JP2003076390A - Method and system for authenticating speaker - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a method and a system for authenticating a speaker for which registered voice input of the minimum time is sufficient without degrading the speaker authentication precision. SOLUTION: The speaker authenticating method for specifying a speaker by arbitrary speech contents takes the voice of the speaker as the input and analyzes the inputted voice of the speaker and extracts and temporarily stores feature parameters and generates or updates a speaker model of the speaker on the basis of the feature parameters and discriminates whether the speaker model has been sufficiently learnt or not on the basis of a prescribed criterion and makes the speaker additionally input his or her voice in the case that it is discriminated that the speaker model has not been sufficiently learnt.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speaker authentication system for identifying a speaker using voice. In particular, the present invention relates to speaker registration in a speaker authentication system capable of specifying a speaker even if the content of input voice is arbitrary.

[0002]

2. Description of the Related Art With the recent rapid development of computer-related technology, a speaker authentication technology capable of identifying a speaker by voice input has been rapidly advanced. And
As the number of applications to which speaker authentication technology is applied has increased rapidly, there has been a strong demand for improving the accuracy of speaker authentication. In the past, the accuracy of speaker authentication was improved by fixing the utterance content.

However, depending on the application, the utterance content may not be fixed, and even in such a case, it is necessary to maintain a certain degree of speaker authentication accuracy. Then, in the case of performing speaker authentication based on an arbitrary utterance content, it is an important factor for improving the recognition accuracy that the registered voice of the speaker registered in advance contains more phonemes. ing. Therefore, the registered voice for a longer recording time is required than the registered voice at the time of speaker authentication in which the utterance content is fixed.

For example, according to Furui's book "Acoustic / Voice Engineering" (Modern Science Co., Ltd. (1992), p.213), generally speaking, 10 to 30 are used for speaker authentication when the utterance content is arbitrary.
It is said that a registered voice of about a second is required, but on the other hand, when a general speaker model is created using a probabilistic model, the registration voice recording time is set to 60 seconds to register the voice. It has also been reported that the identification error was halved compared to when the voice recording time was 30 seconds ("Robust te
xt-independent speaker identification using gaussi
an mixture speaker models ", IEEE Trans. On Speech
and Audio Process. Vol.3 No.1.pp.78 (1995)), it is clear that the registered voice for as long recording time as possible is required to ensure sufficient accuracy of speaker authentication. .

FIG. 1 shows a principle diagram of a conventional speaker authentication system. As shown in FIG. 1, when the voice registration is started, the voice input unit 1 captures the voice and the voice analysis unit 2
Then, the voice is converted into a characteristic parameter and temporarily stored in the RAM area 3 or the like.

Then, the voice registration amount determination unit 4 determines whether or not enough characteristic parameters are maintained to maintain a predetermined authentication accuracy, that is, whether or not a sufficient amount of voice input is registered. If it is determined that the number of voice inputs to be registered is insufficient, the process returns to the voice input unit 1 to output a message prompting the registrant to input additional voice. Finally, when it is determined that the amount of voice input to be registered is sufficient to maintain a predetermined authentication accuracy, the speaker model generation unit 6 generates a speaker model to perform registration processing. Was there.

[0007]

However, in the conventional method as described above, it is determined whether or not enough characteristic parameters are maintained to maintain a predetermined authentication accuracy, that is, a sufficient amount of voice input is registered. Since the voice input amount is used as an index for determining whether or not the speaker model is generated, the speaker model is not generated until the predetermined voice input amount is registered. Therefore, there is a problem that whether or not the speaker model itself maintains sufficient accuracy for speaker authentication can be verified only after the final generation of the speaker model.

Further, when a sufficient amount of voice input cannot be registered, there is also a clear index as to what kind of voice and how much voice should be input in order to sufficiently learn as a speaker model. There was also the problem that it did not exist. Therefore, although the amount of speech input is sufficient, the learning may be insufficient for a specific phoneme, resulting in a speaker model with low authentication accuracy.

