KR100994930B1 - Adaptive voice recognition control method for voice recognition based home network system and the system thereof - Google Patents

Abstract

Disclosed are an adaptive speech recognition control method of a speech recognition based home network system, and a system thereof. An adaptive voice recognition control method of a voice recognition-based home network system according to the present invention includes an internet network, a wall pad connected to the internet network and a home network to perform calls and various control functions, and a voice recognition connected to the wall pad. An adaptive speech recognition control method performed in a speech recognition home network, comprising: a speech recognition unit configured to perform a; and a management server connected to the Internet network to manage an adapted speech recognition model in the speech recognition unit. a) the wall pad accepting a request for the start of the adaptation from the user, (b) the speech recognition unit performing the adaptation in response to the step (a) to generate an adaptation model for the user, and (c) Transmitting the adaptive model generated in step (b) from the wall pad to the management server through the Internet; and (d) the management server. (c) verifying the recognition rate using sufficient speaker's speech data for the adaptive model transmitted through the internet network; and (e) managing the recognition rate if the recognition rate is inappropriate in step (d). The server instructing the deletion of the adaptation model.

 Home network, voice recognition, adaptation, rejection rate

Description

Adaptive voice recognition control method for voice recognition based home network system and the system

The present invention relates to a method and a system for controlling adaptive speech recognition of a speech recognition based home network system. More particularly, the present invention relates to speaker dependent speech recognition using an adaptation technique to supplement speaker independent speech recognition in a speech recognition based home network system. The present invention relates to a method and system for adaptive speech recognition control of a voice recognition-based home network system for efficiently improving a problem of low rejection rate that may occur when using the apparatus.

In general, a home network system refers to a system for controlling lighting, air conditioning, curtains, gas valves, and other security devices in a home by connecting them to one network. Such a home network system includes a user interface called a wall pad that performs a call and various control functions, a device control device in which various devices in the home are connected and monitored and controlled, a door phone installed outside the entrance and / or It is composed of a lobby phone installed in the common entrance to manage access, a guard room phone installed in the guard room, a server for the maintenance and management of the complex and a server for the information communication network.

However, the conventional home network system as described above has a problem in that the user interface is complicated and difficult to use, and thus general users are limited to the interphone function. In order to solve this problem, a voice recognition home network system has been developed to add a speaker independent speech recognition function to a user interface of a home network system and to drive various functions of the home network by speech.

Speech recognition can be divided into a technique using a speaker-independent standard speech model and a speaker-dependent adaptive speech model. The standard speech model is obtained by extracting a common feature from a target word or sentence to be recognized from speech data recorded by a voice of an unspecified majority speaker and using the extracted feature as a reference model for speech recognition. However, all the individual speakers differ slightly in terms of speech habits and intonation, so that the recognition rate is different in speech recognition using the same speaker-independent speech recognition model, and the recognition rate is significantly lower in the case of a person having an unusual tone or speech habit. It can also happen.

In other words, the speaker-independent speech recognition technology also has a significantly lower recognition performance in the case of a person with an unusual speech habit or intonation. In order to improve this problem, the adaptation technique is used as a technique to raise the recognition rate to the level of the general public through a separate speech recognizer learning process even for a person whose recognition rate is significantly lower. Thus, it may be possible to use a modified recognition model optimized for a specific speaker, that is, a speech recognition model combined with a speech recognition model adapted to a standard speech recognition model in order to have a certain recognition rate to a certain speaker.

However, in the adaptation, an adaptation model for a specific speaker may be used in addition to the reference model, and the recognition rate may be improved. However, it is difficult to control the rejection rate and increase the recognition of the error by leaving the developer's hand. In addition, a large sized database storing command voices spoken by a large number of speakers is required, and a complicated control process for verifying recognition rate using the voices stored in the database is complicated. There is a problem that is difficult. In addition, in order to verify the rejection rate, a considerable amount of noise and test time are required, as well as the complexity of the processing program, there is a problem that it is not easy to implement the user to verify directly on the wall pad.

The present invention has been invented to solve the above problems, the technical problem to be achieved by the present invention is to solve the problem that may occur when applying the adaptation technology to increase the speaker independent recognition rate in the voice recognition home network system and It is to provide a voice recognition home network system that can optimize the recognition rate.

