KR100738414B1 - Method for improving performance of speech recognition in telematics environment and device for executing the method - Google Patents

Abstract

A method for improving speech recognition performance under telematics environments and an apparatus for performing the method are provided to minimize an error that a recognition result is recognized again by removing a recognition result receiving a negative response about system confirmation of user's speech from a re-recognition process. A wide POI(Point Of Interest) is inputted if recognition of an inputted detailed POI fails. A POI network associated with the wide POI is activated. A POI having the highest similarity with the detailed POI which is inputted again is searched from the POI network and a searched result is provided as a recognition result. When the POI network associated with the wide POI is activated, an N-best POI result of the inputted wide POI is generated and a POI network corresponding to the N-best POI result is activated.

Description

METHOD FOR IMPROVING PERFORMANCE OF SPEECH RECOGNITION IN TELEMATICS ENVIRONMENT AND DEVICE FOR EXECUTING THE METHOD}

1 is a diagram illustrating an example of a step-by-step speech recognition method according to the prior art.

2 is a diagram illustrating an example of a speech recognition method using a system confirm according to the prior art.

3 is a flowchart illustrating an example of a method of telematics speech recognition.

4 is a flowchart illustrating a method for utilizing inter-step N-best POI results according to an embodiment of the present invention.

5 is a flowchart illustrating a method for using information associated with a system confirmation for user utterance in another embodiment of the present invention.

FIG. 6 is a block diagram illustrating an internal configuration of an apparatus using the N-best POI result between steps according to an embodiment of the present invention.

FIG. 7 is a block diagram illustrating an internal configuration of an apparatus using information associated with a system confirmation about user speech according to another embodiment of the present invention.

8 is a diagram illustrating an example of speech recognition using an N-best POI result between steps according to an embodiment of the present invention.

FIG. 9 is a diagram for one example of speech recognition using information associated with system confirmation of user speech in another embodiment of the present invention.

FIG. 10 is a diagram for one example of using additional classification information of POI according to another embodiment of the present invention.

<Explanation of symbols for main parts of the drawings>

620: activator

621: N-best POI result generator

622: POI network activation unit

720: excluded / recognized

721: negative recognition result provider

722: exclude negative recognition result

723: recognition unit

The present invention relates to a method for improving the performance of speech recognition in a telematics environment and an apparatus for performing the method.

In the case of speech recognition in a vehicle environment such as telematics, if the mere relying on the speech recognizer performance, the actual haptic recognition rate is considerably lowered due to noise. This is because the recognition performance decreases due to confusion of similar point of interest (POI) names in the speech recognition process. In particular, in the vehicle environment as described above, the discrimination between POIs is reduced due to noise such as vehicle noise and ambient noise, so that the speech recognition performance degradation of the similar POI name occurs more frequently.

In the prior art, in order to solve the above problems, the user is required to check the utterance contents at every utterance. The recognition target vocabulary is reduced by recognizing the user's utterance and confirmation as described above, and the recognition performance is partially eliminated due to the similar POI name by recognizing using only the search network of each target recognition vocabulary.

1 is a diagram illustrating an example of a step-by-step speech recognition method according to the prior art. According to the prior art as shown in FIG. 1, in one step, only the POI network 103 associated with one POI 102 that is most similar to the user speech 101 is used to obtain the result 104 as the final goal in the next step. Because of the recognition, recognition fails as shown in the example of FIG. 1. In other words, if an error occurs in one step, the error leads to the next step, and eventually a problem of having to go through the speech recognition process from the beginning again occurs.

2 is a diagram illustrating an example of a speech recognition method using a system confirm according to the prior art. According to the related art as shown in FIG. 2, when a negative response is inputted to a system confirm 202 for a user speech 201, a negative recognition result 203 is displayed again in a stepwise recognition process. Occurs repeatedly. In other words, the recognition result confirmed that the user does not want the recognition result by the user confirmation is re-recognized in the re-recognition process has the problem that the same error appears repeatedly.

