KR101483945B1 - Method for spoken semantic analysis with speech recognition and apparatus thereof - Google Patents

Abstract

The present invention relates to a voice recognizing method capable of analyzing a meaning and a voice recognizing device therefor. The voice recognizing method and the voice recognizing device are capable of: generating a combination network by searching for a meaning at the same time searching for a voice; and analyzing the meaning at the same time recognizing the voice for voice data using the combination network. To this end, according to an embodiment of the present invention, the voice recognizing device comprises: a storing unit to store an acoustic model, a language model, and a meaning model; and a voice recognizing unit to search for a voice using the acoustic model and the language model for inputted voice data, to search for a meaning using the acoustic model and the meaning model, to combine nodes according to the voice search and the meaning search to generate a combination network, and to output voice recognition results and meaning analysis results by performing a final search using the combination network.

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention [0001] The present invention relates to a speech recognition method,

The present invention relates to a speech recognition method capable of performing semantic analysis and a speech recognition apparatus for the same, and is capable of performing semantic search simultaneously with voice search to generate a combined network, And more particularly, to a speech recognition method capable of proceeding with semantic analysis and a speech recognition apparatus therefor.

The contents described in this section merely provide background information on the present embodiment and do not constitute the prior art.

In a speech recognition system, a search process is performed by expressing a target area to be recognized as a search network and finding a word string most similar to an input speech signal (speech data) within the condition of the search space .

There are many ways to create a search space. Among them, a method using a weighted finite state transducer (WFST) is widely spreading. WFST regards a language model, a pronunciation dictionary, a context-dependent acoustic model, and HMM state network as one large weighted automata, and extends the existing automata theory to a weighted automata to apply to speech recognition .

In WFST-based algorithms, when all networks are represented by automata, the algorithm that combines these automata, the crystallization algorithm, and the minimization algorithm that combines the common paths to generate the optimal path are used. .

However, in the speech recognition using the conventional WFST, only the input voice is used for converting to text, and the process of analyzing the meaning is performed using another method such as Natural Language Processing using the converted text I have to. Therefore, if an error occurs in the speech recognition, it is difficult to cope with the error in the language processing belonging to the post-processing.

Korean Patent Laid-Open No. 10-2008-0026951, March 26, 2008 (Name: Speech Recognition Method for Robust Long-Range Speech Recognition System)

As described above, the present invention has been proposed in order to solve the problems of the prior art. The present invention creates a combined network by performing a semantic search simultaneously with a voice search, and uses the same to perform semantic analysis There is provided a speech recognition method capable of proceeding with meaning analysis and a speech recognition apparatus therefor.

However, the object of the present invention is not limited to the above-mentioned objects, and other objects not mentioned can be clearly understood from the following description.

According to an aspect of the present invention, there is provided a speech recognition apparatus including a storage unit for storing an acoustic model, a language model, and a semantic model; And performing a voice search using the acoustic model and the language model with respect to the input voice data, performing a semantic search using the acoustic model and the semantic model, And a speech recognition unit for generating a combined network, performing a final search using the combined network, and outputting a speech recognition result and a semantic analysis result.

The storage unit may further store a pronunciation dictionary, and the speech recognition unit may perform a voice search using the acoustic model, the language model, and the pronunciation dictionary.

At this time, the semantic model may include at least one semantic pattern information including a morpheme and a semantic tag.

Also, the connection network may be at least one of a finite state network (FSN), a word-pair grammar, and an n-gram.

The voice recognition unit may extract the feature data from the input voice data and perform the voice search and the meaning search according to the extracted feature data when the voice data is input.

In addition, the speech recognition unit may perform a final search using the combined network and the acoustic model.

In addition, the speech recognition unit may generate a speech recognition result by weighting the nodes according to the meaning search.

According to another aspect of the present invention, there is provided a speech recognition method capable of performing semantic analysis according to an embodiment of the present invention. Performing speech search using the acoustic model and the language model on the speech data, and performing semantic search using the acoustic model and the semantic model in parallel; Combining the voice search and the nodes according to the semantic search performed by the voice recognition device in parallel to generate a combined network; And performing the final speech recognition and semantic analysis using the combined network and the acoustic model by the speech recognition device.

