KR20200040425A - Speaker recognition apparatus and operation method thereof - Google Patents

Abstract

The present invention relates to a speaker recognition method, which may be applied to an AI home assistant service. The speaker recognition method comprises the steps of: receiving a voice signal of a user at a home, who requests voice registration; measuring similarity between a user voice requesting the voice registration and a previously registered user voice; comparing a measured similarity score and a reference score for determining whether the user belongs to a specific group; and setting a reference score for speaker identification based on a comparison result.

Description

SPEAKER RECOGNITION APPARATUS AND OPERATION METHOD THEREOF}

The present invention relates to a user-specific speaker recognition device and an operation method thereof, and more specifically, to a speaker recognition device and an operation method capable of automatically updating a reference score for speaker identification according to voice similarity between users in the home. It is about.

Recently, with the rapid development of information and communication technology, interest and demand for Internet of Things (IoT) technology is rapidly increasing. Such an IoT may be defined in various ways according to a viewpoint looking at it. However, in essence, IoT is an intelligent information communication technology or service that enables communication between people and things, things and things by connecting various things with a communication network based on the Internet.

The IoT technology is applied in various technology fields such as a smart home, a smart health, and a smart car. In particular, research on smart home services incorporating IoT technology into home network systems is actively being conducted.

Smart home service refers to the overall system that seeks to improve living standards through IoT devices that include IoT functions in a residential environment where a communication network is built. By connecting home appliances such as TV / refrigerator / air conditioner, energy consumption devices such as electricity / water, security services, etc. to a communication network, users can remotely monitor the situation information in the home through a smart phone or voice controller (or AI speaker). Real-time check and control can be enabled. In particular, in recent years, smart home services that can remotely control IoT devices existing in the home through AI speakers have been increasing.

The AI speaker can provide an interactive AI home secretary service by utilizing voice recognition technology and artificial intelligence technology. Here, the interactive AI home secretary service literally means a service in which an artificial intelligence in a home identifies a speaker and serves as a speaker's home secretary. These AI speakers are various services such as personal schedule management service, SNS management service, app execution service, internet shopping service, email management service, messenger management service, multimedia playback service, weather / traffic / travel information service, IoT device control service, etc. Can provide.

The speaker recognition device measures the similarity between the user voice input through the AI speaker and the previously registered user voice, and determines that the user is a pre-registered user when the measured similarity score exceeds a threshold. Is done.

However, since the existing speaker recognition apparatus uses only one reference score for speaker identification, in the case of a special group having similar voices than other ordinary people, such as siblings / sisters / twins, there is a problem in that the speaker recognition rate is lowered. Therefore, there is a need for a method for accurately identifying a speaker even when there are siblings / sisters / twins with similar voices in the home.

The present invention aims to solve the above and other problems. Another object is to provide a speaker recognition apparatus and an operation method thereof that can automatically update a reference score for speaker identification based on a similarity between a user voice attempting voice registration and a pre-registered user voice.

Another object is to provide a speaker recognition apparatus and an operation method thereof, which can identify a speaker by using an automatically updated reference score when pre-registered home users belong to a special group.

According to an aspect of the present invention to achieve the above or another object, receiving a voice signal of a user at home requesting voice registration; Measuring similarity between a user voice requesting the voice registration and a pre-registered user voice; Comparing the measured similarity score with a reference score for determining whether a special group exists; And setting a reference score for speaker identification based on the comparison result.

More preferably, the setting step is characterized in that when the measured similarity score is less than or equal to the reference score for determining whether a special group, the reference score for speaker identification is maintained. In addition, the setting step is characterized in that when the measured similarity score is greater than the reference score for determining whether a special group, the reference score for speaker identification is updated. The reference score for the speaker identification is characterized in that it is updated based on the measured similarity score.

More preferably, the receiving step is characterized in that the voice signal of the user at home is received from the AI speaker. The speaker recognition method further comprises extracting feature values of the voice signal received from the AI speaker, and generating a speaker model corresponding to the user in the home based on the extracted feature values.

