time.

[0008] Furthermore, the timing for updating the registered voice is usually the one that is regularly updated. On the other hand, as the number of updates increases, the possibility of unauthorized registration by others increases, and the time (period) at which the sound changes over time varies from person to person. If the setting of the elapsed time is not set properly and the update period is longer, the voice of the authorized user may change and become unrecognizable.On the other hand, if the update period is too short, the number of updates may be reduced. Due to the increase, the chance of unauthorized registration of the registered voice by others is increased, which is not preferable in any case.

Regarding the update of the registered voice according to the present invention,
An object of the present invention is to provide a speaker recognition device that can update a registered voice at an appropriate timing and can ensure security at the time of updating.

[0010]

A speaker recognition apparatus according to the present invention includes a voice input unit, a voice information storage unit for holding registered voice information as a reference for voice recognition, and a voice information storage unit. It is assumed that the speaker recognition device includes a voice collating unit that determines whether or not the voice information input from the voice input unit is from a legitimate speaker based on the registered voice information. And determining whether the registered voice information stored in the voice information storage means needs to be updated, based on the verification score information indicating the authenticity of the speaker obtained in the verification processing in the voice verification means. Means, and updating means for updating the registered voice information based on the result of the judgment by the judging means.

Here, the "collation score information" may be specified by specific numerical values as shown in the embodiment, as long as it indicates "authority of a normal speaker", or may be information of an individual. May have a range such as "high", "medium", and "low".

If the registered registered voice information is very similar to the input voice information that has been uttered this time, it can be said that there is no aging change and the need for updating is low. On the other hand, if the voice information obtained by uttering the person and the registered voice information are not very similar, aging occurs, and the person himself may not be recognized as it is. There is a need to. Therefore, by determining whether or not it is necessary to update the registered voice information based on the collation score information, the registered voice information can be updated at an appropriate timing, and it is resistant to aging. Further, by performing the update periodically in a meaningless manner, it is possible to prevent the update process from being performed based on the utterance of another person.

[0013] In addition, if correspondence is made between each component and the embodiment, the voice input means is provided in the input section 1, the voice information storage means is provided in the voice data storage section 2, and the determination means is provided in the update necessity determination section 7. , Updating means correspond to the voice registration / management unit 5, respectively.

[0014] In a preferred embodiment of the present invention, the updating means determines that the determination result based on the collation score information in the determining means needs to be updated, and that the consent information of the updating process from the speaker is required. The registered voice information is updated on the condition that there is a certain condition.
In this way, when the voice of the user is inappropriate for updating, for example, in the case of a cold voice, updating can be avoided.

Such functions need not necessarily be provided, and when the determination means determines that updating is necessary, the updating process may be executed as it is. In other words, if a person is registered with a cold voice, etc., the registered voice information will be updated again based on the normal voice by recognizing the person when returning to the normal voice next time. Recognition is no problem. Furthermore, the update determination condition may be added in addition to the above conditions. As an example, by performing the comparison more strictly than usual, it is possible to avoid the danger associated with updating the registered voice.
To perform strict collation, (a) Use with password information collation In addition to voice matching, confirm knowledge that is not known to anyone other than the user, such as passwords and passwords; (b) Change thresholds for collation scores (C) Increasing the number of times of collation Two or three repetitions of collation are performed, and the registered voice can be updated only when the identity is determined by all the collations. is there.

Further, the update weight of the registered voice may be set based on the variation of the feature amount of the utterance at the time of updating, and new registered voice information may be generated in consideration of the registered voice information before updating. Then, even if it is updated by another person's voice, there is a possibility that a legitimate user is collated by retaining the voice information (feature amount) of the person before the update.

The speaker recognition according to the present invention includes not only the speaker verification described in the embodiment but also the speaker identification. That is, in the speaker verification, the registered voice information of one speaker specified in advance is compared with the input voice information, and it is determined whether or not the speakers of each voice are the same (one-to-one recognition). Therefore, in order to identify one speaker, speaker identification information is input. on the other hand,
The speaker identification compares the registered voice information of all speakers registered in advance with the input voice information to identify which registered voice is the same as the speaker of the input voice ( One-to-many recognition).

Each means constituting the speaker recognition device according to the present invention can be realized by a dedicated hardware circuit, or can be realized by a programmed computer.