In addition, in the case of POI names having a wide range among POIs classified in stages, there is a problem in that memory is managed inefficiently because too many detailed POIs are connected to one wide area POI.

The present invention proposes a new technique related to a method for improving the performance of speech recognition in a telematics environment and an apparatus for performing the method, in order to solve the above problems of the prior art.

An object of the present invention is to minimize a step recognition result error caused by an error caused by a similar point of interest (POI) name of the previous step to the current step by using the information of the recognition result of the previous step in the step recognition process. do.

Another object of the present invention is to minimize an error in which the recognition result is re-recognized by excluding a recognition result that has received a negative response to a system confirmation for a user utterance in the re-recognition process.

It is still another object of the present invention to reduce the size of the search network and solve an inefficient use of memory by using additional classification information for large area names.

In order to achieve the above object and to solve the above-mentioned problems of the prior art, the telematics speech recognition method according to an embodiment of the present invention, when the recognition of the input point of interest (POI) fails, Receiving a wide area POI, activating a POI network associated with the wide area POI, and searching for the POI having the highest similarity with the detailed POI re-entered in the POI network and providing the recognition result as a recognition result. do.

According to one aspect of the invention, the step of activating the POI network associated with the wide area POI, generating an N-best POI result of the received wide area POI, and activates the POI network corresponding to the N-best POI It may include the steps to

According to another aspect of the present invention, the N-best POI result may be a list in which N POIs are searched in the active POI network in the order of similarity with the input POIs, and arranged in the searched order. .

According to another aspect of the present invention, the wide area POI is a POI for limiting the range of the detailed POI, and when the wide area POI includes the predetermined number or more of the detailed POI, it may include additional classification information.

Telematics speech recognition method according to another embodiment of the present invention, providing a predetermined system confirmation associated with the recognition result for the input POI, and excludes the recognition result in association with the response of the system confirmation And recognizing.

According to another embodiment of the present invention, the telematics speech recognition apparatus may include a wide area POI input unit receiving a wide area POI, an activation unit for activating a POI network associated with the wide area POI, when recognizing the input detailed POI fails. It may include a recognition result providing unit for searching for the POI having the highest similarity with the input detailed POI in the POI network and providing it as a recognition result.

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

3 is a flowchart illustrating an example of a telematics speech recognition method.

As shown in FIG. 3, in the present invention, when the POI is classified into a wide area, intermediate, or detailed POI and the detailed POI recognition fails (301), the local POI is recognized as a city / county name recognition 302 and the intermediate POI. The target POI 304 is recognized by reducing the range of the POI through step-by-step POI recognition of the old / business type recognition 303. In this case, the wide area POI is a POI for limiting the range of the detailed POI, and when the wide area POI includes the predetermined number of detailed POIs, the wide area POI may include an intermediate POI as additional classification information.

As described above, by using the N-best POI result, which is the recognition result candidate list of the previous step, in the current step between steps, it is possible to minimize the error in the current step due to the error in the previous step.

In addition, a POI provided with a system confirmation for user utterance and a negative response is recognized in the recently registered POI list 305, and if the POI is registered again as a recognition result candidate in a later step, the POI is recalled. By excluding from the recognition result candidate, an error in which the POI is recognized as the recognition result can be minimized.

4 is a flowchart illustrating a method for utilizing inter-step N-best POI results according to an embodiment of the present invention.

In operation S401, the apparatus for recognizing the telematics voice recognizes the POI for the user speech as the detailed POI.

In step S402, the device performs step S409 when the recognition of the detailed POI is successful, and step S403 when the recognition is not successful.

In step S403, the device requests a user to speak a user speech associated with a detailed POI and recognizes the user speech as a wide area POI. In this case, an N-best POI result of the received wide area POI is generated. In addition, the wide area POI is a POI for limiting the range of the detailed POI, and when the wide area POI includes more than a predetermined number of detailed POIs, the wide area POI may include an intermediate POI as additional classification information.