More preferably, the speaker recognition method comprises: receiving a voice signal of a home user requesting speech recognition, and measuring similarity between a user voice requesting speech recognition and a pre-registered user voice. It is characterized by further including. In addition, the speaker recognition method may further include identifying a speaker by comparing the similarity score between the user voice requesting speech recognition and a pre-registered user voice and the set reference score.

According to another aspect of the invention, the process of receiving a voice signal of a user at home requesting voice registration; Measuring a similarity between the user voice requesting the voice registration and a pre-registered user voice; Comparing the measured similarity score with a reference score for determining whether a special group exists; And a program stored in a computer-readable recording medium so that a process of setting a reference score for speaker identification based on the comparison result is executed on a computer.

According to another aspect of the present invention, a voice receiving unit for receiving a voice signal of a home user requesting voice registration; A similarity measuring unit measuring a similarity between the user voice requesting the voice registration and a pre-registered user voice; And a reference score setting unit configured to set a reference score for speaker identification by comparing the measured similarity scores with reference scores for determining whether a special group exists.

More preferably, the speaker recognition apparatus is characterized in that it further comprises a voice feature extraction unit for extracting feature values of the voice signal using a predetermined voice feature extraction algorithm. In addition, the speaker recognition device is characterized in that it further comprises a speaker model generating unit for generating a speaker model corresponding to the user in the home based on the feature values of the voice signal. In addition, the speaker recognition apparatus is characterized in that it further comprises a speaker identification unit for identifying a speaker by comparing the similarity score and a reference score between a user voice attempting speech recognition and a pre-registered user voice.

Referring to the effects of the speaker recognition apparatus and its operation method according to embodiments of the present invention are as follows.

According to at least one of the embodiments of the present invention, by automatically updating a reference score for speaker identification according to a similarity between a user voice attempting to register a voice and a pre-registered user voice, the voice is very like a sibling / sister / twin. Similar special groups have the advantage of being able to accurately identify the speaker.

However, the effects that the speaker recognition apparatus and the operation method thereof according to the embodiments of the present invention can achieve are not limited to those mentioned above, and other effects that are not mentioned are from the following description. It will be clearly understood by those with ordinary knowledge.

1 is a block diagram of a speaker recognition system according to an embodiment of the present invention;
2 is a block diagram of an AI speaker according to an embodiment of the present invention;
3 is a block diagram of a speaker recognition apparatus according to an embodiment of the present invention;
4 is a flowchart illustrating a voice registration method according to an embodiment of the present invention;
5 is a diagram illustrating a method of updating a reference score for speaker identification;
6 is a flowchart illustrating a speaker recognition method according to an embodiment of the present invention.

Hereinafter, exemplary embodiments disclosed herein will be described in detail with reference to the accompanying drawings, but the same or similar elements are assigned the same reference numbers regardless of the reference numerals, and overlapping descriptions thereof will be omitted. The suffixes "modules" and "parts" for the components used in the following description are given or mixed only considering the ease of writing the specification, and do not have meanings or roles distinguished from each other in themselves. That is, the term 'unit' used in the present invention refers to a hardware component such as software, FPGA or ASIC, and 'unit' performs certain roles. However, 'wealth' is not limited to software or hardware. The 'unit' may be configured to be in an addressable storage medium or may be configured to reproduce one or more processors. Thus, as an example, 'part' refers to components such as software components, object-oriented software components, class components and task components, processes, functions, attributes, procedures, Includes subroutines, segments of program code, drivers, firmware, microcode, circuitry, data, database, data structures, tables, arrays and variables. The functionality provided within components and 'parts' may be combined into a smaller number of components and 'parts' or further separated into additional components and 'parts'.

In addition, in describing the embodiments disclosed in the present specification, when it is determined that detailed descriptions of related known technologies may obscure the gist of the embodiments disclosed herein, detailed descriptions thereof will be omitted. In addition, the accompanying drawings are only for easy understanding of the embodiments disclosed in the present specification, and the technical spirit disclosed in the specification is not limited by the accompanying drawings, and all modifications included in the spirit and technical scope of the present invention , It should be understood to include equivalents or substitutes.