[0019]

FIG. 1 shows a preferred embodiment of the present invention. In this system, information such as a voice generated by a speaker is fetched into the apparatus via the input unit 1, and based on dictionary information such as registered voice data indicated by the speaker identification information stored in the voice data storage unit 2. The present invention is applied to a case where the speaker verification is performed to determine whether or not the speaker in use is the principal, and if the identity is confirmed, a predetermined service is executed. Then, based on the input voice data, it is determined whether or not it is necessary to update the registered voice data registered in the voice data storage unit 2 serving as a reference / dictionary data at the time of recognition / collation processing. Performs update processing.
Thus, even if the user's voice changes over time, a high recognition rate can be maintained in response thereto. The specific configuration is as follows.

First, the voice data storage unit 2 is a database for storing speaker holding information and registered voice data in association with each other. It should be noted that, in addition to the registered voice data (feature amount), the recognition result of the collation keyword voice by the input understanding unit 3 to be described later is also held in association with the registered voice data. Updates can also be prevented. For example, if only the registered person knows the content of the keyword, the user who tried to update the registered voice with a different keyword may not be the person himself, in which case the update of the registered voice is stopped. be able to.

Then, as shown in FIG. 1, the input section 1 takes in the voice uttered by the speaker into the apparatus. The input unit 1 captures voice acquired using, for example, a microphone or a telephone. The audio captured by the input unit 1 into the device includes:
There are collation keywords, speaker identification information (such as account number), and answers to system questions. In other words, in the present embodiment, the input of the speaker identification information also uses the voice, so that the input unit 1 also serves as the voice input unit and the speaker identification information input unit in the speaker verification device.

Note that the speaker identification information, which is an utterance other than the collation keyword, and the answer to the question from the system need not necessarily be input by voice, but may be input by, for example, a telephone push button. Also, the input unit 1 needs to have at least a function corresponding to voice input, but may further have a function corresponding to input other than sound. As an example, in the case of an information providing terminal or the like, when the speaker identification information is input from a console such as a numeric keypad, it has an input function corresponding thereto.

The information input from the input unit 1 is provided to an input understanding unit 3, a voice collating unit 4, and a voice registration / management unit 5. First, the input understanding unit 3 recognizes the character information of the voice represented by the voice waveform data received from the input unit 1, and outputs the character string to a predetermined processing unit. More specifically, pattern matching is performed between a feature amount sequence obtained by frequency analysis of a speech waveform and a feature amount sequence of a prepared phrase to be recognized. Here, each number of “0 to 9” required when inputting a personal identification number and other various numbers, and each word such as “Yes” or “No” assumed as an answer of the user (speaker) are described. A feature amount sequence is prepared for each of the words to be recognized.

Of course, when a text is used as an input other than the voice, for example, the same result as in the case of voice recognition can be obtained by using a normalized text as the input content as a recognition result. It is also possible to determine whether the input matching keyword is a predetermined one or matches a currently registered keyword based on the result of the input understanding (speech recognition result). Note that the matching process using specific feature amounts
Since what has been conventionally performed can be applied as it is, a detailed description with a specific example is omitted.

The voice collating unit 4 further includes registered voice data indicated by the speaker identification information registered in the voice data storage unit 2,
The voice data of the speaker input from the input unit 1 or the input understanding unit 3 is compared to determine whether the input voice and the registered voice are the same.

As an example, a collation score for making the determination is calculated. The collation score takes a value from -1.0 to ten 1.0. A positive sign indicates that the speaker of the input speech is the speaker of the registered speech (somewhat), and a negative sign indicates that the speaker is rather another person. Further, the absolute value indicates the likelihood of the content (the person or another person) indicated by each code. Therefore,
It can be said that the greater the positive value (the closer to 1.0), the higher the identity. Then, the obtained matching score is compared with a predetermined threshold value, and if the matching score is larger than the threshold value, it is determined that the user is the person (if the value is equal to or smaller than the threshold value, another person). The collation result is sent to the response generation unit 6 and the update necessity determination unit 7
Give to.