In this case, the step of extracting the feature data from the voice data by the voice recognition device may be performed after the step of receiving the voice data.

In addition, in the parallel performing step, the speech recognition apparatus may perform a voice search and a semantic search on the feature data.

In addition, in the parallel performing step, the speech recognition apparatus may weight a node according to the semantic search to proceed with speech recognition.

The method further comprises: after the step of performing the final speech recognition and the semantic analysis, the speech recognition device identifies the corresponding object information using a semantic tag included in the node according to the semantic search and processes the result as an analysis result; May be further included.

According to the speech recognition method capable of analyzing the meaning of the present invention and the speech recognition apparatus for the same, semantic search is performed simultaneously with speech search to generate a combined network, thereby enabling speech recognition of speech data as well as semantic analysis And does not perform a separate procedure for semantic analysis, so that the semantic analysis result can be provided more quickly.

FIG. 1 is an exemplary diagram for schematically explaining an operation of a speech recognition apparatus according to an embodiment of the present invention.
2 is a block diagram illustrating a main configuration of a speech recognition apparatus according to an embodiment of the present invention.
3 is a block diagram for explaining a main configuration of a speech recognition unit according to an embodiment of the present invention.
4 is a block diagram illustrating a main configuration of a storage unit according to an embodiment of the present invention.
5 is a flowchart for explaining a speech recognition method capable of semantic analysis according to an embodiment of the present invention.
6 and 7 are diagrams for explaining a speech recognition method capable of semantic analysis according to an embodiment of the present invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. However, the detailed description of known functions and configurations incorporated herein will be omitted when it may unnecessarily obscure the subject matter of the present invention. This is to omit the unnecessary description so as to convey the key of the present invention more clearly without fading. While the invention is susceptible to various modifications and alternative forms, specific embodiments thereof are shown by way of example in the drawings and will herein be described in detail. However, it should be understood that the invention is not limited to the specific embodiments thereof, It is to be understood that the invention is intended to cover all modifications, equivalents, and alternatives falling within the spirit and scope of the invention.

In addition, when referring to an element as being "connected" or "connected" to another element, it means that it can be connected or connected logically or physically. In other words, it is to be understood that although an element may be directly connected or connected to another element, there may be other elements in between, or indirectly connected or connected.

Also, the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The singular expressions include plural expressions unless the context clearly dictates otherwise. It is also to be understood that the terms such as " comprising "or" having ", as used herein, are intended to specify the presence of stated features, integers, It should be understood that the foregoing does not preclude the presence or addition of other features, numbers, steps, operations, elements, parts, or combinations thereof.

Now, a speech recognition method capable of semantic analysis according to an embodiment of the present invention and a speech recognition apparatus therefor will be described in detail with reference to the drawings. Here, the same reference numerals are used for similar functions and functions throughout the drawings, and a duplicate description thereof will be omitted.

FIG. 1 is an exemplary diagram for schematically explaining an operation of a speech recognition apparatus according to an embodiment of the present invention.

Referring to FIG. 1, the speech recognition apparatus 100 of the present invention performs speech recognition based on input speech data and outputs recognition results when speech data is input from a user. At this time, the speech recognition apparatus 100 of the present invention can simultaneously output not only speech recognition results, but also semantic analysis results for the speech data. A more specific method of operating the speech recognition apparatus 100 will be described later, and the speech recognition apparatus 100 of the present invention can be implemented as one independent apparatus. When the speech recognition apparatus 100 is implemented as a single independent apparatus that can be used by a user, the speech recognition apparatus 100 may include an input module including a microphone capable of sensing voice signals input from a user and generating voice data, And a display module capable of outputting a speech recognition result and a semantic analysis result.

Also, the speech recognition apparatus 100 of the present invention may be implemented as an embedded type device in a specific hardware device. In this case, the speech recognition apparatus 100 can receive voice data from an input module provided in the apparatus, and can transmit the result to a display module capable of outputting a meaning analysis result and a speech recognition result. At this time, the speech recognition apparatus 100 may be implemented in the form of a program such as an application. For example, it can be embedded in various electronic devices available to the user such as a smart phone, or installed in the device in a program form.

