JP2000181487A - Voice recognition device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve the recognition rate of 2nd and succeeding voice recognition when the voice of the same word or phrase is recognized again. SOLUTION: This device is provided with a voice input part 17 for inputting the voice of a speaker, a 1st voice recognition resource 31 which stores the word or phrase to be recognized in advance, a voice recognition part 32 which recognizes the word or phrase from the voice inputted from the voice input part 17, extracts the recognized word or phrase from the 1st voice recognition resource 31, and outputs it, and a voice recognition resource generation part 34 which narrows down words and phrases stored in the 1st voice recognition resource 31 according to the extracted word or phrase to generate a 2nd voice recognition resource 35. Here, when a word or phrase inputted again from the voice input part 17 is recognized, a word or phrase is extracted from the 2nd voice recognition resource 35.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech recognition apparatus for recognizing an input speech and extracting a phrase corresponding to the recognition result from a previously stored phrase.

[0002]

2. Description of the Related Art As shown in FIG. 15, a conventional voice recognition apparatus has a voice input unit 1 having a microphone 1a for inputting voice and an A / D converter 1b for converting voice from the microphone into a digital signal. A speech recognition resource 2 composed of a word to be recognized and a recognition code defined for each word; a word is recognized based on an output from the speech input unit 1; The speech recognition unit 3 extracts a product based on the recognition code, and an application program unit 4 that registers a product based on the recognition code extracted by the speech recognition unit 3, calculates a sales amount of the product, and the like.

[0003] The speech recognition resource 2 is a meta-language used to formally define a general BN notation (Backus-Naur-form; computer language) as shown in FIG. .).

In FIG. 16, "<SENTENCE>
= <Product price> Yen <Product type> is <Sale quantity>. For example, when the sentence “3 plain 50 yen papers” is input by voice, “50” is the product price, “plain paper” is the product type, and “3” is sold. Quantity)
It means to define.

Also, "<commodity amount> = 1 to 9999
9. , Which means that the price of the merchandise is one of 1 to 99999. Also, "<product type> =
Photo ｜ Card ｜ Paper ｜ Rough paper… ｜ Bromide. "This means that the product types are photo, card, paper, rough paper, ...
It is one of bromides. further,
There is "<sales quantity> = 1 to 99999.", which means that the sales quantity is one of 1 to 99999. These phrases (including the phrase of the amount of money) are defined as phrases recognized in advance.

In such an apparatus, for example, when an operator (speaker) inputs a voice "3 sheets of plain paper of 50 yen" from the voice input unit 1, a predetermined phrase,
That is, "50" is recognized as the product price, "plain paper" as the product type, and "3" as the sales quantity, and a recognition code corresponding to each word is output to the application program unit 4.

However, the speech recognition resources in such a speech recognition apparatus store a huge number of words and phrases, and the number of combinations of items becomes enormous, and one recognition result is output from them. I often did. Therefore, conventionally, after the voice input, the same phrase is input in a conversation format in which a voice input for confirmation can be made again.

[0008] For example, when a customer receives an order for "please give plain paper", he or she inputs a voice saying "this is plain paper," and asks the customer "how long should it be?" If you answered "Please, three sheets", you would say "If you check, three sheets of plain 50 yen paper." As a result, the voice input for the same phrase “plain paper” is performed twice.

[0009]

However, in the conventional apparatus, even if the same word is input for the second and subsequent times, the voice recognition is performed again using the same voice recognition resource 4. At 1
There is a possibility that it will be misrecognized with a probability as large as the second voice input. In this case, there is a problem that the effect of inputting a plurality of voices for confirmation is not sufficiently exhibited.

In general, the greater the number of phrases defined in the speech recognition resource 4, the easier it is to misrecognize, and the more words with similar pronunciations, the easier it is to misrecognize them. It is considered that the recognition rate is improved by narrowing down the words that are likely to be recognized after the second time based on the result of the second voice recognition.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a speech recognition apparatus capable of improving the recognition rate of the second and subsequent speech recognitions when recognizing the same phrase again.

[0012]

According to a first aspect of the present invention, there is provided a voice input means for inputting a voice of a speaker, a voice recognition means for recognizing a phrase from a voice input from the voice input means, and A first speech recognition resource that stores a plurality of phrases to be recognized, a phrase extraction unit that extracts the phrase recognized by the speech recognition unit from the first speech recognition resource,
A speech recognition resource generation unit for narrowing down the words stored in the first speech recognition resource based on the words extracted by the word extraction unit to generate a second speech recognition resource; When recognizing an inputted phrase, the speech recognition apparatus is characterized by extracting a phrase from a second speech recognition resource.

According to a second aspect of the present invention, the voice recognition resource generating means generates a second voice recognition resource as a word to be recognized in advance by using the word extracted from the first voice recognition resource by the word extracting means. 2. The method according to claim 1, wherein
It is a speech recognition device of the description.