The response generation unit 6 determines whether the input comprehension results obtained by the input comprehension unit 3 and the voice verification results performed by the voice verification unit 4 and whether the registered voice update needs to be performed by the update necessity determination unit 7 described below. The contents to be uttered by the system are determined based on the result. Specifically, the contents of the voice guidance issued by the system and the contents uttered in accordance with the verification result are all recorded in advance, and an appropriate value is determined in accordance with the input understanding result, the verification result, and the update necessity determination result at that time. Select the utterance content. And
The selected utterance content (utterance sentence) is provided to the output unit 8.

The output unit 8 converts the generated utterance sentence into a voice and outputs it using a speaker, headphones, or a telephone receiver speaker. In order to convert an utterance sentence into a voice, various methods described below are adopted. * Recording / editing method Record the voice of the utterance contents read out for the whole or each part of the utterance contents that may be generated, and output the sound by playing back the recording file corresponding to the utterance contents when outputting. Is what you do.

* Text-to-speech synthesis system A speech synthesis engine existing as an existing technology is used.
That is, when a character string representing the utterance content is given as an input, a reading voice of the character string can be output.

* Combination method of recording and editing and text-to-speech synthesis Of the utterance contents, a part where a recording part is prepared is output as a utterance by the recording and editing method, and a part where there is no corresponding recording part is uttered by text-to-speech synthesis. Output.

The output section 8 is not limited to audio output as described above. As an output other than voice, for example, text can be used. In this case, the output unit 8 can display the utterance content on the PC display screen.

The structure of each unit for confirming the identity in accordance with the input from the input unit 1 and outputting a predetermined message from the output unit 8 can be basically realized by a conventional one. Omitted.

Here, in the present invention, the update necessity judging section 7 is provided, and it is judged whether or not to constitute speech data as a reference for recognition based on the collation result in the speech collating section 4. . That is, the update necessity determination unit 7
Aging of the voice of the user is detected, and the necessity of updating the registered voice is determined based on the result. The secular change of the voice is detected by referring to the collation score and the feature amounts of the input voice and the registered voice (details will be described later).

When the update necessity determination unit 7 determines that the update is necessary, the determination result is given to the voice registration / management unit 5 and the response generation unit 6. Response generator 6
Outputs a message prompting the user to confirm whether the update can be performed. If the user who has received the message consents to the update, the consent information is given to the voice registration / management unit 5 (input Provided directly via the unit 1 or via the input understanding unit 3).

When the user (speaker) agrees to the renewal registration, the voice registration / management unit 5 associates the voice data in which the speaker uttered the collation keyword with the speaker identification information, and stores the voice data in the voice data storage unit 2. Register with. As a result, the content of the data stored in the audio data storage unit 2 is updated, and the data content is updated according to aging.

Next, the overall operation of this system is shown in FIG.
The detailed functions of the respective processing units will be described as necessary, while using the flowchart shown in FIG. In addition, in order to facilitate understanding of each function, a description will be given with reference to specific examples as appropriate. As a specific example to be cited, consider the following speaker verification device. In other words, voice input / output by telephone is performed, and a 9-digit account number is used as a speaker identification number, and a 4-digit password is used as a matching keyword. Then, speaker verification is performed based on the verification keyword input by voice (after identity verification, a predetermined service and other processes are performed). Furthermore, based on the result of the speaker verification, it is determined whether or not the registered voice data serving as a reference for verification has been updated, and an update process is performed as necessary.

First, speaker identification information input processing is executed (ST1). That is, the user (speaker) is prompted to input speaker identification information. Specifically, the response generation unit 6
The utterance “Thank you for calling. First, tell us your account number” is extracted from the utterance contents stored in advance, and the utterance is output by the output unit 8. Upon receiving this, the user speaks the account number (for example, "596384107") by voice or inputs the account number using a push button on the telephone.

In the case of voice input, the input unit 1 captures the utterance of the user as a voice waveform into the device and passes it to the input understanding unit 3, where the voice waveform data is converted into a phoneme feature quantity of a number prepared in advance. By comparing with the series,
Recognize the uttered digit string. In the case of a push button input, the input unit 1 captures a tone signal corresponding to a numeric string into the device, passes the tone signal to an input understanding unit 3, and converts the input tone signal into a previously prepared numeric signal waveform. Recognize the input digit string by comparing.

Next, a matching keyword input process is executed (ST2). That is, the response generation unit 6 extracts an utterance “Please tell us your collation keyword” from the utterance content prepared in advance, and the output unit 8 outputs the content as a voice. Since the user receiving this utters the collation keyword (personal identification number), the voice data of the uttered collation keyword (for example, "410
7 ”) is input to the device by the input unit 1 and passed to the voice collating unit 4.