In addition, the speech recognition apparatus 100 of the present invention may be implemented as a web server. When the voice recognition apparatus 100 is implemented as a web server, the user can generate voice data in the form of a file using a user terminal such as a smart phone of her own and transmit the voice data to the voice recognition apparatus 100 via the communication network , The voice recognition apparatus 100 can receive voice data transmitted from the user terminal through the communication network. In addition, the speech recognition apparatus 100 may transmit the result of the semantic analysis and the speech recognition to the user terminal via the communication network, and the user terminal that receives the result may output the received result through the display module. In addition, the speech recognition apparatus 100 may operate in conjunction with a web server supporting a specific service such as language learning, call classification, or may be formed integrally with the web server. In addition, the speech recognition apparatus 100 of the present invention may be implemented as a dual processing system in which a module for receiving a voice and a module for recognizing a voice are hardware-classified.

Hereinafter, the main configuration and operation method of the speech recognition apparatus 100 according to the embodiment of the present invention will be described in more detail.

2 is a block diagram illustrating a main configuration of a speech recognition apparatus according to an embodiment of the present invention.

Referring to FIG. 2, the speech recognition apparatus 100 according to the embodiment of the present invention may include an interface unit 10, a speech recognition unit 20, and a storage unit 30.

As described above, the interface unit 10 receives the user's voice data and transfers the voice data to the voice recognition unit 20. The voice recognition unit 20 performs a semantic analysis And a voice recognition result to the user. In this case, when the speech recognition apparatus 100 is implemented as an independent apparatus, the interface unit 10 generates a voice signal of a user input through a microphone as voice data in the form of digital data, And to output the semantic analysis result through a separate display module. In addition, when the speech recognition apparatus 100 is embedded in a specific hardware device or implemented in a form of a program, or in a form of a web server, the interface unit 10 may include voice data transmitted via a separate interface or a wire / And transmits the speech recognition result and the semantic analysis result.

The speech recognition unit 20 recognizes the input speech data and generates a speech recognition result when the speech data is inputted through the interface unit 10. [ In particular, the speech recognition unit 20 according to the embodiment of the present invention recognizes speech data to generate a speech recognition result, and at the same time performs semantic analysis on the speech data and generates a result of the analysis.

The storage unit 30 stores and manages various information for speech recognition and semantic analysis of the present invention. The storage unit 30 may be a flash memory, a hard disk, a memory of a multimedia card micro type (e.g., SD or XD memory), a RAM, a ROM ), And the like.

The speech recognition unit 20 and the storage unit 30 of the present invention will be described in more detail with reference to FIGS. 3 and 4. FIG.

FIG. 3 is a block diagram for explaining a main configuration of a speech recognition unit according to an embodiment of the present invention, and FIG. 4 is a block diagram for explaining a main configuration of a storage unit according to an embodiment of the present invention.

3, the speech recognition unit 20 of the present invention includes a feature extraction module 21, a voice search module 22, a meaning search module 23, a combined network creation module 24, (25).

The feature extraction module 21 extracts feature data from the input voice data and reflects the hearing characteristics of the user and transmits the extracted feature data to the voice search module 22 and the meaning search module 23.

To this end, the feature extraction module 21 first performs an analog to digital conversion (ADC) process for converting a voice signal, which is a continuous analog voice signal, into a discrete voice data value in a digital form. The feature extraction module 21 extracts the feature data based on the voice data converted into the digital form. Here, the feature data means voice and acoustical feature data in the frequency domain of the voice data converted into the digital form. For example, the following parameters may be used: negative duration, negative energy, pitch, power, linear predictive coding (LPC) coefficients, compositional phonemes, formant, RFC Connection / Tilt, Spectrum, VOT (Voice Onset Time), and the like can be extracted as feature data.

