JP2003202894A - Device, method, and program for speech recognition - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To eliminate inconvenience in system function use due to a wrong reserved word as to a system equipped with a speech recognition function. <P>SOLUTION: The speech recognition device which recognizes a speech from input voice data by referring to a system dictionary wherein reserved words are previously registered by a system is equipped with a reserved word list consisting of a set of reserved words and a setting means of setting whether at least one reserved word in the reserved word list is determined as an object of speech recognition according to a user's indication. <P>COPYRIGHT: (C)2003,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice recognition device for performing voice recognition using a system dictionary in which recognition target vocabulary is registered in advance.

[0002]

2. Description of the Related Art Currently, in various information devices such as a car navigation system, those having a voice recognition function in addition to a normal function are increasing. The voice recognition function in the car navigation system can give various command operations such as destination designation and display switching by voice from a speaker such as a driver to the system instead of finger operation on the operation panel. It provides a convenient operation interface for the user.

Regarding the processing related to voice recognition, the system has a unique vocabulary to be recognized as information for identifying each of the above-mentioned various commands based on voice recognition. The recognition target vocabulary consists of a collection of words (hereinafter referred to as “reserved words”) registered in the system in advance.

No matter how many times the reserved words are input by voice,
There are words that are apt to be erroneously recognized as another word, and conversely, words that are erroneously recognized as a specific reserved word regardless of which word is uttered. Reserved words that cause such inconvenience differ depending on the already registered vocabulary and the number of vocabularies, and also differ depending on the speaker. Therefore, identify such words in advance and exclude them from the reserved words. The problem is that it is difficult.

In addition to reserved words, some systems are configured so that the user can register desired words (hereinafter, such words are referred to as "user words"). Words other than or other expressions of reserved words can be registered as user words. Therefore, when a certain reserved word is easily mistaken for another word, another easily recognizable user word can be registered so that the word can be smoothly recognized by voice. However, for a word that is likely to be erroneously recognized as a specific reserved word even if any other word is uttered, the misrecognition cannot be improved unless the reserved word is removed.

[0006]

SUMMARY OF THE INVENTION The present invention has been made in view of the above circumstances, and it is possible to set a specific reserved word out of the target of voice recognition among reserved words in the system, which is inconvenient. An object of the present invention is to provide a voice recognition device, method, and program capable of eliminating inconvenience in using system functions due to reserved words.

[0007]

In order to solve the above problems and achieve the object, the present invention is constructed as follows.

A first voice recognition apparatus according to the present invention is a voice recognition apparatus for voice-recognizing input voice data by referring to a system dictionary in which reserved words are registered in the system in advance, and is composed of a set of the reserved words. A reserved word list, a setting means for setting whether or not at least one reserved word in the reserved word list is a target of voice recognition according to an instruction from the user, and a setting content by the setting means. And a registration means for registering only the reserved words that are the target of the voice recognition in the system dictionary, excluding the reserved words that are not the target of the voice recognition in the reserved word list. It is a device.

A second voice recognition apparatus according to the present invention is a voice recognition apparatus which recognizes input voice data by voice referring to a system dictionary in which reservation words are registered in the system in advance, and is composed of a set of the reserved words. A reserved word list, a setting means for setting whether or not at least one reserved word in the reserved word list is a target of voice recognition according to an instruction from the user, and a setting content by the setting means. And a dictionary reconfiguring unit that reconfigures the system dictionary so that the reserved words that are not the target of the voice recognition are excluded from the reserved word list and only the reserved words that are the target of the voice recognition are configured. It is a voice recognition device characterized by doing.

A third voice recognition apparatus according to the present invention comprises voice recognition means for voice-recognizing input voice data by referring to a system dictionary in which reserved words are registered in the system in advance, and a set of the reserved words. A reserved word list, a setting means for setting whether or not at least one reserved word in the reserved word list is a target of voice recognition according to an instruction from the user, and a setting content by the setting means. Then, from among the plurality of candidates obtained as the voice recognition result by the voice recognition means, except the candidates corresponding to the reserved words that are not the target of the voice recognition in the reserved word list, the reservation that is the target of the voice recognition. And a selection unit that selects a candidate corresponding to only a word.