Next, the process proceeds to speaker verification processing. First,
A collation score is calculated (ST3). That is, the voice collating unit 4 accesses the voice data storage unit 2 using the recognition result of the account number (speaker specific coasting information) given from the input understanding unit 3 as a key, and acquires the corresponding registered voice data. That is, since the voice data storage unit 2 has a data structure as shown in FIG. 4, "Yonichizenna" (voice data when the user generates 4107) corresponding to "596384107" To get. This registered voice data
The data may be audio data itself, or a feature amount extracted therefrom. Then, the characteristic amount of the acquired registered audio data is compared with the characteristic amount of the input audio data, and a matching score (−1.0
値 +1.0). In addition, as a recognition / collation algorithm for calculating the collation score, various conventional ones can be applied.

Further, the voice collating unit 4 determines whether the speaker of the registered voice and the speaker of the input voice are the same based on the calculated collation score value, ie, whether the user is the registrant himself or herself. Is determined. Specifically, it is determined whether or not the obtained matching score is larger than a threshold (ST4). Here, the threshold value is set to 0, and the determination is made simply based on the sign of the collation score value. That is, if the collation score is a positive value, it is determined that the person is the person, and if the collation score is a negative value, the person is determined to be another person. That is, steps 3 and 4 function as the voice collating unit 4.

Then, the result of the determination is passed to the response generation unit 6, and if the result of the branch determination in step 4 is No, that is, if it is determined to be another person, the response generation unit 6 returns to " As a result of the collation, the identity could not be confirmed. Sorry, the customer cannot use this service. ", And the output unit 8 outputs the content as audio. Thereafter, the telephone line is disconnected to end the processing (ST5).

On the other hand, if the branch determination in step 4 is Yes, that is, if the result of the determination is that of the user, the user is notified that the user has been authenticated before determining the necessity of updating the registered voice ( ST6). That is, the response generation unit 6
From the utterance stored in advance, an utterance such as "the collation has been confirmed as the result of collation" is extracted, and the output unit 8 outputs the content as a voice. Thereafter, the process proceeds to step 7 and proceeds to the process of determining the necessity of the voice update.

Next, a process for determining the necessity of updating the registered voice is executed (ST7). That is, the update necessity determination unit 7
Based on the value of the calculated matching score, it is determined whether there is a need to update the registered voice. Here, it can be determined that the registered voice needs to be updated simply when the matching score value is less than a certain threshold. In other words, if the verification score is a positive value and the person is judged to be the person, but the numerical value is small, it is estimated that the voice of the person has become different from the registered voice data due to aging, and if the verification score value is less than a certain threshold, It is determined that there is a need for updating.

As a more complicated judgment method, the history of the collation score value at the time of the past collation and the voice data of the past collation keyword itself are stored and registered by referring to the history information of each case. A method of determining the necessity of the voice update can be adopted. For example, as shown in the flowchart of FIG. 5, the previous and current collation score values are obtained (ST21), the current score is subtracted from the previous score, and the obtained value is set to D (ST22). When the obtained difference D is larger than a certain threshold value, it is determined that aging has occurred and it is determined that updating is necessary. When the difference D is small, it is determined that updating is unnecessary (ST23 to ST2).
5).

As a result, as shown in FIG.
The value of the matching score obtained at the time of the third use is S1, S2,
In the case of S3, the value of S1-S2 is small in the second use, so that it is determined that the update is unnecessary, and in the third use, S3 is used.
Since the value of -S2 is large, it is determined that updating is necessary.

The previous collation score is, for example, as shown in FIG.
As shown in the data structure shown in FIG. 1, the voice data storage unit 2 can store the speaker identification information and the registered voice data in association with each other. The registration can be dealt with by, for example, when the update necessity determining unit 7 determines the necessity of the current update, by registering it in the corresponding storage area for the next determination.

Further, instead of simply using the difference from the previous time as described above, the average value of the collation scores may be obtained, and it may be determined that the update is necessary when the following conditions are satisfied. . Absolute value of (current collation score value−average collation score value)> threshold value It is determined whether or not the update necessity determination result obtained as described above needs to be updated (ST8). If it is determined that there is no need for updating, the current speaker verification process is terminated. After that, the process proceeds to a process depending on a normal application or task.