The feature data may be any one of a Mel-Frequency Cepstrum Codec (MFCC), a Linear Prediction Coefficient Cepstrum (LPCC), a Preceptual Linear Prediction Ceptrum Coeffcient (PLPCC), an Ensemble Interval Histogram (EIH) Lt; / RTI >

The voice search module 22 performs voice search using the acoustic model database 31 and the language model database 32 of the storage unit 30 and transmits the discovered nodes to the join network creation module 24. At this time, the voice search module 22 may convert the notation phoneme into a pronunciation phoneme using the pronunciation dictionary database 33 storing the pronunciation dictionary. More specifically, the voice search module 22 compares the feature data of the voice data received through the feature extraction module 21 with the acoustic model stored in the acoustic model database 31 first. Thereafter, a phoneme string corresponding to the feature data is extracted, and a language corresponding to the extracted phoneme string is extracted through the language model database 32. Through this process, one or more nodes that can be displayed in the combined network can be extracted.

The semantic search module 23 performs a semantic search using the acoustic model database 31 and the semantic model database 32 and delivers the searched nodes to the association network creation module 24. [ More specifically, the meaning search module 23 compares the feature data of the voice data received through the feature extraction module 21 with the acoustic model stored in the acoustic model database 31 first. Thereafter, the phoneme string corresponding to the feature data is extracted, and the semantic pattern information corresponding to the extracted phoneme string is extracted using the semantic model database 32. Through this process, one or more nodes that can be displayed in the combined network can be extracted.

The combined network creation module 24 creates a combined network by combining the nodes derived from the voice search module 22 and the nodes exposed from the meaning search module.

Here, the combined network may be at least one of a finite state network (FSN), a word-pair grammar, and an n-gram. Here, the term "network" refers to words that may appear after a unit, which are fixed by rules or connected by statistical probabilities.

Here, the word-pair grammar is a method of connecting only words that can appear after a specific word, for example, searching by using something that can be connected in the order of "eat" + "want" .

The N-gram utilizes statistical probabilities for linking words to words, and calculates a probability that a word appears after one word by using learning data, thereby searching for a higher probability.

On the other hand, a finite state network (FSN) is a network that bundles all configurable sentences. It consists of a state with a unique name and a transition, which is an operation that changes the state of the FSN.

Hereinafter, for convenience of description, it is assumed that the combined network is an FSN, but the present invention is not limited thereto.

The combined network generation module 24 generates a final combined network by combining nodes when a node according to the search is applied from the voice search module 22 and a node according to the search is transmitted from the meaning search module 23 . At this time, a network including a plurality of vocabulary classes and a transition between the plurality of nodes can be constructed by adding, deleting or changing one or more vocabularies having the same semantical concept in segmented morpheme units.

The final search module 25 may then generate the final speech recognition result and the semantic analysis result using the combined network. In other words, the final search module 25 performs the final search of the feature data of the voice data using the combined network and the acoustic model.

For example, the voice search module 22 may derive one or more vocabulary nodes as a result of the voice search.

Uh, uh, um, membership, point, go, uh, how much, man, it was, okay, where, uh, uh, uh, uh, uh, uh ...

In addition, the semantic search module 23 can derive one or more lexical nodes as a result of the semantic search.

Membership [Service Name], Point [Service Name], How much [Quantity], Remain [Remain], Write [Consumption], Where [Place], Write [Consumption]

The voice search module 22 and the semantic search module 23 may combine the derived one or more nodes. The combining network generation module 24 may use the nodes to create a network. The combined network creation module 24 then passes the generated nodes to the final search module 25.

The final search module 25 performs speech recognition and semantic analysis using the combined network generated through the combined network creation module 24. [ At this time, the final search module 25 extracts the phoneme string by comparing the feature data of the voice data with the acoustic model of the acoustic model database 31, and extracts the voice recognition result and semantic information as a set of nodes corresponding to the extracted phoneme string The analysis result can be generated.

At this time, the final search module 25 assigns weights to the nodes derived through the semantic search module 23, and transmits the nodes derived through the semantic search module 23, i.e., You can also create a sentence around the center.