In step S404, the device activates a POI network corresponding to the N-best POI generated in step S403. In this case, the N-best POI result may be a list in which N POIs are searched in the activated POI network in the order of similarity with the inputted wide area POIs, and arranged in the searched order.

In step S405, the device requests the user to speak the user for the detailed POI and re-recognizes the user speech as the detailed POI.

In step S406, the device performs step S409 when the recognition of the detailed POI succeeds and step S408 when the recognition does not succeed.

In step S407, the device provides the user with information about a failure in speech recognition for user speech.

In operation S408, the device provides the user with a result of recognizing the user's speech.

As described above, when an error due to a similar POI name occurs in the step of recognizing the wide area POI in the step-by-step recognition process, in the step of recognizing a detailed POI by the POI network associated with the wide area POI using the N-best POI result It can be prevented from being excluded to minimize the problem that the error caused by the similar POI name of the previous step continues to the current step.

5 is a flowchart illustrating a method for using information associated with a system confirmation for user utterance in another embodiment of the present invention. As shown in FIG. 5, each step of FIG. 5 may be associated with the step of FIG. 4, and steps S504 to S507 of FIG. 5 may be performed by being included in step S405 of FIG. 4. have.

In step S501, the device provides a system confirmation for the user speech when the detailed POI recognition is successful in steps S401 and S402. In this case, the system confirmation may include a question for confirming whether the recognition result of the user utterance is a POI suitable for the user's intention.

In step S502, when the device recognizes the acknowledgment of the system confirmation, the device performs step S407 to provide the recognition result, and when the negative response is received, step S503.

In step S503, the device stores the negative recognition result for the negative response in the recent registered POI list 305 and recognizes the wide area POI in step S403. In this case, step S403 generates a first N-best POI result through the wide area POI recognition, and step S404 activates a POI network associated with the first N-best POI result.

In step S504, the device searches for the second N-best POI of the detailed POI using the POI network to re-recognize the detailed POI.

In operation S505, the device performs operation S506 when the search result and the second N-best POI result include the negative recognition result stored in the recently registered POI list 305 in operation S503. If it is not included, step S507 is performed.

In operation S506, the device performs operation S507 excluding the negative recognition result from the second N-best POI result.

In operation S507, the apparatus recognizes the detailed POI, and provides information on recognition failure (S407) or providing a recognition result (S408) according to whether the recognition is successful (S406).

As described above, when the negative recognition result received a negative response to the system confirmation for the user utterance is stored, and when the stored negative recognition result is included in the second N-best POI result again during the re-recognition process, the second By excluding the negative recognition result from the N-best POI result, an error in which the negative recognition result is re-recognized in the re-recognition process may be minimized.

FIG. 6 is a block diagram illustrating an internal configuration of an apparatus using the N-best POI result between steps according to an embodiment of the present invention. In this case, the telematics speech recognition apparatus may include a wide area POI input unit 610, an activation unit 620, and a recognition result providing unit 630.

When the recognition of the input detailed POI fails, the wide area POI input unit 610 receives the wide area POI. In this case, the wide area POI is a POI for limiting the range of the detailed POI, and when the wide area POI includes more than a predetermined number of detailed POIs, the wide area POI may include an intermediate POI as additional classification information.

The activation unit 620 activates a POI network associated with the wide area POI. In this case, the activation unit 620 may include an N-best POI generation unit 621 and a POI network activation unit 622.

The N-best POI generation unit 621 generates an N-best POI result of the received wide area POI. In this case, the N-best POI result may be a list in which N POIs are searched in the activated POI network in the order of similarity with the input POIs, and arranged in the search order.