Further, since it is necessary to register a large amount of voice input in order to give sufficient authentication accuracy to the final speaker model, a large capacity RAM area 3 etc. is secured for the entire system. I need to put it. Therefore, there is a problem that computer resources may be consumed more than necessary.

Further, when a registered speaker inputs a voice, it is necessary to speak for 30 seconds or more just for voice input, and the voice input work is a pain for the registered speaker, and it is efficient with a minimum speaking time. It is desirable to be able to finish voice registration.

In order to solve the above problems, the present invention provides
An object of the present invention is to provide a speaker authentication system and method that require a minimum time for registration voice input without lowering speaker authentication accuracy.

[0012]

In order to achieve the above object, a speaker authentication system according to the present invention is a speaker authentication system which specifies a speaker with arbitrary utterance contents, and inputs a voice of the speaker. A voice input unit, a feature parameter storage unit that analyzes the input voice of the speaker, extracts and temporarily stores the feature parameter, and a talk that generates or updates the speaker model of the speaker based on the feature parameter. Model generator / updater,
The speaker model includes a speaker model evaluation unit that determines whether or not the speaker model is sufficiently learned based on a predetermined criterion, and a speaker model storage unit that stores the speaker model as a speaker database. When it is determined that the model learning is insufficient, additional voice input is performed in the voice input unit, and when it is determined that the speaker model learning is sufficient, the speaker model is set to the speaker. It is characterized by storing in a database.

With this configuration, the speaker model is always generated with respect to the voice input, so that the learning degree of the speaker model at the time of input can be grasped and the speaker can be newly input every time the voice is input. Since the model is updated,
The degree of learning of the speaker model advances every time a voice is input, and it is possible to minimize the amount of voice input for the degree of learning to reach a predetermined value.

Further, the speaker authentication system according to the present invention preferably further comprises a utterance content presenting unit for presenting the utterance content to be input to the speaker. This is because the more effective voice input content can be presented, and the learning of the speaker model can be completed by the shorter voice input. It is preferable that the presentation contents include phonemes as wide as possible, and also include phonemes that are insufficient or missing in the already generated speaker model.

Further, the speaker authentication system according to the present invention is insufficient for speaker model generation based on the voice recognition unit that recognizes the input voice of the speaker and the recognition result in the voice recognition unit. It is preferable to further include a utterance content selection unit that selects the utterance content. This is because it is possible to exclude the utterance content that is duplicated with the recognized content from the target of re-input.

Further, the speaker authentication system according to the present invention includes an unspecified speaker model generated based on voice data of the unspecified speaker and a speaker generated or updated by the speaker model generation / update unit. A speaker model temporary storage unit for temporarily storing the model is further provided, in the case of the first voice input, a speaker model is generated based on the unspecified speaker model, and in the case of the second and subsequent voice inputs. The speaker model is updated based on the speaker model stored in the speaker model temporary storage unit, and if it is determined that the speaker model has been sufficiently learned, the speaker model is stored in the speaker model temporary storage unit. It is preferable to store the speaker model in use in the speaker database. This is because the predetermined authentication accuracy can be secured from the first voice input.

The present invention is also characterized by software that executes the functions of the speaker authentication system as described above as processing steps of a computer. Specifically, the speaker is specified by an arbitrary utterance content. A method of authenticating a speaker, comprising: a step of inputting a speaker's voice; a step of analyzing the input speaker's voice, extracting a characteristic parameter and temporarily storing the characteristic parameter; Of the speaker model, the step of determining whether or not the learning of the speaker model is sufficient based on a predetermined criterion, and the step of saving the speaker model as a speaker database. If it is determined that the learning of the speaker model is insufficient, additional voice input is performed in the step of inputting the voice of the speaker, and it is determined that the learning of the speaker model is sufficient. A speaker Characterized in that it is a speaker authentication method and a computer-executable program for implementing such a process to save the Dell speaker database.