Then, the final search module 25 may transmit the generated speech recognition result and semantic analysis result to the interface unit 10 and provide the result to the user.

In this case, if any node constituting the combined network includes a semantic tag, the final search module 25 of the voice recognition apparatus 100 confirms object information corresponding to the semantic tag, Processed and output. Here, the object information may include any one of an object type, an object object, and an object attribute.

Meaning tag Object Information Inquiry, confirmation, request, invitation, ... Object type Service name, place, ... Object target Remaining, Limit, Move, Max, Close, ... Object Properties

For example, if the semantic tag is a service name, the final search module 25 can identify the object information (object object) corresponding to the service name and process it as an analysis result. At this time, it can be processed in the form of "object type: semantic tag".

The storage unit 30 includes an acoustic model database 31, a language model database 32, a pronunciation dictionary database 33, and a semantic model database 34 as described above.

The acoustic model database 31 stores and manages acoustic models statistically modeling phonemes. At this time, the acoustic model may be an HMM (hidden Markov Model), and the basic unit of the acoustic model may be a phoneme string. Therefore, the phoneme string for the feature data can be extracted through the acoustic model database 31. [

The language model database 32 stores and manages the language model, and supports the selection of sentences matching grammar rather than arbitrary sentences in learning and searching. Here, the language model may be implemented as a network of at least one of an FSN, a word-pair grammar, and an n-gram.

The pronunciation dictionary database 33 stores and manages the pronunciation dictionary. A pronunciation dictionary defines a simple rule based on the standard pronunciation method, or a definition that takes into account the characteristics of a specific environment, a speaker and dialect.

The semantic model database 24 stores and manages semantic models. At this time, the semantic model may be generated according to a LSP (Lexico-Synthetic Pattern). A process of generating a semantic model according to an LSP will be briefly described. First, one or more morpheme candidates are generated in one sentence, and a morpheme is selected in consideration of dictionary search and word formation rules for the generated candidates. Then, the part mark tag is analyzed in the selected morpheme. Then, one or a plurality of morphemes are combined, and one or more morphemes are assigned a semantic tag. Since the process of generating a semantic model according to the LSP uses a known technique, a detailed description thereof will be omitted, and the semantic pattern information stored and managed in the semantic model database 24 may include one or a plurality of morphemes And a semantic tag. For example, it can be stored in the form of membership [service name], point [service name], amount [quantity], remaining [remaining], consumption [consumption], place [ ], For example, "membership" means the morpheme of the smallest unit of language. [Service name] is a meaning tag indicating what the morpheme (membership) means. For another example, the "maximum", "near", "near", "near", and "near" with respect to "gas station" and "Seoul" Quot; proximity " can be assigned to the " close "

The main configuration of the speech recognition apparatus 100 according to the embodiment of the present invention has been described above.

The speech recognition apparatus 100 according to an embodiment of the present invention includes only the interface unit 10, the speech recognition unit 20 and the storage unit 30, (100) may further include a preprocessing unit (not shown) and a post-processing unit (not shown).

At this time, the preprocessing unit (not shown) may perform preprocessing of the inputted voice data to be suitable for voice recognition. For example, unnecessary noise cancellation, voice enhancement, and the like can be performed. The post-processing unit (not shown) corrects the spacing and spelling errors with respect to the speech recognition result, aligns the consistency of the foreign word display, and deletes the speech if the speech that can not be discriminated is included.

Hereinafter, a speech recognition method capable of semantic analysis according to an embodiment of the present invention will be described.

5 is a flowchart for explaining a speech recognition method capable of semantic analysis according to an embodiment of the present invention.

2 and 5, a speech recognition method capable of semantic analysis according to an exemplary embodiment of the present invention includes first extracting feature data from the speech data when the speech recognition apparatus 100 receives speech data (S101) (S103). In this case, the speech recognition apparatus 100 may be configured to perform various tasks such as Mel-Frequency Cepstrum Coefficients (MFCC), Linear Predictive Cepstral Coefficients (LPCC), Ensemble Interval Histogram (EIH), Short-Time Modified Coherence (SMC), and Perceptual Linear Prediction Feature data can be extracted by any one technique.