A first voice recognition method according to the present invention is a voice recognition method for voice-recognizing input voice data by referring to a system dictionary in which reserved words are registered in advance in the system, in response to an instruction from a user. And a setting step for setting whether or not at least one reserved word in the reserved word list composed of the set of reserved words is a target of voice recognition, and the reservation based on the setting content in the setting step. A registration step of registering only the reserved words, which are the target of the voice recognition, in the system dictionary, excluding the reserved words, which are not the targets of the voice recognition in the word list, is a voice recognition method.

A second voice recognition method according to the present invention is a voice recognition apparatus for voice-recognizing input voice data by referring to a system dictionary in which reserved words are registered in advance in the system, in accordance with an instruction from a user. The reserved word list is set based on the setting step of setting whether or not at least one reserved word in the reserved word list composed of the set of reserved words is a target of voice recognition, and the setting content in the setting step. A dictionary reconstructing step of reconstructing the system dictionary so that reserved words that are not the target of the voice recognition are removed, and the system dictionary is composed only of the reserved words that are the target of the voice recognition. This is a voice recognition method.

A third voice recognition method according to the present invention is a voice recognition step of voice-recognizing input voice data by referring to a system dictionary in which reserved words are registered in advance in a system, and a voice recognition step according to an instruction from a user. A setting step for setting whether or not at least one reserved word in the reserved word list composed of the set of reserved words is to be subjected to voice recognition, and the voice based on the setting content in the setting step. From the plurality of candidates obtained as the voice recognition result in the recognition step, except the candidate corresponding to the reserved word that is not the target of the voice recognition in the reserved word list, corresponding to only the reserved word that is the target of the voice recognition And a selection step of selecting a candidate to perform the voice recognition method.

The first voice recognition program according to the present invention is a voice recognition program for voice-recognizing input voice data by referring to a system dictionary in which reserved words are registered in advance in the system. In response to an instruction, a setting step of setting whether or not at least one reserved word in the reserved word list composed of the set of reserved words is a target of voice recognition, and based on the setting content in the setting step A registration step of registering only the reserved words that are the target of the voice recognition in the system dictionary, excluding the reserved words that are not the target of the voice recognition in the reserved word list, is a voice recognition program.

A second voice recognition program according to the present invention is a voice recognition program for voice-recognizing input voice data by referring to a system dictionary in which reserved words are registered in advance in the system. Accordingly, a setting step of setting whether or not at least one reserved word in the reserved word list composed of the set of reserved words is a target of voice recognition, and based on the setting content in the setting step, A dictionary reconstructing step of reconfiguring the system dictionary so that reserved words that are not the target of voice recognition are excluded from the reserved word list and only the reserved words that are the target of voice recognition are executed. It is a program.

A third voice recognition program according to the present invention provides a computer with a voice recognition step of voice-recognizing input voice data by referring to a system dictionary in which reserved words are registered in the system in advance, and an instruction from a user. Accordingly, a setting step of setting whether or not at least one reserved word in the reserved word list composed of the set of reserved words is a target of voice recognition, and based on the setting content in the setting step, From the plurality of candidates obtained as the voice recognition result in the voice recognition step, except for the candidate corresponding to the reserved word that is not the target of the voice recognition in the reserved word list, only the reserved word that is the target of the voice recognition. A speech recognition program for executing a selection step of selecting a corresponding candidate.

[0017]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to the drawings.

(First Embodiment) FIG. 1 is a block diagram showing the schematic arrangement of a voice recognition system according to the first embodiment of the present invention. The voice recognition system of this embodiment is incorporated in various information devices such as a car navigation system, and is connected to a system bus (not shown).
It is configured with a computer including storage devices such as U, RAM, and ROM. In this case, the present invention can be implemented by a computer program, and the program stored in a secondary storage device such as a hard disk or ROM or the like is read into a main memory such as RAM or the like and executed by the CPU. The computer includes a built-in type computer in the above information equipment and a general-purpose computer (PC or the like). For simplification of description, FIG. 1 shows only components related to the present invention, and other components are omitted. As shown in the figure, this speech recognition system is provided with an overall control unit 1
0, a voice recognition unit 11, a reserved word list 12, a word setting unit 14, and a display unit 13.