The present invention proposes a speaker recognition apparatus and an operation method thereof capable of automatically updating a reference score for speaker identification based on the similarity between a user voice attempting voice registration and a pre-registered user voice. In addition, the present invention provides a speaker recognition apparatus and an operation method for identifying a speaker using an automatically updated reference score when pre-registered home users belong to a special group.

Hereinafter, various embodiments of the present invention will be described in detail with reference to the drawings.

1 is a block diagram of a speaker recognition system according to an embodiment of the present invention.

Referring to FIG. 1, the speaker recognition system 100 according to an embodiment of the present invention includes a router 110 and an AI speaker 120 located inside a home, an Internet network 130 located outside the home, and an Internet service It may include a provision server 140. The Internet service providing server 140 may include a speaker recognition server (or speaker recognition device) for identifying voices (voices) of users in the home.

The router (Access Point, 110) is a device that connects wired / wireless communication so that the AI speaker 120 located in the home and a plurality of IoT devices (not shown) can be connected to the Internet network 130. The router 110 may perform a function of sharing an Internet address (IP address) provided by an Internet service provider so that multiple terminals existing in the home can share with each other.

The AI speaker (or voice controller, 120) can be connected to a router 110 or a communication modem (not shown) through a wired / wireless communication interface, and provides an Internet service located externally through the router 110 or a communication modem It may communicate with the server 140.

The AI speaker 120 is directly connected to a plurality of IoT devices through a wired / wireless communication interface to communicate with the IoT devices. On the other hand, although not shown in the drawing, in another embodiment, the AI speaker 120 may be connected to a home hub (not shown) through a wired / wireless communication interface, and a plurality of IoT devices may be connected through the home hub. It can also communicate with.

The AI speaker 120 may provide an AI home secretary service corresponding to voice commands of home users. For example, the AI speaker 120 may check the status of a plurality of IoT devices existing in the home in response to a user's voice command or control the operation of the devices.

When a plurality of user accounts are registered in the AI home secretary service, in order to identify (differentiate) the voices of users in the home, the AI speaker 120 provides the user voice input through the microphone to the Internet service providing server 140 can do.

The Internet service providing server 140 may access the router 110 existing in the home through the Internet network 130 and communicate with the AI speaker 120 through the router 110.

The Internet service providing server 140 may identify a speaker by analyzing a user's voice signal received from the AI speaker 120 through the speaker recognition server. As a result of the speaker identification, if the user requesting speech recognition is a pre-registered home user, the Internet service providing server 140 works with the AI speaker 120 to provide an AI home secretary service corresponding to the voice commands of the home users. Can provide.

2 is a block diagram of an AI speaker according to an embodiment of the present invention.

Referring to FIG. 2, the AI speaker 200 according to an embodiment of the present invention may include a communication unit 210, an input unit 220, an output unit 230, a memory 240, and a control unit 250. . The components shown in FIG. 2 are not essential for implementing the AI speaker 200, so the AI speaker described herein may have more or fewer components than those listed above.

The communication unit 210 may include a wired communication module for supporting wired communication and a short range communication module for supporting short-range wireless communication. The wired communication module is a wired communication network constructed according to technical standards or communication methods (for example, Ethernet, Power Line Communication (PLC), Home PNA (Home PNA), IEEE 1394, etc.) for wired communication. It transmits and receives a wired signal to and from at least one of a communication modem, a router, and an IoT device. The short-range communication module is for short-range communication (Bluetooth?), Radio Frequency Identification (RFID), Infrared Data Association (IrDA), Ultra-Wideband (UWB), ZigBee, NFC ( Near Field Communication (Wireless-Fidelity), Wi-Fi Direct, and Wireless Universal Serial Bus (USB) technology can be used to support short-range wireless communication.