An adaptive speech recognition control method of a voice recognition based home network system according to an aspect of the present invention for solving the technical problem,

A wall pad connected to an internet network, the internet network and a home network to perform calls and various control functions, a voice recognition unit connected to the wall pad to perform voice recognition, and a voice recognition connected to the internet network An adaptive speech recognition control method performed in a speech recognition home network comprising a management server that manages an adapted speech recognition model in a unit, the method comprising: (a) receiving, by a wall pad, a start request for adaptation from a user; (b) the speech recognition unit performing adaptation in response to step (a) to generate an adaptation model for the corresponding user; (c) transmitting the adaptation model generated in step (b) from the wall pad to the management server through the internet network: (d) the management server is applied to the adaptation model transmitted through the internet network in step (c) Verifying the recognition rate using enough speaker's speech data; And (e) if it is verified that the recognition rate is inappropriate in step (d), the management server instructing the deletion of the adaptation model.

The method may further include (f) verifying the rejection rate by using the living noise data of a sufficient time or more, if the recognition rate is determined to be appropriate in step (e); (g) if it is determined that the rejection rate is not suitable in step (f), the management server generating an adaptive model supplemental patch including a filler model required to secure the rejection rate; (h) the management server transmitting an adaptation model supplemental patch to the wall pad through the Internet; And (i) the wall pad receiving the adaptive model supplemental patch transmitted in step (h) and the speech recognition unit updating the adaptive model using the received adaptive speech model supplemental patch. .

In addition, the adaptive voice recognition control system of the voice recognition-based home network system according to another aspect of the present invention for solving the above technical problem, in the home network system using the voice recognition, to the Internet network and home network devices A wall pad that is connected and includes a guide that provides a user with a call and various controls, the set of instructions for adaptation; A first module which processes speech recognition by being connected to the wall pad and a predetermined adaptive speech recognition algorithm based on speaker-dependent speech recognition in response to the guidance of the wall pad generate an adaptive model specialized for the user's speech. And transmits the generated adaptive speech model to the management server through the communication network, and performs the adaptive speech model management process including deleting the adaptive model and / or updating the adaptive model using the adaptive model complementary patch using the verification result from the management server. A speech recognition unit having a second module; And a test voice database for storing speech data for a command to be recognized by a large number of unspecified plural speakers, a feature extractor for extracting features by importing voices stored in the test voice database, A comparison search unit for comparing and searching using an adaptive model, a determination unit for determining whether a recognized voice corresponds to a predefined command according to the search and comparison result in the comparison search unit, and the comparison search unit; And a management server including a verification unit configured to verify the recognition rate based on the recognition result of the determination unit, and to generate an adaptation model deletion command for deleting the adaptation model as a verification result and to transmit it to the wall pad. The wall pad may recognize the voice when receiving the adaptive model delete command. In it characterized in that to control the removal of the adaptation model.

In addition, the test voice database stores voice data for a large number of unspecified multiple speaker speech recognition commands and living noise data for a sufficiently long time, and the verification unit is further configured to recognize the recognition results of the comparison search unit and the determination unit. Based on the verification of the recognition rate, if it is determined that it is not appropriate, the generation of the adaptive model deletion command to delete the adaptive model is generated as a verification result and transmitted to the wall pad. Generates an adaptive model supplemental patch including a verification result and transmits it to a wall pad, wherein the wall pad controls the speech recognition unit to delete the adaptive model and receives the adaptive model deleting command and receives the adaptive model deleting patch. When receiving the adaptive model in the speech recognition unit Using the completed patch is also possible to control so as to update an adaptive model.

The adaptive voice recognition control method and system of the voice recognition based home network system according to the present invention verify that the recognition rate is secured by using a sufficient number of speaker's voice data in the management server, so that a predetermined criterion, for example, at least 90% or more If it is properly recognized, the adaptive speech model is adopted as an adaptive speech model that complements the standard speech model. If not, the adaptive speech model is considered to be too different from the standard speech model, so that it is highly likely to be misidentified. This reduces the likelihood of false recognition occurring in the process. That is, the problem that may occur when applying adaptive speech recognition by wall pad alone is compensated for.
Optionally, if the recognition result is not rejected and is misrecognized, the management server transmits an adaptation model supplemental patch to complement the adaptation model including a filler model that is not misrecognized by the management server. By including a rejection rate verification process for updating a speech model using a model complementary patch, it is possible to more efficiently alleviate the problem of a decline in rejection rate that may occur in the process of improving the recognition rate.