With such a configuration, by loading and executing the program on a computer, a speaker model is always generated for voice input. Therefore, the learning degree of the speaker model at the time of input can be grasped. Since the speaker model is updated each time a new voice is input, the learning degree of the speaker model will advance each time a voice is input, and the learning degree reaches a predetermined value. It is possible to realize a speaker authentication system that requires only minimal voice input.

[0019]

BEST MODE FOR CARRYING OUT THE INVENTION (First Embodiment) A speaker authentication system according to a first embodiment of the present invention will be described below with reference to the drawings. First, the first embodiment of the present invention
FIG. 2 shows a principle diagram of the speaker authentication system according to the present invention.

In FIG. 2, 1 is a voice input unit for inputting a voice of a registrant at the time of voice registration, 2 is a voice analysis unit for analyzing the input voice and converting it into a characteristic parameter, and 3 is a temporary feature parameter. The RAM areas that are temporarily stored are shown.

Reference numeral 5 denotes a speaker model generation means selecting section for selecting whether to create or update the speaker model according to the number of times the registered voice is input. If the registered voice input is the first time, the speaker is selected. The model generation unit 6 generates a speaker model,
If it is the second time or later, the speaker model updating unit 7 updates the speaker model.

Further, 8 is a speaker model evaluation unit for judging whether or not the speaker model has sufficient authentication accuracy, that is, whether or not learning as a speaker model is sufficient, based on a predetermined judgment criterion. When it is determined that learning is sufficient as the speaker model, it is determined that the generated speaker model has a certain degree of authentication accuracy. On the other hand, when it is determined that the learning as the speaker model is insufficient, it is determined that the speaker model being generated does not have a certain level of authentication accuracy,
The voice input unit 1 inputs the second and subsequent voices.

With such a configuration, it is possible to first create and update the speaker model for each voice input unit. That is, the first voice input is made,
The voice is captured by the voice input unit 1. The captured voice is converted into a characteristic parameter by the voice analysis unit 2. The created characteristic parameter will be stored in the RAM area 3. Therefore, it is sufficient for the RAM area 3 to store the characteristic parameters for each input unit.

In the case of the first voice input, the speaker model generator 6 receives the characteristic parameter from the RAM area 3 and generates a speaker model. In the case of the second and subsequent voice inputs, the speaker model updating unit 7 receives the characteristic parameters from the RAM area 3 and updates the already generated speaker model.

Further, after the speaker model is generated or updated, the speaker model is evaluated as to whether or not learning is sufficient as the speaker model, that is, whether or not a predetermined recognition accuracy is secured as the speaker model. When it is determined that the learning is insufficient as the speaker model in the part 8,
Since it can be determined that the predetermined recognition accuracy has not been ensured as the speaker model, the next voice input required for improving the authentication accuracy is fetched from the voice input unit 1 as an additional voice input.

When it is determined that the speaker model is sufficiently learned, it can be determined that the speaker model has a predetermined recognition accuracy, and thus no more voice registration is required.

Therefore, by judging whether or not learning is sufficient as a speaker model for each voice input unit based on a predetermined criterion, it is possible to prevent the registrant from being forced to make unnecessary voice input. In addition, it is possible to prevent the voice registration from ending in a state where learning is insufficient as a speaker model. Furthermore, by generating and updating the speaker model for each voice input unit, the RAM area required for storing the characteristic parameters can be made small, and effective use of computer resources can be achieved. Becomes

More specifically, the system is as shown in FIG. FIG. 3 is a configuration diagram of the speaker authentication system according to the first exemplary embodiment of the present invention. In FIG. 3, the voice input unit 31 captures the voice of the registered speaker, and the voice analysis unit 3
Pass to 2. The voice analysis unit 32 converts it into a characteristic parameter for speaker authentication.

The characteristic parameters extracted by the voice analysis unit 32 are stored in the RAM 33 for generating or updating the speaker model.
Etc. are temporarily stored. Of course, the characteristic parameter storage unit 3
It may be stored in a disk area such as 4.