The input unit 220 may include a microphone for inputting an audio signal. The microphone processes external sound signals as electrical voice data. The processed voice data may be variously used according to a function (or a running application program) performed by the AI speaker 200. Meanwhile, various noise cancellation algorithms for removing noise generated in the process of receiving an external sound signal may be implemented in the microphone.

The output unit 230 is for generating output related to visual, auditory, or tactile senses, and may include at least one of a display unit, an audio output unit, a hap tip module, and an optical output unit. The audio output unit may output audio data stored in the memory 240. The sound output unit may also output sound signals related to functions performed by the AI speaker 200. The sound output unit may include a speaker and a buzzer.

The memory 240 stores data supporting various functions of the AI speaker 200. The memory 240 may store application programs or applications driven by the AI speaker 200 and data and commands for the operation of the AI speaker 200.

The control unit 250 controls the operation related to the application program stored in the memory 240 and generally the overall operation of the AI speaker 200. Furthermore, in order to implement various embodiments described below on the AI speaker 200 according to the present invention, the controller 250 may control by combining at least one of the above-described components.

The control unit 250 may further include a speech recognition module 255. The voice recognition module 255 recognizes an external voice input through a microphone by driving a voice recognition engine to which a voice recognition algorithm is applied. That is, the voice recognition module 255 converts the external voice input through the microphone into digital data, amplifies the converted digital data (Pre-emphasis), and then detects the start and end points of the digitally converted voice. do. Subsequently, the speech recognition module 255 extracts speech feature values for speech between the detected start point and end point to recognize a unique voice or tone.

On the other hand, in the present embodiment, the voice recognition module 255 is illustrated to be implemented in the control unit 250, but is not limited thereto, and may be configured independently of the control unit 250. Furthermore, the voice recognition module may be implemented on the Internet service providing server 140 interworking with the AI speaker 200.

3 is a block diagram of a speaker recognition apparatus according to an embodiment of the present invention.

Referring to FIG. 3, a speaker recognition apparatus (or speaker recognition server, 300) according to an embodiment of the present invention includes a voice reception unit 310, a voice feature extraction unit 320, a speaker model generation unit 330, and similarity measurement It may include a unit 340, a reference score setting unit 350, a speaker identification unit 360 and a database 370. The components illustrated in FIG. 3 are not essential for implementing the speaker recognition apparatus 300, so the speaker recognition apparatus described herein may have more or fewer components than those listed above. .

The voice receiver 310 may receive a voice signal of a user in the home from the AI speaker 200. The voice receiver 310 may store the voice signal received from the AI speaker 200 in a database (or memory, 370). In addition, the voice receiver 310 may provide the voice signal received from the AI speaker 200 to the voice feature extraction unit 320.

The voice feature extraction unit 320 may extract feature values (or feature vectors) of the received voice signal using a predetermined voice feature extraction algorithm. As the speech feature extraction algorithm, any one of a Mel-frequency cepstrum coefficients (MFCC) algorithm, a perceptual linear prediction (PLP) algorithm, a Gammatone Cepstral Coefficients (GTCC) algorithm, and a Zero-Crossings with Peak Amplitudes (ZCPA) algorithm may be used. .

More preferably, an MFCC algorithm with high accuracy compared to the computational amount can be used. The MFCC algorithm can extract feature vectors of speech signals using MFC coefficients. Here, MFC (Mel-frequency cepstrum) is discrete in the mel-scale frequency domain (mel-scale frequency domain) that simulates the frequency response of the power spectrum (power spectrum) calculated for speech data in a unit speech frame (frame) It means the value of taking the cosine transform (DCT).

The speaker model generation unit 330 may generate a speaker model by using voice characteristics of a user who wants to register a voice (voice) as training data. The speaker model generation unit 330 may store a speaker model corresponding to a user in the home in the database 370. Speaker models stored in the database 370 may be used to identify the voices of users in the home.