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

1 is a block diagram illustrating a schematic structure of a system for performing an adaptive speech recognition control method of a speech recognition based home network system according to the present invention. Referring to FIG. 1, a system for performing an adaptive voice recognition control method of a voice recognition-based home network system according to the present invention is connected to an internet network 1 and home network devices 104 and includes a call and various controls. A wall pad (10) comprising a guide (102) for providing a set of instructions for adaptation to a user;

The first module 122 which processes speech recognition by being connected to the wall pad 10 and the speaker dependent speech recognition based on the speaker-dependent speech recognition based on the standard speech recognition model 124 in response to the guidance of the wall pad 10. By performing a predetermined adaptive speech recognition algorithm, an adaptive model specific to a user's speech is generated, and the generated adaptive speech model 128 is transmitted to the management server 14 through the Internet network 1 and from the management server 14. A speech recognition unit 12 including a second module 126 for performing an adaptive speech model management process including deleting an adaptive model and / or updating an adaptive model using an adaptive model supplemental patch using the verification result, and

A test voice database 142 which stores voice data for a command to be recognized by a large number of unspecified multiple speakers, a feature extracting unit 144 which extracts a feature by importing voices stored in the test voice database 142, and extracts A comparison search unit 146 for comparing the received feature with the adaptive model 145 transmitted through the Internet, and a voice corresponding to a predetermined command as a recognized voice according to the search and comparison result of the comparison search unit 146. The adaptive model for determining the recognition and the adaptive model to delete the adaptive model if it is determined that it is not appropriate by verifying the recognition rate based on the recognition results of the comparison search unit 146 and the determination unit 148. The management server 14 includes a verification unit 150 that generates a delete command delete_adaptmodel as a verification result and transmits the result to the wall pad.

In addition, the test voice database 14 preferably stores voice data for a large number of unspecified multiple speaker speech recognition commands and living noise data for a sufficiently long time, and the verification unit 150 performs the comparison search. Based on the recognition result of the unit 146 and the determination unit 148, the recognition rate is verified, and if it is determined that it is not appropriate, an adaptive model delete command (delete_adaptmodel) for deleting the adaptation model is generated as a verification result and used as a wall pad. In addition to the transmission, if the recognition rate is verified and determined to be appropriate, an adaptive model complement patch (complementpatch_adaptmodel) including a filler model is generated from the data in which misrecognition occurs as a verification result and transmitted to the wall pad 10. The wall pad 10, in response to receiving the adaptive model delete command (delete_adaptmodel) in the speech recognition unit 12 When the control unit deletes the adaptation model and receives the adaptation model supplement patch (complement_adaptmodel), it is more preferable to control the speech recognition unit 12 to update the adaptation model using the adaptation model supplement patch (complement_adaptmodel).

In such a system, the adaptive voice recognition control method of the voice recognition-based home network system according to the present invention is distributed in the wall pad 10, the voice recognition unit 12, and the management server 14. Such a control method is implemented by storing and retrieving a program, program code, code segments, and executing a series of steps processed in each component and can be programmed by those skilled in the art.

2 is a flowchart illustrating main processing steps of an adaptive speech recognition control method of a speech recognition based home network system according to the present invention. 1 is often referenced below. 2, the adaptive voice recognition control method of the voice recognition-based home network system according to an embodiment of the present invention is connected to the Internet network 1, the Internet network 1 and the home network to make calls and various control functions. The voice recognition unit 12 connected to the wall pad 10 to perform voice recognition, and the voice recognition unit 12 connected to the Internet network 1. An adaptive speech recognition control method performed in a speech recognition home network comprising a management server 14 for managing an adapted speech recognition model,

(a) step S200 of receiving, by the user, the wall pad 10 from the user to start the adaptation;

(b) the speech recognition unit 12 performing adaptation in response to step (a) (S200) to generate an adaptation model for the corresponding user (S202);

(c) transmitting (S204) the adaptive model generated in step (b) to the management server 14 through the internet network 1 in the wall pad;

(d) the management server 14 verifying the recognition rate using enough speaker's speech data with respect to the adaptive model transmitted through the Internet network 1 in step (c) (S204), And

(e) if it is verified that the recognition rate is inappropriate in step (d) (S206), the management server 14 includes a step (S208) of generating an adaptive model delete command (delete_adaptmodel) instructing the deletion of the adaptive model. .