The overall control unit 10 is a main component of the voice recognition system and includes a voice recognition unit 11. The overall control unit 10 includes a processing unit such as an input / output interface that controls data input used for voice recognition and data output of a voice recognition processing result, and a user interface that controls user operation on the system. The data used for voice recognition may be, for example, voice data from a speaker input through the microphone 2 shown in FIG. 1 or line input of a voice data file or recorded voice. Further, the data output of the voice recognition processing result is passed to a higher-level function of the system such as a voice command input function. The user interface includes a graphical interface (GUI) that controls the display unit 13 including a touch panel and the like.

The interface processing section and the voice recognition section 11 can be realized by using a known technique. Phoneme text (voiced strings) from voice data
The speech recognition unit 11 that recognizes and outputs the input speech data is acoustically analyzed, and the degree of similarity to a small unit such as a phoneme is calculated based on the parameter (feature amount) obtained by this analysis. By arranging those phonemes optimally, candidates as words or sentences are output. In the voice recognition process, the reserved word registered in the voice recognition unit 11 is referred to as a recognition target. The present embodiment is configured so that the word setting unit 14 sets and changes the reserved word that is the target of the voice recognition.

FIG. 2 is a diagram for explaining the word setting section 14. The word setting unit 14 is the registration processing unit 1.
40 and setting flag 141. The reserved word list 12 is provided from the initial state, for example, by a ROM or the like in the system, and represents a plurality of words to be voice-recognized by the system (synonymous with “vocabulary”).
Is. The registration processing unit 140 determines whether or not to register the item in the voice recognition unit 11 according to the setting content of the setting flag 141 associated with each item (item) in the reserved word list 12. In the default state, the setting contents of the setting flag 141 are all set to "1", which means that all words are targets to be recognized by the voice recognition unit 11. Then, when the content of the setting flag 141 corresponding to a specific item in the reserved word list 12 is set to “0”, the word indicated by the item is not recognized by the voice recognition unit 11.

Needless to say, the truth of the flag may be reversed ("0" is a recognition target and "1" is a non-recognition target).

The setting to the setting flag 141 can be performed by the user via the user interface of the overall control unit 10. A screen configuration for this will be described with reference to FIG.

FIG. 3 shows an example of the structure of a screen displaying a list of reserved words.

On the display screen of the display unit 13, for example, 50
Reserved words are displayed in a list in an appropriate order such as the order of sounds or alphabets. This reserved word corresponds to an item in the reserved word list 12. A non-recognition check box 21 is provided next to each reserved word box 22, and a check is given when the user presses this with a finger or the like. When the check is added, this reserved word is not subject to voice recognition. Here, when the check box 21 is pressed again, the set check is removed, and this reserved word can be registered again as a recognition target word.

Check box 2 on such a display screen
The setting content of 1 is the setting flag 141 in the word setting unit 14.
Reflected in.

In FIG. 3, the first five words of all the reserved words are displayed. Among them, "Iritariori (Italian food)" and "Super (supermarket)" are checked, and these two words are not recognized. In this example, the word column 22 is set to "Hiragana display", but the present invention is not limited to this, and may be a mixture of Chinese characters and alphanumeric characters depending on the system functional configuration. In the case of a foreign language voice recognition system, appropriate characters such as alphabets are used according to each language.

When all the words are many and cannot be displayed on one screen, when the next button 24 is pressed, reserved words on and after the next screen are sequentially displayed. Further, when the previous button 23 is pressed, the reserved word on the previous screen is displayed. Then, when the end button 25 is pressed, the reserved word setting process on the list display screen ends.

In the system, it is preferable that the user does not carelessly change the above-mentioned setting for a specific reserved word that is essential as a voice recognition target. In order to distinguish such a reserved word whose setting cannot be changed from other reserved words, the setting flag 1
The value of 41 is set to “2” in advance, for example. Word setting section 1
4 corresponds to the set value of the setting flag 141 as described above.
With respect to the reserved word, the user operation on the non-recognition check box 21 is prohibited. Alternatively, such reserved words whose settings cannot be changed may be excluded from the list display in FIG.