The speaker model generation unit 330 may generate a speaker model using a predetermined statistical model technique. As the statistical model technique, any one of a Gaussian Mixture Model (GMM) technique and a Support Vector Machine (SVM) technique may be used, but is not limited thereto.

The similarity measurement unit 340 may measure the similarity between the voice of the user attempting voice registration or voice recognition and the speaker models stored in the database 370. The similarity measurement unit 340 may measure the similarity in units of voice frames for robust speaker identification.

The similarity measurement unit 340 may measure a similarity between a user voice attempting voice registration or voice recognition and a pre-registered user voice using a predetermined similarity measurement method. As the method for measuring the similarity, either a VQ (Vector Quantization) -based similarity measurement method or a DTW (Dynamic Time Warping) -based similarity measurement method may be used, but is not limited thereto.

The reference score setting unit 350 may set a reference score (or threshold, threshold) for general speaker identification. The reference score is set to a value for optimizing the speaker recognition rate. Also, the reference score may be compared with a similarity score for speaker identification.

The reference score setting unit 350 may determine whether the corresponding users are a general group or a special group based on the similarity of voices between pre-registered home users. That is, when the voice similarity between the users in the house exceeds a reference score (hereinafter referred to as 'first reference score' for convenience of description) for determining whether a special group is present, the reference score setting unit 350 determines the corresponding users It can be classified (recognized) as a 'special group'. On the other hand, if the voice similarity between the users in the home is less than or equal to the first reference score, the reference score setting unit 350 may classify (recognize) the users as a 'normal group'.

The reference score setting unit 350 maintains a reference score (hereinafter referred to as 'second reference score' for convenience of description) for speaker identification depending on whether or not pre-registered home users are a general group / special group, or applicable The baseline score can be updated. That is, the reference score setting unit 350 maintains a reference score for speaker identification when pre-registered home users are a general group, and a reference for speaker identification when pre-registered home users are a special group. The score can be updated.

The updated reference score may be set based on the similarity score between users in the home belonging to the special group. For example, the updated reference score may be set to 110% of the similarity score between user voices belonging to a special group, but is not limited thereto.

The speaker identification unit 350 may identify a speaker by comparing the similarity score between the user's voice attempting speech recognition and pre-registered speaker models and the second reference score. For example, if the similarity score between the user's voice attempting speech recognition and a specific speaker model exceeds a second reference score, the speaker identification unit 350 may convert the user attempting speech recognition into a user corresponding to the specific speaker model. Can be recognized.

The database (or memory, 370) stores data supporting various functions of the speaker recognition device 300. The database 370 may store a number of application programs or applications that are driven by the speaker recognition device 300 and data and commands for the operation of the speaker recognition device 300.

The database 370 is generated through the voice signal information of the home users acquired through the voice receiver 310, information about the reference score set through the reference score setting unit 350, and the speaker model generation unit 330. Information about the old speaker models.

As described above, as described above, the speaker recognition apparatus according to the present invention automatically updates the reference score for speaker identification according to the similarity between the user voice attempting to register the voice and the pre-registered user voices, thereby sibling / sister / Even special groups with very similar voices, such as twins, can accurately identify the speaker.

4 is a flowchart illustrating a voice registration method according to an embodiment of the present invention.

Referring to FIG. 4, the user at home may request to register his voice (voice) with the AI home secretary service (S410). At this time, the user at home may input his / her voice signal through the microphone of the AI speaker 200.

The speaker recognition device 300 may receive a voice signal of the user at home from the AI speaker 200 and store data regarding the received voice signal in a database (S420).

The speaker recognition device 300 may analyze the voice signal received from the AI speaker 200 and extract feature values (or feature vectors) of the voice signal (S430). At this time, the speaker recognition apparatus 300 may extract voice feature values of the user at home using a predetermined voice feature extraction algorithm.

The speaker recognition apparatus 300 may generate a speaker model of the corresponding user using the voice feature values of the user in the home as training data, and may store the generated speaker model in a database (S440). At this time, the speaker recognition apparatus 300 may generate a speaker model using a GMM statistical model technique or an SVM statistical model technique.