Then, the speaker model generation means selecting section 35 confirms whether or not the speaker model of the registered speaker already exists. When the speaker model of the registered speaker does not exist, that is, when the voice input is the first time, a new speaker model is generated by passing the stored characteristic parameter to the speaker model generation unit 36. Will be done. After the speaker model is generated, the speaker model evaluation unit 38 verifies the learning degree of the speaker model.

When the speaker model of the registered speaker already exists, that is, when the voice input is the second or later, the speaker model updating unit 37 uses the stored characteristic parameter. Update the speaker model. When the speaker model is updated, the speaker model evaluation unit 38 also determines whether or not the learning degree of the updated speaker model is sufficient.

Then, in the speaker model evaluation section 38,
When it is determined that the learning by the input voice is insufficient, that is, the speaker model does not secure sufficient authentication accuracy, the process returns to the voice input unit 31 and the voice is input again. In this case, it is preferable to provide a re-input promoting unit 39 that outputs a message prompting the speaker to re-input. This is because the speaker himself can recognize the degree of learning of the speaker model.

On the other hand, if it is determined that the learning is sufficient, that is, the speaker model has sufficient authentication accuracy, it is stored in the speaker model storage unit 40 as the speaker model database 41. It will be.

Various types of parameters can be considered as the characteristic parameters. For example, LPC (Linear Pre
dictive Coding) Cepstrum and MFCC (Mel Freque
ncyCepstral Coefficients) or any other characteristic parameter used in speaker authentication by normal voice input.

Further, it is sufficient to store only the characteristic parameters, and it is sufficient that the time taken as voice input is about 5 to 15 seconds (one sentence to a few sentences in a short sentence).

On the other hand, various methods can be considered for the speaker model itself and its generation method. For example, GMM
A method using a probabilistic model such as (Gaussian Mixture Model) may be used, or a method may be used in which a codebook is created by performing clustering using the k-means method or the LBG method. The method of generating the speaker model itself is not particularly limited, and any method that can perform speaker authentication by voice input may be used.

Similarly, various methods can be considered for the method of updating the speaker model. For example, MAP (maximum ap
osteriori) method and a method using a model update algorithm such as Bayesian adaptation. The method for updating the speaker model is also not particularly limited, and any method that can perform speaker authentication by voice input may be used.

In the first embodiment, the choice of whether to newly generate a speaker model or update an existing speaker model is made by selecting the number of times of voice input (whether it is the first time or not). However, the present invention is not limited to this, and the determination may be made based on whether or not the speaker model of the registrant has already been generated. Furthermore, even if the speaker model has already been generated,
If the speaker wants to re-register itself, or
If it is determined that the authentication accuracy of the speaker model itself has not reached the predetermined level, a new speaker model may be generated.

Whether or not the generated or updated speaker model is sufficiently learned is determined using the following criteria.

First, the distance difference or the likelihood difference in the vector space with respect to the input voice before and after the speaker model is updated is obtained. If the change in the distance difference or the likelihood difference is small, it can be determined that the degree of learning of the speaker model has advanced, and the learning itself ends.

For example, FIG. 4 is a diagram showing the difference in log likelihood before and after the speaker model is updated. Here, the log-likelihood difference before and after the speaker model is updated when the initial speaker model is generated based on four sentences and then the speaker model is updated for each sentence is shown. The horizontal axis shows the total number of sentences used for learning the speaker model.

As can be seen from FIG. 4, the log-likelihood difference before and after the speaker model update tends to become smaller as the learning progresses. Therefore, it is sufficient to judge that the learning has been sufficiently performed at the time when the decrease in the logarithmic likelihood difference before and after the speaker model update is saturated. In FIG. 4, the threshold is determined at the time when there are two voice inputs with a log likelihood difference of 3 or less before and after the speaker model update.

The criterion for determining whether or not the speaker model has been sufficiently learned is not limited to the above method. For example, the speaker model is G
In the case of MM, it may be considered to use the value of variance as an index. That is, when the variance value is too small, it is considered that the feature parameter is not sufficiently extracted, and the user is prompted to input additional voice.

