KR101042499B1 - Apparatus and method for processing speech recognition to improve speech recognition performance - Google Patents

Abstract

The present invention relates to a speech recognition processing apparatus and a method for improving speech recognition performance. The present invention relates to a speech recognition process at a low cost by performing a speech recognition process by limiting a range of a pronunciation dictionary based on user evaluation of a speech recognition result. We want to significantly improve performance (the speed and success rate of speech recognition).

To this end, the present invention, the speech recognition processing apparatus for improving the speech recognition performance, the user interface means for receiving a user evaluation for the user speech or speech recognition results for speech recognition from the user; And a voice recognition means for repeating the voice recognition for the input user voice while limiting the range of the pronunciation dictionary according to the user evaluation, and providing the result of the voice recognition to the user. do.

Speech Recognition, Speech Recognition Performance, Pronunciation Dictionary, Corrected Pronunciation Dictionary, Interaction, DTMF

Description

Speech recognition apparatus and method for improving speech recognition performance {APPARATUS AND METHOD FOR PROCESSING SPEECH RECOGNITION TO IMPROVE SPEECH RECOGNITION PERFORMANCE}

The present invention relates to the field of speech recognition technology, and more particularly, to perform speech recognition processing through interaction with a user, which can significantly improve speech recognition performance at low cost, and speech recognition processing for improving speech recognition performance. An apparatus and a method thereof are provided.

Convenience and necessity of speech recognition are sympathetic to many people, and are always introduced as one of the 10 technologies of the future, and much effort is being made to improve the performance of such speech recognition. In particular, the voice recognition technology for applying to a voice recognition service that provides a voice recognition function through the user interface is still being developed.

However, current voice recognition technology has a lot of difficulties in providing the level of service desired by consumers in the market. The reason is that speech recognition basically performs speech recognition on a speech recognition target based on a pronunciation dictionary.

In detail, in the service scenario, voice data input by a user in a service scenario is changed into a probabilistic representation of the digitized data of the voice in the voice recognition module, and then the pre-trained phoneme-based Probability values and probability values for the speech recognition target word based on the given pronunciation dictionary are compared with the probability values for the input voice.

As a result of the comparison, the nearest target word is found in the pronunciation dictionary and output as a voice recognition result. That is, since speech recognition is approaching probabilistic, there is a probability that even 0.0001% is wrong, which means that the quality of service is different. Therefore, due to such a problem, there is a limit of speech recognition even in today's evolving speech recognition technology.

The above problem is an essential problem to be solved in the word-based speech recognition service. That is, in the situation where the limitation of the voice recognition technology exists, even if a new service is developed and provided through the existing voice recognition method, it will eventually fail to satisfy the user's needs.

1 is a flowchart of a conventional voice recognition service method, and illustrates a method performed by a voice recognition processing device (also referred to as a voice recognition server).

In the conventional voice recognition service, as shown in FIG. 1, after a brief description of the service (service overview guidement operation) is performed to a user who is connected to the service (telephone connection) 100 (102), the service scenario Require a voice input for the progress of a particular step (104).

Accordingly, when a user voice for voice recognition is input from the user (106), voice recognition is performed on the input user voice (108). After the voice recognition is performed, the voice recognition result is presented to the user (110) and the user is selected whether it is correct or not (112).

As a result of checking whether the voice recognition result is correct (112), if the voice recognition result is correct, the service scenario corresponding to the voice recognition result is continued (114). On the other hand, if it is wrong, the number of voice recognition failures so far is three times, for example, after driving the service termination message to the user and ending (116), but if the number of voice recognition failures is less than two times, The voice input is requested again (104), and the subsequent process is repeatedly performed.

As described above, in the case of repeating speech recognition through voice re-entry, if the pronunciation dictionary, which is a domain for speech recognition, was 10,000 words, the speech recognition is continuously performed in the range of 10,000 words whenever repeated. After all, even if the speech recognition is repeated according to the retry, the speech recognition is likely to fail the first time, the second time, or the third time.