3 is a diagram for explaining in detail the process of updating the speech recognition model by adaptation. The process described with reference to FIG. 3 corresponds to step (b) of FIG. 2. Referring to Fig. 3, the voice data stored by the voice of a particular speaker whose recognition rate is significantly lowered, preferably voiced for a predetermined number of sentences, is recorded 304 through input signal processing 302 such as level adjustment. The feature is extracted 308 from 306 and modeled 310 to update 312 the speech recognition model. The updated voice recognition model 314 includes a voice model, a Lexicon, and a voice object ID. Here, the process described above to the standard speech model that extracts the standard features of the constituent phonemes of the target word or sentence to be recognized from the speech data recorded by the voice of the unspecified majority speaker by updating the speech recognition model and serves as a reference model for speech recognition. The adaptive speech model modeled by is reflected.

However, the adaptive speech model that updates the standard speech model through the adaptation process as described above is a speech model that has not been verified by the developer. There is a problem of being high. Therefore, according to the present invention, the adaptive voice model generated by the adaptation process in the wall pad 10 and the voice recognition unit 12 is transmitted to the management server 14 and the management server 14 verifies the recognition rate with respect to the adaptive voice model. If it is determined that the recognition rate is not appropriate, an adaptive model delete command (delete_adaptmodel) is generated to transmit the adaptation model to the wall pad 10. The process of verifying the recognition rate is done using enough speaker, e.g., at least 100 voice data for one instruction. Voice data is pre-stored in the test voice database 142 of the management server 14, and if a predetermined criterion, i.e., the target performance is 90%, more than 90 data should show recognition success in the voice recognition test by 100 speaker data. do. That is, the user with low recognition rate manipulates the wall pad 10 to apply the adaptive speech model generated through the adaptation process, so that the management server 14 uses the speech data of a sufficient number of speakers stored in the test speech database 142. If at least 90% of speech data is properly recognized by verifying that the recognition rate is secured, the adaptive speech model is adopted as an adaptive speech model that complements the standard speech model, but otherwise it is too different from the standard speech model. It is determined that the likelihood is high and the adaptive speech model is deleted.

On the other hand, in the case of the voice recognition that is always operated, it may be recognized as a command also by the sound of phone calls, chats, other noises. In order to prevent such misunderstandings, it is necessary to secure an appropriate rejection rate. In order to secure a rejection rate, a voice recognition is attempted using the noise data of a life long enough to verify whether it is recognized as a command. However, such a living noise data is large and the rejection rate verification process using the living noise data is also not suitable for performing in the wall pad 10. Therefore, according to the present invention, the management server 14 verifies whether the false recognition occurs using the living noise data stored in the test voice database 142 with respect to the speech recognition model generated in the adaptation process, and when the false recognition occurs, Insert a filler model in the form of a patch to reject it.

That is, in the adaptive speech recognition control method of the voice recognition-based home network system according to the present invention, if (f) the recognition rate is determined to be appropriate in step (e) (S206), the management server 14 lives more than enough time. Verifying a rejection rate using the noise data (S220);

(g) if it is determined that the rejection rate is not suitable in step (f) (S220), the management server 14 generates an adaptive model supplemental patch including a filler model required to secure the rejection rate. (S222);

(h) the management server 14 transmitting the adaptation model supplemental patch to the wall pad 10 through the Internet network 1; And

(i) the wall pad receives the adaptive model supplemental patch transmitted in step (h) (S224), and the speech recognition unit 12 updates the adaptive model using the received adaptive speech model supplemental patch (S226). It is more preferable to further include;

Next, FIG. 4 is a diagram illustrating a program processing process for implementing the control method of FIG. 2. Referring to FIG. 4, when the user senses that the user presses the adaptation button provided on the wall pad (S402) and the user presses the adaptation button, the wall pad 10 transmits the adaptation mode initiation protocol to the speech recognition unit 12. The recognition unit 12 receives the protocol and transmits the acknowledgment (ACK) (S408), and the wall pad 10 receives the acknowledgment (ACK) (S410) and proceeds to the adaptation mode (S412). As a result, control for the adaptation mode is started, and the speech recognition unit 12 proceeds to the adaptive operation in response to the control (S414). When the adaptation operation is completed, the speech recognition unit 12 transmits an adaptation operation completion protocol to the wall pad 10 informing that the adaptation operation is completed (S416). The wall pad 10 receives the adaptive calculation completion protocol (S418) and ends the adaptation mode (S420), and the voice recognition unit 12 now switches to the recognition mode for the voice recognition operation.