On the other hand, if it is determined that there is a need to update the registered voice, the process proceeds to step 9, where the update is recommended and the user is confirmed (ST9). That is, the matching score of this time is low or greatly changed, and the similarity to the registered voice data may be low because the reason that the necessity of updating is determined is that the user happens to have a cold or the like. . In such a case, if the registered voice data is automatically updated, the user may not be recognized as the person himself / herself the next time the cold is corrected and the voice returns to the normal voice. The person knows best whether the generated voice is a normal voice for the person or not. Therefore, by obtaining the consent of the update to the person, update based on incorrect information is suppressed. Specifically, the response generation unit 6 extracts the following utterance from the utterance stored in advance, and the output unit outputs the content as a voice.

[We recommend that you update the registered voice. If you do not update the registered voice, it may not be possible to perform correct verification in the future. However, if your voice is currently out of order, such as when you have a cold, Please do not update ...
Do you want to update your registered voice? Upon receiving this, the user makes an utterance indicating affirmation or denial such as “yes” or “no”, so that the utterance is received by the input unit 1 and the utterance is taken into the apparatus as voice waveform data. Pass to input understanding unit 3. The input understanding unit 3 determines whether the input is “positive” or “negative” from the audio waveform data, and passes the result to the audio registration / management unit 5. The input method of the user's answer is
Not only voice but also push button input or the like may be used.

In this case, for example, an utterance of the system such as "Press the push button 1 when updating the voice, and press the push button 2 when not updating" is added to the utterance and output to the user. You can be prompted to respond with a push button input. Then, the input tone signal is recognized by the input understanding unit 3 in the same manner as in the case of the push button input of the account number shown in step 1.

Then, the voice registration / management unit 5 determines whether or not the user has agreed to update the registered voice (ST10), and when the user has agreed (recognition result of the user answer =
If "yes", the process proceeds to step 11 to update the voice data. That is, the voice registration / management unit 5 registers the voice data input in step 2 as new registered voice data in the corresponding storage area of the voice data storage unit 2. This registration may be overwritten on the currently registered audio data, or may be updated with the average value of the feature amounts of the previously registered audio data and the input audio data. You can also. If the user does not agree with the voice update, the process of speaker verification ends.

FIG. 8 shows a second embodiment of the present invention. Since the block configuration is the same as that of FIG. 1, the function of the corresponding processing unit will be described while describing its operation based on a flowchart.

The second embodiment is basically the same as FIGS. 3 and 4 of the first embodiment, and improves the security at the time of updating. Specifically, the processing functions of the flowchart shown in FIG. 3 are the same in the present embodiment (the functions of each processing unit are also the same in that respect). In the first embodiment, when the user agrees to the update in the branch determination in step 10 in the flowchart shown in FIG. 4, the update process is immediately performed. In the present embodiment, additional collation is performed. This ensures that you are who you are. Further, even when updating, by setting weights, more accurate registered voice data is created, and the recognition rate in speaker verification performed later is increased. In other words, the function for executing such processing is added to the registration update / management unit 5.

Then, specifically, as shown in FIG.
When the user accepts the update (Yes in step 10)
Then, the processing after step 31 is performed. That is, first, the number of utterances / threshold for additional collation is determined (ST31). This processing is also performed by the voice registration / update processing unit. The specific processing of this step is as shown in FIG. As shown in the drawing, the current collation score is referred to, and it is determined whether or not it is smaller than the additional collation count determination threshold value (ST41 to ST43).

If it is smaller than the threshold, that is, if it is determined that the person is the person and the similarity with the registered voice data is low, the process proceeds to step 44, where the number of times of additional collation is set to 2, and the collation threshold is set to a strict value. Is set to On the other hand, if the current matching score is equal to or greater than the additional matching number determination threshold,
Proceeding to step 45, the number of additional matches is set to one,
The collation threshold is set to a normal value.

That is, according to the present embodiment, the number of utterances at the time of additional collation and the collation threshold are set according to the collation score (the current score) when the person is determined to be the person. At this time, if the collation score is low, the number of utterances and the threshold value are set higher for more strict checking. With this, when it is determined that the user is the person, but there is a possibility that the person is not the person (the personality is low), the determination is made strictly, and the security is improved by performing the update process when the person is still verified.