Since the prior art as described above confirms whether the voice recognition result is correct or incorrect, the voice recognition is performed again from the beginning. Even though the voice recognition process is repeated several times, the result of the voice recognition is the processing capability of the voice recognition processing server. There is a problem that depends on the (Computing Power) or the size of the pronunciation dictionary.

That is, even if repeated processing is not performed, the voice recognition result will not be significantly changed unless the processing capacity of the voice recognition server or the size of the pronunciation dictionary is improved.

Here, the processing capacity of the speech recognition processing server and the size of the pronunciation dictionary have a great influence on the speech recognition performance. However, once it is established, a task for improving the speech recognition cost is enormous.

Therefore, especially for telecommunication carriers (eg, Katie Co., Ltd.) that provide services throughout the country, there is an urgent need to improve voice recognition performance without expanding processing capacity or pronunciation dictionary size to reduce the cost of system construction. It is required.

Accordingly, the present invention provides a speech recognition processing apparatus and method for improving speech recognition performance, which can improve speech recognition performance without expanding the computing power of the speech recognition processing server or the pronunciation dictionary. There is this.

The objects of the present invention are not limited to the above-mentioned objects, and other objects and advantages of the present invention, which are not mentioned above, can be understood by the following description, and will be more clearly understood by the embodiments of the present invention. Also, it will be readily appreciated that the objects and advantages of the present invention may be realized by the means and combinations thereof indicated in the claims.

In order to achieve the above object, the present invention is characterized by performing a voice recognition process through interaction with a user.

That is, the present invention is characterized in that the speech recognition process is performed again by limiting the range of the pronunciation dictionary based on the user evaluation of the speech recognition result.

More specifically, the present invention provides a speech recognition processing apparatus for improving speech recognition performance, comprising: user interface means for receiving a user evaluation of a user speech for speech recognition or a result of speech recognition from a user; And a voice recognition means for repeating the voice recognition for the input user voice while limiting the range of the pronunciation dictionary according to the user evaluation, and providing the result of the voice recognition to the user. do.

The present invention also provides a voice recognition processing method for improving voice recognition performance, comprising: a first voice recognition step of performing a voice recognition by receiving a user voice for voice recognition from a user; A user evaluation step of receiving a user evaluation by providing a voice recognition result of the user voice to the user; And a second speech recognition step of redefining the pronunciation dictionary by limiting the range of the pronunciation dictionary according to the user evaluation and then re-performing the speech recognition process for the user's voice.

The above invention has the effect of significantly improving the speech recognition performance at low cost by performing the speech recognition process through interaction with the user. That is, the present invention can remarkably improve the speed and accuracy (success rate) of speech recognition by re-performing the speech recognition process by limiting the range of the pronunciation dictionary based on user evaluation of the speech recognition result, thereby further improving to consumers. It is effective to provide better quality voice recognition service.

In addition, the present invention, it is possible to significantly increase the speech recognition performance by only a small modification of the existing speech recognition system without expansion of the pronunciation dictionary or processing power (Computing Power), thereby improving the performance of the existing speech recognition system In addition to this, it is possible to activate the development of a new service that provides a voice recognition interface.

The above objects, features, and advantages will become more apparent from the following detailed description taken in conjunction with the accompanying drawings, whereby those skilled in the art may easily implement the technical idea of the present invention. There will be. In the following description, well-known functions or constructions are not described in detail since they would obscure the invention in unnecessary detail. Hereinafter, a preferred embodiment of the present invention will be described in detail with reference to the accompanying drawings.

2 is a block diagram of an embodiment of a speech recognition processing apparatus (server) for improving speech recognition performance according to the present invention.

Speech recognition processing apparatus (server) according to the present invention, as shown in Figure 2, the user interface unit 221, the user voice DB 222, voice recognition unit 223, the first pronunciation dictionary DB 224, and It includes a modified pronunciation dictionary DB (225). Hereinafter, each component will be described in detail.