5A and 5B show an example of an adaptation process that can be applied to the control method of FIG. 4 as a flowchart. 5A and 5B, the wall pad 10 displays a description of adaptation and provides a selection of gender and a module on the screen (S502). It is desirable to provide guidance text, gender selection button display, mode selection button, and next / cancel button in the form of a graphical user interface. If necessary, adaptation descriptions, adaptation notes, and gender selection guides may be included. Next, the user checks the number of adaptations that the user has adapted (S504) and, for example, when the number is greater than 3, provides a selection of whether or not to initialize the model (S506). It is desirable to provide a GUI with a Continue / Cancel button to select whether or not. In other words, it is possible to guide the initiation of the model and the inability to adapt more than three times per gender. On the other hand, if the number of adaptations is less than 3, the method selects whether to perform full adaptation or selective adaptation (S508). When the entire adaptation is selected, after checking the entire instruction list, branching is performed to start adaptation (S510). In the case of selecting selective adaptation, the method branches to step S512 of selecting an instruction to be adapted from the instruction list and then starting the adaptation. For example, if the user checks the green LED, which is provided in the wall pad 10, and the user speaks a command (S520), whether the recognition is performed, the speech length, and the signal-to-noise ratio (SNR) are displayed. It is determined whether the correct utterance is performed based on the ratio) check, the silence length, and the like (S522). At this time, it would be more desirable to further provide a function of playing the recording utterance, a guide function for determining whether or not the utterance. If it is determined that the correct utterance is repeated for the next instruction, and if it is determined that the correct utterance is not the correct utterance is guided to re-speak the same instruction to execute (S524), check whether it is uttered three times (S526) utterance less than three times In one case, the above process is repeated with respect to the current command, and in the case of uttering three times, the utterance is registered by uttering one utterance (S528). When all / selection commands are uttered or even when not all utterances are pressed, when the 'adaptation button' is pressed, the process moves to the adaptation performing step (S530). In addition, when vocalization ends, the adaptation is performed (S540).

On the other hand, the concept disclosed in the present invention should be distinguished from the concept of distributed speech recognition. Distributed speech recognition is a speech recognition in a system of server-client structure through wired / wireless network and is called DSR (Distributed Speech Recognition). In this method, it is essential to execute the preprocessing unit and the recognition unit of speech recognition on different computers, and the preprocessing unit that extracts the speech feature is usually performed on the client computer, and the recognition unit, that is, the language processing process, is performed on the server computer where the large database exists. To be handled. Running the entire process of speech recognition on a client computer usually suffers from insufficient data or processing power. Conversely, running the entire process of speech recognition on a server computer introduces noise and distortion when sending audio data to the server. You will have difficulty. Distributed speech recognition can overcome this problem and is a suitable model for the recent decentralized system environment.

However, according to the present invention, all the speech recognition processes for adaptation are performed by the wall pad corresponding to the client computer alone, but the adaptive speech model is transmitted to the management server and the management server side is applied to the adaptive speech model performed by the wall pad. Completion of the problem that may occur when applying adaptive speech recognition by wall pad alone by generating and transmitting supplemental patches to delete the adaptive speech model by performing recognition rate verification and rejection rate verification only, and to prevent false recognition. It should be noted that

The adaptive speech recognition control method and system for a speech recognition-based home network system according to the present invention as described above are provided with a management server, and the recognition rate is determined using a test speech database in which a sufficient number of speaker's voice data is stored in advance in the management server. Verification rate is verified by using the recognition rate verification process to verify and delete the adaptive model that has low recognition rate in the verification result, and the test voice database that stores the living noise data for a sufficient time in advance. In this case, the management server transmits an adaptation model supplemental patch to supplement the adaptation model, including the filler model, and the wall pad and the speech recognition unit reject the update of the speech model using the adaptation model supplemental patch. Po rate verification process By this rejection rate that may occur in the process of improving the recognition rate, but that improves the problem efficiently.

1 is a block diagram schematically showing the structure of a system for performing an adaptive speech recognition control method of a speech recognition based home network system according to the present invention;

2 is a flowchart illustrating main processing steps of an adaptive speech recognition control method of a speech recognition based home network system according to the present invention;

3 is a diagram for explaining in detail the process of updating the speech recognition model by adaptation;

4 is a view for explaining a program processing procedure for implementing the control method of FIG.

5A and 5B are flowcharts illustrating an example of an adaptation process that may be applied to the control method of FIG. 4.