In addition to displaying the list of reserved words as shown in FIG. 3 according to the instruction from the user, the reserved words in the reserved word list 12 that are not recognized are extracted and displayed. If the display switching function is provided in the display unit 13, it becomes easier to search for a desired word when the word that is not the recognition target is returned to the recognition target again, and the convenience is improved.

FIG. 4 is a flowchart showing a schematic operation of the voice recognition system according to this embodiment.

First, after initializing the hardware and activating the system when the power is turned on, a transition is made to a waiting state for a user operation or the like. In this waiting state, user's instructions other than word setting and voice recognition (system termination, etc.) are accepted, but in FIG. 4, only steps directly related to the present invention are shown, and other steps are omitted. is there.

When the request from the user to change the setting of the reserved vocabulary is accepted in step S31, the process proceeds to step S32, the reserved word list shown in FIG. 3 is displayed, and the above-mentioned voice recognition target is present. A word setting process is performed.

According to the setting made in step S32, the recognition target word is registered in the voice recognition unit 11, and the registered vocabulary becomes the system dictionary (step S33).
After that, when voice is input in step S34,
The voice recognition process is executed (step S35), and the process according to the recognition result is executed (step S36).

According to the present embodiment described above, only the reserved words to be recognized can be registered in the voice recognition unit 11 as the vocabulary to be recognized. That is, the user can explicitly designate a reserved word that is not a recognition target and exclude it from the system dictionary. Therefore, it becomes possible to eliminate the inconvenience in using the system function due to an inconvenient reserved word that is erroneously recognized as a similar word in the voice recognition process. The usability can be improved.

(Second Embodiment) Next, a second embodiment of the present invention will be described.

Depending on the type of the voice recognition system, FIG.
Instead of directly registering a word in the voice recognition unit 11 shown in, a system dictionary is first constructed by subjecting the recognition target word to processing such as compilation, and registered in the voice recognition unit 11 in the form of this system dictionary. There is something to do.
The second embodiment applies the present invention to such a voice recognition system.

FIG. 5 is a block diagram showing a schematic configuration of such a voice recognition system according to the second embodiment of the present invention. 5, components indicated by reference numerals 40 to 45 are the reference numerals 10 to 15 in FIG. 1, respectively.
It corresponds to the components shown in and has the same functions as these. Reference numeral 45 is a dictionary creating unit, and 46 is a system dictionary, which is different in structure from the first embodiment in that these components are added.

When the user sets the recognition target / non-target of the word by using the screen configuration of FIG. 3 described in the first embodiment and the word setting end button 25 is pressed, the dictionary creating unit 45 Of the reserved words in the reserved word list 12, only the reserved words registered as the recognition target words in the word setting unit 44 are used to reconstruct the system dictionary 46. Further, the word setting unit 44 replaces the reconstructed system dictionary 46 with the system dictionary that has been used so far.

Naturally, the reconstructed system dictionary 4
Since reserved words that are not recognized in 6 are not registered, they do not become recognition results. Therefore, also in the second embodiment, as in the first embodiment,
You can explicitly exclude inconvenient words that cause misrecognition.

When the continuous word recognition such as the place name recognition (eg, Kanagawa prefecture / Kawasaki city / Sachi ward) found in a car navigation system can be performed, only a part of the continuous words can be recognized. It may be set outside, which may cause inconvenience on the side of the voice recognition unit 41. In such a case, it is advisable to provide a word that cannot be set as a recognition target even in the reserved words. This can be dealt with by setting the value of the setting flag to a special value (“2” in the first embodiment), as described in the first embodiment.

Further, when a word in the middle is set as a non-recognition target, it is preferable to automatically set a word following the word as a non-recognition target because the setting operation is simplified.

(Third Embodiment) The third embodiment of the present invention will be described below.

In the first embodiment described above, the voice recognition unit 11 is used.
To prevent the registration of unrecognized reserved words in
In the second embodiment, when the system dictionary 46 is reconstructed, words that are not recognition targets are excluded. In these embodiments, the user reserves inconvenient reserved words in advance,
It is a work to remove it explicitly.

On the other hand, in the third embodiment of the present invention, the word exclusion operation is not performed in the process of re-registering the word or reconstructing the system dictionary, but recognizing the vocabulary which is not the recognition target. It is to prevent the result.