The speaker recognition apparatus 300 may measure the similarity between the user voice requesting voice registration and the pre-registered user voices using a predetermined similarity measurement method (S450).

The speaker recognition apparatus 300 may check whether the measured similarity score exceeds a reference score (that is, a first reference score) for determining whether a special group is present (S460).

As a result of the check in step 460, if the measured similarity score is less than or equal to the first reference score, the speaker recognition device 300 recognizes a user requesting voice registration and a pre-registered home user as a general speaker. A reference score for identification (ie, a second reference score) may be maintained (S470).

On the other hand, as a result of the check in step 460, when the measured similarity score is greater than the first reference score, the speaker recognition device 300 recognizes a user requesting voice registration and a pre-registered home user as a special group and is a speaker The reference score for identification may be updated (S480). At this time, the speaker recognition device 300 may update the reference score for speaker identification using the measured similarity score.

If the pre-registered home users are a general group, the speaker recognition device 300 may identify the speaker using the initially set reference score. On the other hand, if the pre-registered home users are a special group, the speaker recognition device 300 may identify the speaker by updating the initially set reference score.

As described above, as described above, the speaker recognition apparatus according to the present invention extracts a voice feature of a user requesting voice registration to generate a speaker model, and creates a speaker model between the user voice requesting voice registration and the pre-registered user voices. The reference score for speaker identification can be automatically updated according to the similarity.

5 is a diagram illustrating a method of updating a reference score for speaker identification.

Referring to FIG. 5, it is assumed that voices of twins B and fathers among members of the house are registered in the AI home secretary service, and twins A are requesting voice registration for the corresponding service.

When the twin A requests voice registration through the AI speaker 200, the speaker recognition device 300 may generate a speaker model corresponding to the twin A based on the voice signal received from the AI speaker 200 ( S510).

The speaker recognition apparatus 300 may measure the similarity between the voice of the twin A and the voice of the pre-registered father (S520). In this embodiment, it is assumed that the similarity score between the voice of the twin A and the voice of the father is 20, and the reference score for determining whether a special group is 70.

The speaker recognition apparatus 300 may check whether the measured similarity score exceeds a reference score for determining whether a special group is present (S530). As a result of the check, since the similarity score 20 between the twins A and the father is smaller than the reference score 70 for determining whether or not a special group, the speaker recognition device 300 recognizes the twins A and the father as a general group to identify the speaker The reference score for can be maintained as it is (S540).

Thereafter, the speaker recognition apparatus 300 may measure the similarity between the voice of the twin A and the voice of the previously registered twin B (S550). In this embodiment, it is assumed that the similarity score between the voice of twin A and the voice of twin B is 80.

The speaker recognition device 300 may check whether the measured similarity score exceeds a reference score for determining whether a special group is present (S560). As a result of the check, since the similarity score 80 between the twins A and the twins B is larger than the reference score 70 for determining whether or not a special group, the speaker recognition device 300 recognizes the twins A and twins B as a special group. The reference score for speaker identification may be updated (S570). In this case, the speaker recognition apparatus 300 may update the reference score for speaker identification based on the similarity score 80 between twins A and twins B.

As described above, if one of the pre-registered in-house users is in a special group relationship with a user requesting voice registration, the reference score for speaker identification may be updated. The updated reference score may be used as a new reference score for future speaker identification.

6 is a flowchart illustrating a speaker recognition method according to an embodiment of the present invention.

Referring to FIG. 6, the speaker recognition device 300 may receive a voice signal of a user at home from the AI speaker 200 (S610).

The speaker recognition device 300 may analyze the voice signal received from the AI speaker 200 and extract feature values (or feature vectors) of the voice signal (S620). At this time, the speaker recognition apparatus 300 may extract voice feature values of the user at home using a predetermined voice feature extraction algorithm.

The speaker recognition apparatus 300 may measure the similarity between a voice feature of a user requesting speech recognition and a voice feature of pre-registered users using a predetermined similarity measurement method (S630).