Then, the speech recognition apparatus 100 performs a voice search using an acoustic model and a language model, and performs a semantic search using an acoustic model and a semantic model (S105).

That is, the speech recognition apparatus 100 first compares the feature data with an acoustic model, extracts a phoneme string corresponding to the feature data, extracts a language corresponding to the extracted phoneme string through a language model, And extracts at least one corresponding node.

At the same time, the speech recognition apparatus 100 compares the feature data with an acoustic model, extracts a phoneme string corresponding to the feature data, extracts semantic pattern information corresponding to the extracted phoneme string through a semantic model, Extract one or more nodes according to semantic search.

Thereafter, the speech recognition apparatus 100 creates a combined network using the derived node through step S105 (S107). Here, the combined network may be at least one of a finite state network (FSN), a word-pair grammar, and an n-gram.

Then, the speech recognition apparatus 100 performs a final voice search and a semantic search using the combined network (S109). At this time, the speech recognition apparatus 100 extracts a phoneme string by comparing the feature data with an acoustic model, and generates a speech recognition result and a semantic analysis result as a set of nodes corresponding to the extracted phoneme string.

Here, the speech recognition apparatus 100 may assign a weight to nodes derived through semantic searching and generate a sentence based on the weighted nodes.

Then, the speech recognition apparatus 100 can output the speech recognition result and the semantic analysis result (S111).

In this case, if any node constituting the combined network includes a semantic tag, the speech recognition apparatus 100 confirms the object information corresponding to the semantic tag, and processes it as an analysis result. Here, the object information may include any one of an object type, an object object, and an object attribute.

Referring to FIGS. 6 and 7, FIGS. 6 and 7 are diagrams for explaining a speech recognition method capable of semantic analysis according to an embodiment of the present invention.

That is, an example of a state transition diagram in which a predetermined sentence is expressed in the form of an FSN type network includes a node for a total of six states and seven transitions. In addition, for convenience of explanation, each node includes only a single word, but is not limited thereto, and may be a list of similar words. For example, a node including the word "writing" may further include words similar to "writing ", such as " writing ", " writing"

Also, for convenience of description, an example including only six states and seven transitions is described as an example, but may be made up of more nodes and transitions.

Through this network, the following sentence can be generated.

Membership points are a while away

However, as described above, if the node of the network includes more words and more transitions are made, various sentences such as the following may be generated.

How many membership points are there?
How many membership points are there?
How much is your membership?
How much time did you spend writing your membership points?
Where do I write my membership points?
Where do I write my membership points?
Membership Points Well, where do I write it?

In addition, FIG. 7 shows a state in which only a general voice search is performed and a network is created using only the nodes. FIG. 8 shows a state in which a network is created using a node according to semantic search together with voice search will be.

In other words, as described above, only a general voice search of FIG. 7 can be used to generate only the "membership point is short" by speech recognition result. However, the combination network according to the embodiment of the present invention is shown in FIG. As shown in the figure, further includes nodes according to the semantic search, i.e., "membership [service name] "," point [service name] ", " And the result of the semantic analysis such as "query type: inquiry, query target: membership point, target attribute: residual" is extracted together with the speech recognition result of " That is, as described above, the speech recognition apparatus 100 confirms the semantic tags included in the node, verifies the object information corresponding to the semantic tags, and outputs the processed result as an analysis result.

As described above, the present invention performs semantic search simultaneously with voice search to generate a combined network, thereby making it possible to perform semantic analysis as well as voice recognition on voice data at the same time, and does not perform a separate procedure for semantic analysis There is an excellent effect that the semantic analysis result can be provided more quickly.

The speech recognition method capable of semantic analysis according to the embodiment of the present invention has been described above.