In accordance with the above settings, the user is additionally verified (ST32). That is, the user is made to speak again and speaker verification is performed. Then, as a result of the additional collation, it is determined whether or not the identity has been confirmed (ST33). If the conditions are not satisfied, the update process is not performed this time. As a result, the update registration of the voice data can be reliably performed based on the utterance of the person, and the security is improved.

On the other hand, as a result of the additional collation, if it is confirmed that the user is the person, the process proceeds to step 34, where the update weight is set. Here, a weight is set for determining a mixture ratio between the feature amount of the existing registered voice and the feature amount of the newly input uttered voice in the updated registered voice. Then, specifically, the flowchart shown in FIG. 10 is performed.

First, the current matching score when the person is determined to be the person and the score at the time of additional matching are obtained, and the difference D2 is obtained (ST51, ST52). Then, the obtained difference D
It is determined whether or not 2 is smaller than the update utterance count determination threshold value (ST53). If the difference D2 is small (Yes in the branch determination of step 53), the process proceeds to step 54, and 2 is set to the number of utterances for updating. Conversely, when the difference D2 is large (No in the branch determination of step 53), the process proceeds to step 55, and 1 is set to the number of utterances for updating.

After the number of utterances for updating is determined, weighting is performed next (ST56). As is clear from the figure, the weight becomes smaller as the difference D2 increases. That is, when the difference D2 is large, it is determined that "variation in voice" of the speaker is large, and the weight of the new input is set low. That is, since the variation is large, weighting is performed so that past speech feature amounts remain as much as possible.

When the speech feature is expressed in a vector, each feature data (the feature vector of each element) is substituted into the equation shown in FIG. 11 to obtain the updated feature vector. .

When the number of utterances for registration and the update weight are determined as described above, the process proceeds to step 35, where the given registration voice is obtained, and a score is calculated based on the obtained voice (ST35, ST36). . When the obtained matching score exceeds the threshold (ST37), the registered voice is updated (ST11). If the score is less than the threshold, the process returns to step 35, and the registration voice is input again.

[0064]

As described above, according to the present invention, the necessity of updating the registered voice information is determined by using the matching score information, so that the updating process can be performed at an appropriate timing.

[Brief description of the drawings]

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

FIG. 2 is a part of a flowchart illustrating an operation.

FIG. 3 is a part of a flowchart illustrating an operation.

FIG. 4 is a diagram showing a data structure of an audio data storage unit.

FIG. 5 is a flowchart illustrating a function of an update necessity determining unit.

FIG. 6 is a diagram illustrating an example of determining whether or not update is necessary;

FIG. 7 is a diagram showing another data structure of the audio data storage unit.

FIG. 8 is a flowchart illustrating functions of a voice registration / management unit that is a main part of the second embodiment of the present invention.

FIG. 9 is a flowchart showing a detailed processing procedure of step 31 in FIG. 8;

FIG. 10 is a flowchart showing a detailed processing procedure of step 34 in FIG. 8;

FIG. 11 is a diagram illustrating a calculation example of new update registration voice data (feature amount).

[Description of Signs] 1 input unit 2 voice data storage unit 3 input understanding unit 4 voice collation unit 5 voice registration / management unit 6 response generation unit 7 update necessity determination unit 8 output unit

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Hiroshi Nakajima 10F, Hanazono Todo-cho, Ukyo-ku, Kyoto-shi, Japan OMRON Corporation F-term (reference) 5D015 AA03 GG01 HH04 9A001 BB03 HH17

Claims (2)

[Claims] 1. A voice input means, a voice information storage means for holding registered voice information serving as a reference for voice recognition, and a voice input means based on the registered voice information stored in the voice information storage means. A speaker recognizing device that includes a voice collating unit that determines whether or not the input voice information is issued by a valid speaker. Based on the collation score information indicating the individuality,
Determining means for determining the necessity of updating the registered voice information held in the voice information storage means; and updating means for updating the registered voice information based on the determination result of the determining means. Speaker recognition device. 2. The method according to claim 1, wherein the updating unit determines that the determination result based on the collation score information in the determining unit needs to be updated, and that the registered voice is provided on condition that there is consent information of an updating process from a speaker. The speaker recognition device according to claim 1, wherein the information is updated.