The POI network activator 622 activates a POI network corresponding to the N-best POI. In this case, even if the wide area POI associated with the detailed POI is not selected due to an error by the similar POI name, the wide area POI network is activated by the N-best POI, so that the detailed POI recognition can succeed.

The recognition result providing unit 630 searches for the POI having the highest similarity with the re-entered detailed POI in the POI network and provides the result as a recognition result.

FIG. 7 is a block diagram illustrating an internal configuration of an apparatus using information associated with a system confirmation about user speech according to another embodiment of the present invention. In this case, the telematics speech recognition apparatus may include a system confirmation provider 710 and an exclusion / recognition unit 720.

The system confirmation provider 710 provides a predetermined system confirmation associated with the recognition result of the input POI. In this case, the system confirmation may include a question for confirming whether the recognition result is a POI suitable for a user's intention.

The exclusion / recognition unit 720 recognizes the result of the recognition in association with the response of the system confirmation. In this case, the exclusion / recognition unit 720 may include a negative recognition result storage unit 721, a negative recognition result exclusion unit 722, and a recognition unit 723.

When a negative response to the system confirmation is recognized, the negative recognition result storage unit 721 stores the recognition result as a negative recognition result.

The negative recognition result exclusion unit 722 excludes the negative recognition result from the N-best POI result when the negative recognition result is included again in the N-best POI result in the re-recognition process.

The recognition unit 723 recognizes the N-best POI from which the negative recognition result is excluded.

8 is a diagram illustrating an example of speech recognition using an N-best POI result between steps according to one embodiment of the present invention.

As shown in FIG. 8, in the step of recognizing the wide area POI, an N-best POI result 802 is generated for the wide area POI 801 and a POI network 803 associated with the N-best POI result 802 is generated. Activate it. In this case, since the POI is searched using the activated POI network 803 in the step of recognizing the detailed POI 804, even if an error due to a similar POI occurs in the step of recognizing the wide area POI, the desired detailed POI 804 ) Can be recognized.

FIG. 9 is a diagram for one example of speech recognition using information associated with system confirmation of user speech in another embodiment of the present invention.

As shown in FIG. 9, when a similar POI name is recognized for a desired POI name 901, a system confirm 902 is provided to receive a negative response by the user, a similar POI corresponding to the system confirm 902 is provided. The target POI name 901 can be recognized by storing the name and 904 excluding the similar POI name from the N-best POI 903 upon re-recognition.

FIG. 10 is a diagram for one example of using additional classification information of POI according to another embodiment of the present invention. As shown in FIG. 10, when the wide area POI 1002 for limiting the range of the detailed POI 1001 includes a certain number of detailed POIs 1001, the range of the wide area POI 1002 is too large, further classification. Information 1003 may be included.

As described above, the size of the search network can be reduced by using additional classification information for large area names, and the problem of inefficient use of memory can be solved.

The method for improving the performance of speech recognition in a telematics environment according to the present invention may be implemented in the form of program instructions that can be executed by various computer means and recorded in a computer readable medium. The computer readable medium may include program instructions, data files, data structures, etc. alone or in combination. Program instructions recorded on the media may be those specially designed and constructed for the purposes of the present invention, or they may be of the kind well-known and available to those having skill in the computer software arts. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tape, optical media such as CD-ROMs, DVDs, and magnetic disks, such as floppy disks. Magneto-optical media, and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like. The medium may be a transmission medium such as an optical or metal wire, a waveguide, or the like including a carrier wave for transmitting a signal specifying a program command, a data structure, or the like. Examples of program instructions include not only machine code generated by a compiler, but also high-level language code that can be executed by a computer using an interpreter or the like. The hardware device described above may be configured to operate as one or more software modules to perform the operations of the present invention, and vice versa.

In the present invention as described above has been described by the specific embodiments, such as specific components and limited embodiments and drawings, but this is provided to help a more general understanding of the present invention, the present invention is not limited to the above embodiments. For those skilled in the art, various modifications and variations are possible from these descriptions.