The speech recognition method according to the embodiment of the present invention can be implemented as a computer-readable code on a computer-readable recording medium. The computer readable recording medium may include program instructions, data files, data structures, and the like, alone or in combination, and includes all kinds of recording apparatuses in which data that can be read by a computer system is stored. Examples of the computer-readable recording medium include an optical recording medium such as a magnetic medium such as a hard disk, a floppy disk and a magnetic tape, a compact disk read only memory (CD-ROM), and a digital video disk (ROM), random access memory (RAM), flash memory, and the like, such as a magneto-optical medium such as a magneto-optical medium and a floppy disk, And hardware devices that are specifically configured to perform the functions described herein.

In addition, the computer-readable recording medium may be distributed over network-connected computer systems so that computer readable codes can be stored and executed in a distributed manner. In addition, functional programs, codes, and code segments for implementing the present invention can be easily inferred by programmers of the technical field to which the present invention belongs.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed exemplary embodiments, but, on the contrary, It will be appreciated by those skilled in the art that numerous changes and modifications can be made to the invention. And all such modifications and changes as fall within the scope of the present invention are therefore to be regarded as being within the scope of the present invention.

The present invention relates to a speech recognition method capable of performing semantic analysis and a speech recognition apparatus for the same, and is capable of performing semantic search simultaneously with voice search to generate a combined network, And more particularly, to a speech recognition method capable of proceeding with semantic analysis and a speech recognition apparatus therefor.

According to the present invention, a semantic search is performed simultaneously with a voice search to create a combined network. By using this, a voice recognition for voice data as well as a semantic analysis can be performed at the same time. It is possible to provide quick semantic analysis results, which can contribute to the development of the speech recognition industry.

In addition, since the present invention is not only possible to be marketed or operated, but also can be practically and practically carried out, it is industrially applicable.

10: interface unit 20: voice recognition unit 21: feature extraction module
22: Voice search module 23: Meaning search module
24: Combined Network Generation Module 25: Final Search Module
30: storage unit 31: acoustic model database
32: language model database 33: pronunciation dictionary database
34: semantic model database 100: speech recognition device

Claims (12)

A storage unit for storing an acoustic model, a language model, and a semantic model; And
A voice search is performed on the input voice data using the acoustic model and the language model, a semantic search is performed using the acoustic model and the semantic model, and then the voice search and the node according to the semantic search are performed A speech recognition unit for generating a combined network by performing a final search using the combined network to output a speech recognition result and a semantic analysis result;
/ RTI >
Wherein the speech recognition unit generates a speech recognition result by weighting nodes according to the semantic search when performing a final search using the combined network. The method according to claim 1,
The storage unit
More pronunciation dictionary,
The speech recognition unit
Wherein the speech recognition apparatus performs a speech search using the acoustic model, the language model, and the pronunciation dictionary. The method according to claim 1,
Wherein the semantic model includes at least one semantic pattern information including a morpheme and a semantic tag. The method according to claim 1,
The combining network
Wherein the network is at least one of a Finite State Network (FSN), a word-pair grammar, and an n-gram. The method according to claim 1,
The speech recognition unit
Extracts the feature data from the input voice data when voice data is input, and performs the voice search and the meaning search according to the extracted feature data. The method according to claim 1,
The speech recognition unit
And performs a final search using the combined network and the acoustic model. delete Receiving speech data from the speech recognition apparatus;
Performing speech search using the acoustic model and the language model on the speech data, and performing semantic search using the acoustic model and the semantic model in parallel;
Combining the voice search and the nodes according to the semantic search performed by the voice recognition device in parallel to generate a combined network; And
The speech recognition apparatus performing final speech recognition and semantic analysis using the combined network and the acoustic model;
/ RTI >
Wherein the speech recognition apparatus performs a speech recognition by assigning a weight to a node according to the semantic search when the final speech recognition is performed using the combined network and the acoustic model. 9. The method of claim 8,
After receiving the input,
Extracting feature data from the speech data;
The speech recognition method comprising the steps of: 10. The method of claim 9,
The step of performing in parallel
Wherein the speech recognition apparatus performs a speech search and a semantic search for the feature data. delete 9. The method of claim 8,
After performing the final speech recognition and semantic analysis,
Identifying the corresponding object information using the semantic tag included in the node according to the semantic search and processing the result as an analysis result;
The speech recognition method comprising the steps of:

Images