When the speaker model is a codebook, it can be considered that the number of samples assigned to the centroid is used as an index. That is, when the number of samples assigned to the centroid is small, the sample point selected as the representative point is more likely to be a singular point, and therefore the user is prompted to input additional voice. Become.

Further, it is preferable that the characteristic parameters and the speaker model are created for each user. That is, it is possible to identify a plurality of speakers by registering the user identifier at the same time as the speaker registration and generating an independent feature parameter and speaker model for each user identifier.

Next, the processing flow of the program for realizing the speaker authentication system according to the first embodiment of the present invention will be described. FIG. 5 shows a flow chart of processing of a program that realizes the speaker authentication system according to the exemplary embodiment of the present invention.

In FIG. 5, first, voice data of the speaker is input (step S501), and characteristic parameters are extracted from the input voice data (step S502). Then, when the voice input is the first time (step S50
3: Yes), a new speaker model is generated (step S
504) If the voice input is for the second time or later (step S503: No), the speaker model is updated (step S505).

Next, it is judged whether or not the speaker model is sufficiently learned, based on the judgment criteria indicating the degree of learning (step S506), and it is judged that the learning of the speaker model is insufficient. In this case (step S506: No), the additional input prompting message is output (step S507), and the speaker inputs the voice again (step S50).
1).

On the other hand, when it is judged that the learning of the speaker model is sufficient (step S506: Yes), the generated / updated speaker model is stored in the database as a new speaker model. (Step S508).

As described above, according to the first embodiment, it is necessary to force the registrant to perform unnecessary voice input by determining whether or not each voice input unit is sufficiently learned as a speaker model. It is possible to prevent the voice registration from occurring and to prevent the voice registration from ending in a state where the learning is insufficient as the speaker model. Furthermore, by creating and updating the speaker model for each voice input unit, the storage area required to store the characteristic parameters can be made small, and effective use of computer resources can be achieved. Becomes

(Second Embodiment) Hereinafter, a speaker authentication system according to a second embodiment of the present invention will be described with reference to the drawings. First, FIG. 6 shows a block diagram of a speaker authentication system according to a second exemplary embodiment of the present invention.

In FIG. 6, a point different from the speaker authentication system according to the first embodiment is that a utterance content presenting section 61 and a utterance content control section 62 for presenting the utterance content to be input for registration to the registrant are provided. That is the point. That is, the utterance content presentation unit 61 presents the utterance content to be input by the user, and the utterance content control unit 62 controls the content presented to the utterance content presentation unit 61.

First, the utterance content presented by the utterance content presenting section 61 is selected and presented from the text database 63 which stores text data in which phonemes are distributed in a well-balanced manner. The presentation method is not particularly limited, and may be displayed and output on a display device such as a display, or may be output and presented by synthetic voice using a telephone line or the like.

Next, the utterance content control unit 62 analyzes the voice recognition result of the text data presented in the past and the voice input data corresponding to the text data, based on the speaker model already generated. By
Phonemes already included in the speaker model and not included,
Or it will clearly recognize phonemes that are only slightly included. Then, the presentation content in the utterance content presentation unit 61 is controlled by selecting from the text database 63 text data that widely includes phonemes that are not included in the speaker model or phonemes that are only slightly included. Become.

Further, as shown in FIG. 7, a voice recognition unit 71 for recognizing the registered voice input is further provided, and the voice recognition result in the voice recognition unit 71 is used in the utterance content control unit 62 to make the speaker It is also conceivable to control to select the utterance content that is insufficient for model creation.

That is, in FIG. 7, the first input voice is taken in by the voice input unit 31, the characteristic parameters are extracted by the voice analysis unit 32, and the voice is input by the speaker model generation unit 36 or the speaker model update unit 37. The speaker model is generated or updated every time. After the speaker model is generated or updated, the voice recognition unit 71 performs voice recognition on the input voice data, and the recognition result is used as the utterance content control unit 62.
Send to.