Therefore, the spirit of the present invention should not be limited to the described embodiments, and all the things that are equivalent to or equivalent to the claims as well as the following claims will belong to the scope of the present invention. .

According to the present invention, the sensory recognition rate can be maximized through the effective design of the user interface, not merely depending on the performance of the voice recognition device.

According to the present invention, an error caused by a similar point of interest (POI) name of the previous step is generated up to the current step by using the information of the recognition result of the previous step in the step recognition process of the telematics speech recognition. It is possible to improve the performance of speech recognition by minimizing the error of the recognition result.

According to the present invention, if the recognition result received a negative response to the system confirmation (System Confirm) for the user utterance, and if the recognition result is included again in the re-recognition process, the recognition result by excluding the recognition result, Can improve the performance of speech recognition by minimizing the recognition errors that are recognized again.

According to the present invention, it is possible to reduce the size of the search network and solve an inefficient memory use problem by using additional classification information for a large area name.

Claims (13)

In the telematics speech recognition method, If the recognition of the inputted detailed point of interest (POI) fails, receiving a wide area POI; Activating a POI network associated with the wide area POI; And Searching for the POI having the highest similarity with the detailed POI re-entered in the POI network and providing the result as a recognition result Telematics speech recognition method comprising a. The method of claim 1, Activating a POI network associated with the wide area POI, Generating an N-best POI result of the received wide area POI; And Activating a POI network corresponding to the N-best POI Telematics speech recognition method comprising a. The method of claim 2, The N-best POI result is a list of retrieval of the N POIs in the activated POI network in the order of similarity with the input POIs, and a list arranged in the search order. The method of claim 1, The wide area POI is a POI for limiting the range of the detailed POI, and when the wide area POI includes the predetermined number or more of the detailed POI, the telematics speech recognition method may include additional classification information. In the telematics speech recognition method, Providing a predetermined System Confirm associated with a recognition result for the input POI; And Recognizing except for the recognition result in association with the response of the system confirmation Including, Recognizing except for the recognition result in association with the response of the system confirmation, Storing the recognition result as a negative recognition result when a negative response to the system confirmation is recognized; Excluding the negative recognition result from the N-best POI result when the negative recognition result is included again in the N-best POI result during the recognizing process; And Recognizing using the N-best POI from which the negative recognition result is excluded Telematics speech recognition method comprising a. delete The method of claim 5, The system confirmation, the telematics speech recognition method characterized in that it comprises a question for confirming whether the recognition result is a POI in accordance with the user's intention. A computer-readable recording medium in which a program for executing the method of any one of claims 1 to 5 or 7 is recorded. In the telematics speech recognition device, A wide area POI input unit configured to receive a wide area POI when recognition of the input detailed POI fails; An activation unit for activating a POI network associated with the wide area POI; And Recognition result providing unit for retrieving the POI having the highest similarity with the detailed POI re-entered in the POI network and providing it as a recognition result. Apparatus comprising a. The method of claim 9, The activator, An N-best POI result generator for generating an N-best POI result of the received wide area POI; And POI network activation unit for activating the POI network corresponding to the N-best POI Apparatus comprising a. The method of claim 10, And the N-best POI result is a list in which N POIs are searched in the active POI network in the order of similarity with the input POIs, and arranged in the search order. In the telematics speech recognition device, A system confirmation provider for providing a predetermined system confirmation associated with the recognition result of the input POI; And Exclusion / recognition unit for recognizing the result of the recognition in association with the response of the system confirmation Including, The exclusion / recognition unit, A negative recognition result storage unit that stores the recognition result as a negative recognition result when a negative response to the system confirmation is recognized; A negative recognition result exclusion unit that excludes the negative recognition result from the N-best POI result when the N-best POI result is included again in the N-best POI result during the recognizing process; And Recognizing unit using the N-best POI without the negative recognition result Apparatus comprising a. delete

Images