The speaker recognition apparatus 300 may compare the measured similarity score with the reference score for speaker identification (S640). Here, the reference score for speaker identification may be updated to a value greater than the initially set reference score to increase the accuracy of speaker recognition when at least two of the pre-registered home users are a special group.

As a result of the comparison in step 640, when the measured similarity score exceeds a reference score for speaker identification, the speaker recognition device 300 is a home corresponding to a specific speaker model that pre-registers a user who is attempting speech recognition. I can recognize it as my user.

As described above, the speaker recognition apparatus according to the present invention accurately identifies the voices of users belonging to the special group by applying the updated reference score to the speaker recognition algorithm when the pre-registered users in the house are a special group. (Distinction) can do it.

The above-described present invention can be embodied as computer readable codes on a medium on which a program is recorded. The computer-readable medium includes any kind of recording device in which data readable by a computer system is stored. Examples of computer-readable media include a hard disk drive (HDD), solid state disk (SSD), silicon disk drive (SDD), ROM, RAM, CD-ROM, magnetic tape, floppy disk, and optical data storage device. There is this. In addition, the computer may include a control unit of the terminal. Accordingly, the above detailed description should not be construed as limiting in all respects, but should be considered illustrative. The scope of the present invention should be determined by rational interpretation of the appended claims, and all changes within the equivalent scope of the present invention are included in the scope of the present invention.

100: speaker recognition system 110: router
120/200: AI speaker 130: Internet network
140: Internet service providing server 300: speaker recognition device
310: voice receiving unit 320: voice feature extraction unit
330: speaker model generation unit 340: similarity measurement unit
350: reference point setting unit 360: speaker identification unit
370: database

Claims (13)

Receiving a voice signal of a home user requesting voice registration;
Measuring similarity between a user voice requesting the voice registration and a pre-registered user voice;
Comparing the measured similarity score with a reference score for determining whether a special group exists; And
And setting a reference score for speaker identification based on the comparison result. The method of claim 1, wherein the receiving step,
A speaker recognition method characterized by receiving the voice signal of the user at home from an AI speaker. The method of claim 1, wherein the setting step,
If the measured similarity score is less than or equal to the reference score for determining whether or not a special group, the speaker recognition method characterized by maintaining the reference score for the speaker identification. The method of claim 1, wherein the setting step,
If the measured similarity score is greater than the reference score for determining whether a special group, the speaker recognition method characterized in that for updating the reference score for the speaker identification. According to claim 4,
The reference score for the speaker identification is updated based on the measured similarity score. According to claim 1,
And extracting feature values of the received voice signal, and generating a speaker model corresponding to the user in the home based on the extracted feature values. According to claim 1,
Receiving a voice signal of a home user requesting speech recognition; And
And measuring the similarity between the user voice requesting the voice recognition and the pre-registered user voice. The method of claim 7,
And comparing the similarity score between the user's voice requesting the voice recognition and the pre-registered user's voice and the set reference score to identify a speaker. A program stored on a computer-readable recording medium such that the method according to any one of claims 1 to 8 is executed on a computer. A voice receiver that receives a voice signal of a user in the home requesting voice registration;
A similarity measuring unit measuring a similarity between the user voice requesting the voice registration and a pre-registered user voice; And
A speaker recognition apparatus including a reference score setting unit that compares the measured similarity score with a reference score for determining whether a special group is set, and sets a reference score for speaker identification. The method of claim 10,
A speaker recognition apparatus further comprising a voice feature extraction unit for extracting feature values of the received voice signal using a predetermined voice feature extraction algorithm. The method of claim 11,
A speaker recognition device further comprising a speaker model generator configured to generate a speaker model corresponding to the user in the home based on feature values of the voice signal. The method of claim 10,
A speaker recognition device further comprising a speaker identification unit for identifying a speaker by comparing the similarity score between the user voice attempting speech recognition and the pre-registered user voices and the set reference score.

Images