The present embodiment is similar to the first and second embodiments up to the point where the user sets the recognition target / non-recognition target, but only the setting is stored in the word setting section 44. However, unlike the other embodiments, the system dictionary is not changed and words are not re-registered in the voice recognition unit.

At the time of voice recognition, the voice recognition unit of the present embodiment performs the recognition process using the system dictionary from the beginning as it is. The voice recognition unit of the present embodiment is premised on being able to return a plurality of ranked recognition results for one voice input. In this case, since the system dictionary also includes words that have not been recognized, the recognition result may also include words that have not been recognized. A recognition result including a word that is not a recognition target will be referred to as a temporary recognition result.

The voice recognition unit refers to the recognition target / non-target flag of each word stored in the word setting unit, and deletes the reserved word that is not the recognition target from the temporary recognition result. And
The speech recognition unit outputs the remaining words ranked from the first rank again as a final recognition result, and the system performs processing according to the recognition result. If all the reserved words of the temporary recognition result are not recognized and no words remain, the user's utterance is treated as rejected (invalid or rejected). FIG. 6 is a flowchart showing a series of processes described above that exclude reserved words that have not been recognized from the recognition result.

When one recognition result is composed of a word string as in continuous word recognition, the recognition result containing at least one reserved word that is not a recognition target in the word string may be deleted. .

According to the third embodiment as described above, since the system dictionary is not changed, the utterance is rejected when all the recognition results are out of the recognition target, but the means for reconstructing the dictionary is unnecessary. As compared with the second embodiment, there are advantages that the system configuration is simpler and the memory and calculation amount can be reduced.

The present invention is not limited to the above-described embodiment, but can be variously modified and carried out.

[0052]

As described above, according to the present invention,
Provided are a voice recognition device, method, and program capable of setting a specific reserved word out of the target of voice recognition among reserved words in the system and eliminating inconvenience in using system functions due to an inconvenient reserved word. can do.

[Brief description of drawings]

FIG. 1 is a block diagram showing a schematic configuration of a voice recognition system according to a first embodiment of the present invention.

FIG. 2 is a block diagram showing an internal configuration of a word setting unit.

FIG. 3 is a diagram showing a configuration example of a screen that displays a list of reserved words.

FIG. 4 is a flowchart showing a schematic operation of the voice recognition system of the first embodiment.

FIG. 5 is a block diagram showing a schematic configuration of a voice recognition system according to a second embodiment of the present invention.

FIG. 6 is a flowchart showing a flow of processing for removing a reserved word that has not been recognized from the recognition result in the speech recognition system according to the third embodiment of the present invention.

[Explanation of symbols]

10 ... Overall control unit 11 ... Voice recognition unit 12 ... Reserved vocabulary list 13 ... Display 14 ... Word setting section

─────────────────────────────────────────────────── ─── Continuation of front page (51) Int.Cl. ⁷ Identification code FI theme code (reference) G10L 15/28

Claims (13)