The utterance content control unit 62 selects a text including a necessary phoneme from the received voice recognition result from the text database 63 and sends it to the utterance content presentation unit 61. Then, the utterance content presenting section 61 displays the most effective text data for improving the authentication accuracy as the utterance content, and performs the second voice input.

That is, since the speaker model that has already been generated is generated based on the previously input speech input, first, speech recognition is performed on the previously input speech input, and the speech database 63 is stored in the text database 63. By comparing with the prepared text data and selecting the text data with a small degree of matching, by selecting the text data so that all phonemes in a specific language, such as Japanese, can be comprehensively captured. , It is possible to present text data containing more phonemes not included in the speaker model.

By doing so, it is possible to perform the input work without waste for the speaker, and it is possible to expect the maximum effect with the minimum utterance input time.

Next, the flow of processing of the program that realizes the speaker authentication system according to the second embodiment of the present invention will be described. FIG. 8 shows a flow chart of processing of a program that realizes the speaker authentication system according to the exemplary embodiment of the present invention.

In FIG. 8, first, voice data of a speaker is input (step S801), and characteristic parameters are extracted from the input voice data (step S802). In parallel with the extraction of the characteristic parameter, voice recognition is performed on the input voice data (step S803), and the recognition result is passed to the utterance content control unit.

When the voice input is the first time (step S804: Yes), a speaker model is newly generated (step S805), and when the voice input is the second time or later (step S804: No), the speaker model is updated (step S806).

Next, it is judged whether or not the speaker model is sufficiently learned, based on the judgment criterion indicating the degree of learning (step S807), and it is judged that the learning of the speaker model is insufficient. In this case (step S807: No), the text database 63 prepared in advance as the utterance content is searched based on the content of the voice recognition (step S807).
808).

Then, the text data having the lowest degree of coincidence with the speech recognition content (text data most covering all phonemes of a specific language) is selected (step S809),
The text data is presented to the speaker as the next utterance content (step S810), and the speaker inputs the content of the text data again by voice (step S80).
1).

On the other hand, when it is determined that the learning of the speaker model is sufficient (step S807: Yes), the generated / updated speaker model is stored in the database as a new speaker model. (Step S811).

As described above, according to the second embodiment, by providing the registrant with the utterance content to be input for registration, the registrant needs to utter in consideration of the utterance content to be input. As a result, the load at the time of voice input can be minimized.

Further, since it is possible to control so as to include more phonemes in the presented utterance content, even if the voice input time is short, the phoneme uttered by the registrant is most efficiently incorporated into the speaker model. It becomes possible.

Furthermore, a means for recognizing the voice uttered when the first registered voice is input is provided, and the utterance content which is insufficient for creating the speaker model is selected using the voice recognition result. It is possible to finish the voice registration in a short registration time by presenting it when the registered voice is input.

(Third Embodiment) A speaker authentication system according to a third embodiment of the present invention will be described below with reference to the drawings. First, FIG. 9 shows a block diagram of a speaker authentication system according to a third exemplary embodiment of the present invention.

In FIG. 9, a point different from the speaker authentication system according to the first embodiment is that, in the first voice input, a speaker model generated in advance based on a voice of an unspecified speaker is used. The point is to generate a speaker model.
That is, the RAM 33 is referred to by referring to the unspecified speaker model 91 generated in advance by the speaker model generation unit 36.
A new speaker model is generated by updating the unspecified speaker model 91 based on the characteristic parameter temporarily stored in the.

First, when the unspecified speaker model 91 is generated, the voice data of a large number of speakers of 100 to 1000 or more are input, and the models such as the GMM and the codebook described above are generated.

Then, as shown in FIG. 9, the first voice input is taken in by the voice input unit 31, the characteristic parameters are extracted by the voice analysis unit 32, and the speaker model generation unit 36 makes the unspecified speaker model. Based on 91 and the extracted feature parameter, a temporary speaker model for each voice input is generated. Then, the generated speaker model is temporarily stored in the speaker model temporary storage unit 92.