The user interface unit 221 receives a user voice for voice recognition from the user terminal 20 connected through the telephone network 21 or provides various announcements to the user terminal 20 and accordingly to the voice recognition result. Performs a function to receive an evaluation of the evaluation. Here, the user terminal refers to a terminal capable of transmitting user voice, such as a wired telephone, a mobile communication terminal, and an internet telephone. In addition, the telephone network 21 refers to a network connected to the user terminal 20 and capable of transmitting a user's voice to the voice recognition processing device (server) 22, which includes a mobile communication network, a public telephone network, an IP network, and the like. Included.

The user voice database (DB) 222 stores user voices received through the user interface unit 221. The reason for storing the user voices is that the user voices are newly generated even when the voice recognition is re-defined by redefining the pronunciation dictionary. This is to use the user's voice initially input without input. That is, once the user's voice is input, the voice recognition processing apparatus (server) repeatedly performs the voice recognition process several times while redefining the pronunciation dictionary for the input user voice.

The voice recognition unit 223 performs a voice recognition on the user voice input through the user interface unit 221 and provides a corresponding result (voice recognition result), but the evaluation input from the user (evaluation of the voice recognition result) According to the phonetic dictionary, the phonetic dictionary is re-defined by reproducing the phonetic dictionary.

That is, the voice recognition unit 223 performs voice recognition on the user's voice using the first pronunciation dictionary 224 and provides the result to the user terminal 20. In response, when the user terminal transmits an evaluation of the result (user evaluation) to the speech recognition result, the speech recognition unit 223 redefines the pronunciation dictionary based on the received user evaluation and defines the pronunciation dictionary (hereinafter, The voice recognition is re-executed using 'correction pronunciation dictionary' (225).

Next, when the evaluation (user evaluation) method for the speech recognition result is described, the user evaluation is performed by pressing a number of user terminal buttons corresponding to the number of syllables normally recognized among the speech recognition results. That is, when the user presses the button of the number 3, it is converted into a DTMF (Dual-Tone Multi-Frequency) signal and transmitted to the voice recognition apparatus (server) 22, and the voice recognition unit 223 transmits the DTMF thus transmitted. The "three syllables" in the speech recognition result through the signal is to determine that the speech recognition was made normally.

The evaluation method is not limited to the evaluation method by the above button input. It is also possible for the user to directly speak "the number of syllables recognized normally" by voice (evaluation method by voice input).

Meanwhile, referring to a process of generating a corrected pronunciation dictionary, the speech recognition unit 223 excludes the speech recognition result, which is the target of user evaluation, up to the number of syllables (the number of syllables in which speech recognition is normally performed) indicated by the user evaluation. Redefine the scope of the phonetic dictionary up to words containing letters.

For example, when the user says "Hong Gil-dong", the voice recognition unit 223 performs voice recognition on the first pronunciation dictionary and, as a result, provides "Hong Gil-seo" as a voice. When the user who listens to the voice recognition result presses the button of the number 2 (because it was normally recognized until "Hong Gil"), the voice recognition unit 223 excludes the voice recognition result ("Hong Gil Seo") that was the subject of user evaluation. , And all words beginning with the word "Hong Gil" (eg, Hong Gil Nam, Hong Gil Dong, Hong Gil North, etc.). The pronunciation dictionary in which the object is limited thus becomes a corrected pronunciation dictionary.

3 is a flowchart illustrating an example of a speech recognition processing method for improving speech recognition performance according to the present invention, and illustrates a method performed by a speech recognition processing apparatus (server) as shown in FIG. 2.

First, the speech recognition processing method according to the present invention will be described generally.

After performing the voice recognition on the user's voice input for the voice recognition, the user is asked how many times the recognition result is correct.

If the user presses the button 2 when the second syllable is correct, the word is searched in the pronunciation dictionary to include the second syllable of the speech recognition result and the searched content (the word string including the second syllable) is searched. Redefine as a temporary pronunciation dictionary. Then, voice recognition is performed on the pre-stored user's voice using the redefined pronunciation dictionary.

After the voice recognition is re-executed, the result is again shown to the user, and the number of syllables is checked. If all the voice recognition results are correct, the user jumps to the corresponding menu. Jump to a routine.

For example, the present invention will be described below.

Suppose a user utters "future research institute". At this time, if the voice recognition result is "future technology", the user presses the button four times.