[Claims] 1. A voice recognition device for recognizing input voice data by referring to a system dictionary in which reserved words are registered in advance in a system, a reserved word list including a set of the reserved words, and an instruction from a user. According to the setting means for setting whether or not at least one reserved word in the reserved word list is the target of voice recognition, and based on the setting content by the setting means, the voice in the reserved word list A speech recognition apparatus comprising: a registration unit that registers only reserved words that are the target of voice recognition in the system dictionary, excluding reserved words that are not recognition targets. 2. A voice recognition device for recognizing input voice data by referring to a system dictionary in which reserved words are registered in advance in a system, a reserved word list including a set of the reserved words, and an instruction from a user. According to the setting means for setting whether or not at least one reserved word in the reserved word list is the target of voice recognition, and based on the setting content by the setting means, the voice in the reserved word list A voice recognition device, comprising: a dictionary reconfiguring unit configured to reconfigure the system dictionary so that reserved words that are not recognition targets are removed and the reserved words that are the target of voice recognition are configured only. 3. A voice recognition means for recognizing input voice data by referring to a system dictionary in which reserved words are registered in advance in the system, a reserved word list composed of a set of the reserved words, and an instruction from a user. Depending on the setting means for setting whether or not at least one reserved word in the reserved word list is the target of the voice recognition, based on the setting content by the setting means, the voice recognition result by the voice recognition means From among the plurality of candidates obtained as above, a candidate corresponding to only the reserved word that is the target of the voice recognition is selected in the reserved word list, except a candidate that corresponds to the reserved word that is not the target of the voice recognition. A speech recognition apparatus comprising: 4. The method according to claim 1, further comprising a prohibition unit that prohibits the user from setting a specific reserved word not to be subject to the voice recognition. The voice recognition device according to the item. 5. The setting means comprises, for each reserved word in the reserved word list, at least a binary flag indicating whether or not the reserved word is a target of the voice recognition. The voice recognition device according to claim 1. 6. A voice recognition method for recognizing input voice data by referring to a system dictionary in which reserved words are registered in advance in a system, comprising a reservation composed of a set of the reserved words according to an instruction from a user. A setting step of setting whether or not at least one reserved word in the word list is to be subjected to voice recognition, and based on the setting content in the setting step, a reserved word that is not the target of the voice recognition in the reserved word list. And a registration step of registering only the reserved words that are the target of the voice recognition in the system dictionary. 7. A voice recognition device for voice recognition of input voice data by referring to a system dictionary in which reserved words are registered in advance in a system, wherein a reservation comprising a set of the reserved words according to an instruction from a user. A setting step of setting whether or not at least one reserved word in the word list is to be subjected to voice recognition, and based on the setting content in the setting step, a reserved word that is not the target of the voice recognition in the reserved word list. And a dictionary reconstructing step of reconstructing the system dictionary so that the system dictionary is composed only of reserved words that are the target of the voice recognition. 8. A voice recognition step of recognizing input voice data by referring to a system dictionary in which reserved words are registered in advance in the system, and a reservation composed of a set of the reserved words according to an instruction from a user. A setting step for setting whether or not at least one reserved word in the word list is to be subjected to voice recognition, and based on the setting contents in the setting step, a plurality of voice recognition results obtained in the voice recognition step A selection step of selecting, from the candidates, candidates corresponding to the reserved words that are not the target of the voice recognition in the reserved word list, and selecting only candidates corresponding to the reserved words that are the target of the voice recognition. A speech recognition method characterized by the above. 9. The method according to claim 6, further comprising a prohibition step of prohibiting the user from setting a specific reserved word not to be subject to the voice recognition. Speech recognition method described in paragraph. 10. A voice recognition program for recognizing input voice data by referring to a system dictionary in which reserved words are registered in advance in a system, comprising a set of the reserved words to a computer according to an instruction from a user. A setting step of setting whether or not at least one reserved word in the reserved word list to be a target of voice recognition, and a target of the voice recognition in the reserved word list based on the setting content in the setting step. A registration step of registering only the reserved words that are the target of the voice recognition in the system dictionary excluding the reserved words that are not registered in the system recognition program. 11. A voice recognition program for recognizing input voice data by referring to a system dictionary in which reserved words are registered in advance in the system, comprising a set of the reserved words according to an instruction from a user to a computer. A setting step of setting whether or not at least one reserved word in the reserved word list to be a target of voice recognition, and a target of the voice recognition in the reserved word list based on the setting content in the setting step. Non-reserved words are excluded, and a dictionary reconstructing step of reconstructing the system dictionary so that the system dictionary is composed only of reserved words that are the target of the speech recognition. 12. A voice recognition step of recognizing input voice data into a computer by referring to a system dictionary in which reserved words are registered in a system in advance, and a set of the reserved words according to an instruction from a user. Based on the setting step of setting whether or not at least one reserved word in the reserved word list to be the target of the voice recognition, and the content of the setting in the setting step, it is obtained as the voice recognition result in the voice recognition step. From the plurality of candidates, a selection step of selecting candidates corresponding to only the reserved words that are the target of the voice recognition, excluding candidates corresponding to the reserved words that are not the target of the voice recognition in the reserved word list, A voice recognition program that executes. 13. The method according to claim 10, further comprising a prohibition step of prohibiting the user from setting a specific reserved word not to be subject to the voice recognition. A speech recognition program according to item.