Similarly, the characteristic parameters are extracted from the second and subsequent input voices, but by updating the speaker model stored in the speaker model temporary storage unit 92 in the speaker model updating unit 37, It will be updated to a new speaker model. When the speaker model is updated, the speaker model evaluation unit 38 also updates the updated speaker model.
It will be determined whether or not the degree of learning is sufficient.

Then, as in the first embodiment, the speaker model evaluation unit 38 determines that the learning by the input voice is insufficient, that is, the speaker model does not have sufficient authentication accuracy. In this case, the voice input unit 31 is returned to and voice input is performed again. In this case, the re-input promoting unit 3 which outputs a message prompting the speaker to re-input.
It is preferable to provide 9. This is because the speaker himself can recognize the degree of learning of the speaker model.

On the other hand, if it is determined that the learning is sufficient, that is, the speaker model has sufficient authentication accuracy, the speaker model storage unit 40 stores the speaker model database 41. It will be.

By doing so, it is possible to expect a certain degree of authentication accuracy from the first voice input, and it is possible to expect the maximum effect with respect to additional input with the minimum utterance content.

In the third embodiment, the temporary speaker model is stored in the speaker model temporary storage unit 92.
The speaker model may be directly generated in the speaker model database 41, and the speaker model may be updated while repeatedly inputting voice until the authentication accuracy can be ensured to a predetermined level.

The program for realizing the speaker authentication system according to the embodiment of the present invention is, as shown in FIG. 10, a CD-ROM 102-1 and a flexible disk 1.
In addition to the portable recording medium 102 such as 02-2, another storage device 101 provided at the end of the communication line, or a recording medium 104 such as a hard disk or a RAM of the computer 103.
Any of the above may be stored, and when the program is executed, the program is downloaded and executed on the DSP.

Further, regarding the speaker model and the like generated by the speaker authentication system according to the embodiment of the present invention,
As shown in FIG. 10, not only the portable recording medium 102 such as the CD-ROM 102-1 and the flexible disk 102-2 but also another storage device 10 provided at the end of the communication line.
1, a hard disk or RAM of the computer 103,
Alternatively, it may be stored in any recording medium 104 represented by a non-volatile memory such as a flash memory, and is read by the computer 103 when using the speaker authentication system according to the present invention, for example.

[0080]

As described above, according to the speaker authentication system of the present invention, it is unnecessary for the registrant to judge whether or not the speaker model is sufficiently learned for each voice input unit. It is possible to prevent the voice input from being forced and to prevent the voice registration from ending in a state where learning is insufficient as a speaker model.

Further, according to the speaker authentication system of the present invention, the speaker model is generated / updated for each voice input unit, thereby reducing the storage area required for storing the characteristic parameters. Therefore, it is possible to effectively use computer resources.

[Brief description of drawings]

FIG. 1 Principle diagram of a conventional speaker authentication system

FIG. 2 is a principle diagram of a speaker authentication system according to the first embodiment of the present invention.

FIG. 3 is a configuration diagram of a speaker authentication system according to the first exemplary embodiment of the present invention.

FIG. 4 is a diagram showing a log likelihood difference before and after a speaker model update in the speaker authentication system according to the first exemplary embodiment of the present invention.

FIG. 5 is a flowchart of processing in the speaker authentication system according to the first exemplary embodiment of the present invention.

FIG. 6 is a configuration diagram of a speaker authentication system according to a second embodiment of the present invention.

FIG. 7 is a configuration diagram of a speaker authentication system according to a second embodiment of the present invention.

FIG. 8 is a flowchart of processing in the speaker authentication system according to the second exemplary embodiment of the present invention.

FIG. 9 is a configuration diagram of a speaker authentication system according to a third embodiment of the present invention.

FIG. 10 is an exemplary diagram of a computer environment.