Claims (4)

A wall pad connected to an internet network, the internet network and a home network to perform calls and various control functions, a voice recognition unit connected to the wall pad to perform voice recognition, and a voice recognition connected to the internet network An adaptive speech recognition control method performed in a speech recognition home network comprising a management server that manages an adaptive speech recognition model in a part, (a) the wall pad accepting a request for initiation of the adaptation from the user; (b) generating an adaptive model specialized for the user's utterance by performing a recording process for adaptation according to the guidance of the wall pad and performing a predetermined adaptive speech recognition algorithm based on speaker-dependent speech recognition with the recorded data. ; (c) transmitting the adaptive model generated in step (b) from the wall pad to the management server through the Internet; (d) the management server retrieves a voice stored in a test voice database storing voice data of a command to be recognized by a plurality of speakers, extracts a feature, and extracts the extracted feature and the adaptive model transmitted through the Internet in step (c). Verifying the recognition rate based on the recognition result for comparing and searching using and determining whether the recognized speech is a voice corresponding to a predefined command according to the comparison and search result; And (e) if it is verified that the recognition success rate is less than a predetermined criterion in speech recognition using the adaptive model in step (d), instructing the management server to delete the corresponding adaptive model generated in the wall pad; An adaptive speech recognition control method of a voice recognition based home network system, characterized in that the. The method of claim 1, (f) If it is verified in step (e) that the recognition success rate is higher than or equal to a predetermined criterion in the speech recognition using the adaptive model of step (d), the management server rejects the living noise data previously stored in the management server. Verifying the rate; (g) an adaptation including a filler model required by the management server to secure the rejection rate if it is determined in step (f) that the rejection rate, which is a rate at which the living noise data is not recognized and rejected as a command, is less than a predetermined criterion; Generating a model supplemental patch; (h) the management server transmitting an adaptation model supplemental patch to the wall pad through the Internet; And (i) the wall pad receives the adaptive model supplemental patch transmitted in the step (h) and the speech recognition unit updates the adaptive model using the received adaptive speech model supplemental patch; Adaptive Speech Recognition Control Method of Recognition-based Home Network System. A wall pad including an internet network, a guide unit connected to the internet network and a home network, and providing a series of guides necessary for adaptation to the user together with a call and various control functions, and a voice pad connected to the wall pad to perform voice recognition. A voice recognition home network system comprising a voice recognition unit and a management server connected to the Internet network to manage an adapted voice recognition model in the voice recognition unit, A first module which processes speech recognition by being connected to the wall pad and a predetermined adaptive speech recognition algorithm based on speaker-dependent speech recognition in response to the guidance of the wall pad generate an adaptive model specialized for the user's speech. And transmit the generated adaptive speech model to the management server through a communication network and perform the adaptive speech model management process including the deletion of the adaptation model or the update of the adaptation model using the adaptation model supplement patch using the verification result from the management server. A speech recognition unit having two modules; And Using a test voice database for storing voice data of a command to be recognized by a plurality of speakers, a feature extractor for retrieving features by retrieving voices stored in the test voice database, an extracted feature and an adaptive model transmitted through the Internet A comparison search unit for comparing and searching, a determination unit determining whether a recognized voice corresponds to a predefined command according to the search and comparison result in the comparison search unit, and the comparison search unit and the determination unit In the case of the voice recognition using the adaptive model based on the recognition result of the recognition result that the recognition success rate is less than a predetermined criterion includes a verification unit for generating an adaptive model deletion command to delete the adaptive model as a verification result and transmits it to the wall pad A management server; The wall pad, And the voice recognition unit controls to delete the adaptive model when the adaptive model deletion command is received. The method of claim 3, The test voice database, Saves voice and living noise data for multiple speaker recognition commands; The verification unit, An adaptive model deletion command for verifying a recognition rate based on the recognition result of the comparison search unit and the determination unit and deleting the adaptive model when the recognition success rate is less than a predetermined criterion in speech recognition using the adaptive model. Generated as a verification result and transmitted to the wall pad, and when the recognition rate is verified and it is verified that the recognition success rate is higher than or equal to a predetermined criterion in the speech recognition using the adaptive model, an adaptive model supplemental patch including a filler model is included from the data in which false recognition occurs. Generated as a verification result and sent to the wall pad, The wall pad, When the adaptive model deletion command is received, the voice recognition unit controls to delete the corresponding adaptive model, and when the adaptive model complementary patch is received, the voice recognition unit controls to update the adaptive model using the adaptive model complementary patch. Voice recognition home network system.

Images