Then, the speech recognition processing apparatus searches for a word included in the pronunciation dictionary of 10,000 words starting with "future technology" and redefines a new pronunciation dictionary for the found speech recognition target word.

Then, the voice recognition engine is initialized using the redefined pronunciation dictionary as the input of the voice recognition engine, and the voice recognition is performed again using the previously stored user voice as the input of the voice recognition engine. In this case, the user does not need to input the voice again, and the recognition rate can be remarkably improved since the number of the words to be recognized is greatly reduced.

Hereinafter, a voice recognition processing method according to the present invention will be described in detail with reference to FIG. 3.

The voice recognition processing device 22 makes a brief description of the service (service overview guidement drive) to the user terminal dialed 300 (302), and then requests the user for voice input for voice recognition (304). .

Accordingly, when the user voice for voice recognition is input from the user, the voice recognition processing apparatus 22 stores the input user voice (306) and performs voice recognition on the user voice (308).

After performing the speech recognition (308), the speech recognition processing apparatus 22 checks whether the speech recognition result is correct while hearing the speech recognition result to the corresponding user (310). Specifically, it will give a comment like "Press # if the voice recognition result is correct, or press an asterisk (*) if it is incorrect".

The user's response confirms whether the voice recognition result is correct (312), if correct, performs the corresponding service scenario (314), and if wrong, checks whether the number of voice recognition errors so far corresponds to the reference value (for example, 4 times). (316).

If the number of speech recognition errors is less than 4, the message that induces the evaluation (detailed evaluation) of the speech recognition results ("the number of syllables in the comment" is checked and the button corresponding to the number of correct syllables). Press ”) (318). According to an embodiment, the process of checking whether the voice recognition result is correct 312 and the process of providing a comment 318 for inducing user evaluation may be implemented through one process. In other words, if the voice recognition result is correct, press #, and if it is wrong, check the number of syllables and press the button corresponding to the number of syllables. You can do it accordingly. If the DTMF signal input by the user terminal indicates "0", the voice recognition is all wrong, and thus the feedback will be fed back to "304" in order to receive re-entry.

The user who listens to the comment that induces user evaluation evaluates the voice recognition result. The evaluation method includes a button input evaluation method and a voice input evaluation method. The evaluation method will be described as an example.

The speech recognition processing device 22 receives the DTMF signal from the user terminal 20 to check the "normally recognized syllable number" (320), and corresponds to the confirmed syllable number (the number indicated by the DTMF signal). The word including the syllable is found in the pronunciation dictionary and redefined as a new pronunciation dictionary (corrected pronunciation dictionary) (322).

The speech recognition processing apparatus 22 re-initializes the speech recognition engine based on the redefined pronunciation dictionary (corrected pronunciation dictionary) (324), and performs the speech recognition by inputting the user's speech stored at "306". After 326, the process of "310" or less is repeated by feeding back to "310".

On the other hand, if the voice recognition error check result 316, if the voice recognition error number is four times, the user confirms the voice re-entry intention (328), if there is a re-entry intention feedback to "304", and re-enter If there is no intention, the service termination message is driven (330). Here, in the process of confirming the voice re-entry intention, the voice recognition processing device 22 "The voice recognition error has exceeded the standard number of times, press # to re-enter the voice recognition target word, and to terminate the asterisk (* Press ") to confirm re-entry.

As described above, the present invention can be applied to various services using voice recognition, such as a service proceeding while selecting a service menu of each stage through voice recognition, an automatic telephone connection service using voice recognition, and the like.

On the other hand, the voice recognition processing apparatus and method for improving the voice recognition performance according to the present invention is to perform the user voice and user evaluation from the user terminal 20 connected to the network, such as a telephone network as shown in FIG. Although it may be implemented in a "server-client" manner of input, depending on the embodiment is not limited to this may also be implemented through a "single voice recognition processing device".

That is, when implemented as a "single system" such as a user terminal (mobile phone, PDA, electronic dictionary, etc.) or various information processing devices having a voice recognition function, the user's voice is input from the user or operator through the input unit of the single system. Input and voice recognition is performed. In this case, user evaluation (user evaluation of the voice recognition result) is performed through various inputs such as a button input, a touch input (for a touch pad and a touch screen), a voice input, and the like in the single system.