[Explanation of symbols]

1, 31 voice input unit 2, 32 voice analysis unit 3 RAM area 4 voice registration amount determination unit 5, 35 speaker model generation means selection unit 6, 36 speaker model generation unit 7, 37 speaker model update unit 8, 38 Speaker model evaluation unit 33 RAM 34 Characteristic parameter storage unit 39 Re-entry promotion unit 40 Speaker model storage unit 41 Speaker model database 61 Speech content presentation unit 62 Speech content control unit 63 Text database 71 Speech recognition unit 91 Unspecified speaker Model 92 Talker model temporary storage unit 101 Storage device 102 at line destination Portable recording medium 102-1 CD-ROM or flexible disk 102-1 CD-ROM 102-2 Flexible disk 103 Computer 104 Recording on RAM / hard disk etc. on computer Medium

   ─────────────────────────────────────────────────── ─── Continued front page    (72) Inventor Chiharu Kawai             4-1, Kamiodanaka, Nakahara-ku, Kawasaki-shi, Kanagawa             No. 1 within Fujitsu Limited F-term (reference) 5D015 AA03 GG06

Claims (6)

[Claims] 1. A speaker authentication system for identifying a speaker with arbitrary utterance content, comprising: a voice input unit for inputting a voice of the speaker; analyzing the input voice of the speaker; A characteristic parameter storage unit that extracts and temporarily stores the speaker model, a speaker model generation / update unit that generates or updates the speaker model of the speaker based on the characteristic parameter, and learning of the speaker model is sufficient. A speaker model evaluation unit that determines whether or not the speaker model is stored based on a predetermined determination criterion; and a speaker model storage unit that stores the speaker model as a speaker database, and learning of the speaker model is insufficient. If it is determined that there is additional voice input in the voice input unit, if it is determined that the learning of the speaker model is sufficient, save the speaker model in the speaker database Characterized by And speaker authentication system. 2. The speaker authentication system according to claim 1, further comprising a speech content presentation unit that presents speech content to be input to the speaker. 3. A voice recognition unit for recognizing an input voice of the speaker, and a voicing for selecting a voicing content which is insufficient for generating the speaker model based on a recognition result in the voice recognition unit. The speaker authentication system according to claim 1, further comprising a content selection unit. 4. An unspecified speaker model generated based on voice data of an unspecified speaker, and a speaker model for temporarily storing the speaker model generated or updated by the speaker model generation / update unit. A temporary storage unit is further provided, wherein a speaker model is generated based on the unspecified speaker model in the case of the first voice input, and the speaker model temporary storage unit is generated in the case of the second and subsequent voice inputs. The speaker model is updated based on the speaker model stored in the speaker model, and when it is determined that the learning of the speaker model is sufficient, the speaker model is temporarily stored in the speaker model temporary storage unit. The speaker authentication system according to claim 1, wherein the speaker model is stored in the speaker database. 5. A speaker authentication method for identifying a speaker with arbitrary utterance content, comprising the steps of inputting a voice of a speaker, analyzing the input voice of the speaker, and extracting characteristic parameters. Temporarily storing, based on the characteristic parameter, generating or updating the speaker model of the speaker, based on a predetermined determination criteria whether the learning of the speaker model is sufficient Including the step of determining, the step of storing the speaker model as a speaker database, in the step of inputting the voice of the speaker, if it is determined that the learning of the speaker model is insufficient. A speaker authentication method, characterized in that when additional voice input is performed and it is determined that the learning of the speaker model is sufficient, the speaker model is stored in the speaker database. 6. A program to be executed by a computer that embodies a speaker authentication method for identifying a speaker with arbitrary utterance content, the step of inputting a voice of a speaker, and the input voice of the speaker. The step of generating and updating the speaker model of the speaker on the basis of the characteristic parameter by extracting the characteristic parameter and temporarily storing the characteristic parameter, and whether or not the learning of the speaker model is sufficient. Including a step of determining whether or not based on a predetermined determination criteria, and a step of storing the speaker model as a speaker database, if it is determined that the learning of the speaker model is insufficient, In the step of inputting the voice of the speaker, additional voice input is performed, and when it is determined that the learning of the speaker model is sufficient, the speaker model is set to the speaker database. Programs to be executed by storing the computer characterized by.