On the other hand, the method of the present invention as described above can be written in a computer program. And the code and code segments constituting the program can be easily inferred by a computer programmer in the art. In addition, the written program is stored in a computer-readable recording medium (information storage medium), and read and executed by a computer to implement the method of the present invention. The recording medium may include any type of computer readable recording medium.

That is, the present invention provides a first voice recognition function for performing voice recognition by receiving a user voice for voice recognition from a user for a voice recognition processing apparatus having a processor for voice recognition processing for improving voice recognition performance. ; A user evaluation function of receiving a user evaluation by providing a voice recognition result of the user voice to the user; And limiting the range of the pronunciation dictionary according to the user evaluation, and redefining the pronunciation dictionary, and then executing a second speech recognition function for re-executing the speech recognition process for the user's voice. To provide.

The present invention described above is capable of various substitutions, modifications, and changes without departing from the technical spirit of the present invention for those skilled in the art to which the present invention pertains. It is not limited by the drawings.

1 is a flow chart for a conventional voice recognition service method,

2 is a configuration diagram of an embodiment of a speech recognition processing apparatus (server) for improving speech recognition performance according to the present invention;

3 is a flow chart of an embodiment of a speech recognition processing method for improving speech recognition performance according to the present invention.

* Explanation of symbols on the main parts of the drawings

20: user terminal 21: telephone network

22: speech recognition device 221: a user interface unit

222: user voice DB 223: speech recognition unit

224: first pronunciation dictionary DB 225: modified pronunciation dictionary DB

Claims (13)

In the speech recognition processing device for improving the speech recognition performance, User interface means for receiving a user's evaluation of a user's voice or a voice recognition result for voice recognition from a user; And Speech recognition means for repeating the speech dictionary by limiting the range of the pronunciation dictionary in accordance with the user evaluation while repeating the speech dictionary for the input user voice, to provide the speech recognition result to the user Speech recognition processing device comprising a. The method of claim 1, The user rating is, Speech recognition processing device, characterized in that indicating the number of syllables normally recognized in the speech recognition results. The method of claim 2, The voice recognition means, And a speech recognition device that redefines the scope of the phonetic dictionary to words including letters up to the number of syllables indicated by the user evaluation and excludes the voice recognition result that is the target of the user evaluation. The method of claim 1, The voice recognition means, Speech recognition processing device for repeating the speech recognition using the user voice initially input through the user interface means. The method of claim 2, The user rating is, Speech recognition processing device, characterized in that input via the DTMF signal or the user's voice. The method of claim 5, The user, A voice recognition processing device, characterized in that the user of the telephone terminal connected through the telephone network. The method of claim 1, The user interface means, Voice recognition processing device for inducing the input of the user voice or the user evaluation through the announcement. In the speech recognition processing method for improving the speech recognition performance, A first voice recognition step of receiving a user voice for voice recognition from the user and performing voice recognition; A user evaluation step of receiving a user evaluation by providing a voice recognition result of the user voice to the user; And A second speech recognition step of redefining the pronunciation dictionary by limiting the range of the pronunciation dictionary according to the user evaluation and then re-performing the speech recognition process for the user's voice; Speech recognition processing method comprising a. The method of claim 8, The user rating is, Speech recognition processing method characterized in that it indicates the number of syllables normally recognized in the speech recognition results. The method of claim 9, The second voice recognition step, The voice recognition processing method of redefining the scope of the phonetic dictionary to words including letters up to the number of syllables indicated by the user evaluation is excluded, excluding the voice recognition result that is the target of the user evaluation. The method of claim 8, The second voice recognition step, And re-performing the voice recognition process and feedback to the user evaluation step. The method of claim 11, The second voice recognition step, The voice recognition processing method is repeatedly performed until the voice recognition is successful within a preset number of times. The method of claim 9, The user rating is, Speech recognition processing method characterized in that it is input through a DTMF signal or the user's voice.

Images