According to a third aspect of the present invention, the first speech recognition resource stores a plurality of phrases to be recognized in advance for each phrase group, and the speech recognition resource generating means includes: 2. The speech recognition apparatus according to claim 1, wherein a speech recognition resource of a phrase group including the phrase extracted from the first speech recognition resource by the phrase extraction unit is generated as a second speech recognition resource. It is.

According to a fourth aspect of the present invention, there is provided a voice input means for inputting a voice of a speaker, a voice recognition means for recognizing a phrase from a voice input from the voice input means, and a plurality of voice recognition means to be recognized in advance. A first speech recognition resource for storing a phrase, a phrase extracting unit for extracting from the first speech recognition resource a phrase for which the phrase recognized by the speech recognition unit is to be recognized in advance, and a first speech for the phrase extracting unit A phrase extracted from the recognition resource is set as a candidate phrase A, and a phrase stored in the first speech recognition resource is set as a candidate phrase B.
And a voice recognition resource generating means for generating a voice recognition resource of the second type. When voice recognition of a word re-input from the voice input means is performed, the word from one of the candidate words A and B of the second voice recognition resource is used. Is a speech recognition device characterized by extracting

According to a fifth aspect of the present invention, there is provided a voice input means for inputting a voice of a speaker, a voice recognition means for recognizing a phrase from a voice input from the voice input means, and a plurality of voice recognition means to be recognized in advance. A first speech recognition resource in which words are stored for each word group, and
And a phrase extracted from the first speech recognition resource by the phrase extraction unit as a candidate phrase A, and a phrase extraction unit of a phrase group of the first speech recognition resource. And a voice recognition resource generating means for generating a second voice recognition resource that uses the word group including the word extracted in step 2 as a candidate word B.
A phrase is extracted from any one of the candidate phrases A and B of the speech recognition resource.

According to a sixth aspect of the present invention, there is provided a voice input means for inputting a voice of a speaker, a voice recognition means for recognizing a phrase from the voice input from the voice input means, A first speech recognition resource that stores a phrase of a price as a phrase, a phrase extraction unit that extracts a phrase recognized by the speech recognition unit from the first speech recognition resource,
Voice recognition resource generating means for generating, as a second voice recognition resource, what is obtained by deleting a phrase having a price lower than the sales amount of the product from the price term of the first voice recognition resource; A speech recognition apparatus characterized in that when speech is input, a price phrase is extracted from a second speech recognition resource.

[0018]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, a first embodiment of the present invention will be described with reference to FIGS. FIG. 1 is a block diagram showing the configuration of a speech recognition apparatus according to the present embodiment. Reference numeral 11 denotes a CPU (central processing unit) constituting a control unit main body, and 12 denotes a ROM (ROM) storing program data to be executed by the CPU 11. Read-only memory), 13
Is a RAM (random access memory) provided with a memory used for various data processing, 14 is a hard disk drive (HDD), and 15 is a key provided for a worker (speaker) to perform various operations. The keyboard control unit 17 connected to the keyboard 16 includes a microphone 18 for inputting audio as an analog signal and an A / D converter 19 for converting audio input from the microphone 18 as an analog signal to a digital signal. A voice input unit 20 as a voice input unit is a display control unit that controls a display unit 21 that displays a result of recognition of the input voice.

The CPU 11, ROM 12, RAM 1
3, hard disk device 14, keyboard control unit 15,
The A / D converter 19 and the display control unit 20 are connected to each other by bus lines such as a data bus, a control bus, and an address bus.

FIG. 2 is a functional block diagram showing the configuration of the speech recognition apparatus according to the present embodiment. Reference numeral 31 denotes a word to be recognized in advance and a recognition code defined (associated) for each word. A first speech recognition resource consisting of
32 recognizes a phrase corresponding to the input voice based on the output from the voice input unit 17 (voice recognition means), and outputs a recognition code corresponding to the phrase to the first voice recognition resource 31.
Alternatively, the speech recognition unit 33 extracts and outputs from a second speech recognition resource 35 to be described later and outputs a recognition text obtained from the recognition result (phrase extraction means), based on the recognition code from the speech recognition unit 32. The application program unit that displays the corresponding product and passes the recognition text from the speech recognition unit 32 to a speech recognition resource generation unit 34, which will be described later. The second speech recognition resource 35 is narrowed down by recognizing the recognition resource 31.
This is a speech recognition resource generation unit as a speech recognition resource generation unit that generates.

Although not shown, the words to be speech-recognized are stored in advance in association with the speech characteristic data of the standard speaker (unspecified speaker-compatible type). However, the voice feature data actually uttered by the user may be associated (specific speaker correspondence type).

The first speech recognition resource 31 stores the original data used for performing the first speech recognition, and is constructed in advance in, for example, the hard disk device 14 or the like. Specifically, words to be recognized by speech are defined. Among them, FIG. 3 shows an example in which the product type is described in general BN notation.

In FIG. 3, "<SENTENCE> =
<Product type>? Is it? . However, this means that, for example, when the sentence "Is plain paper, photo and card" is spoken, "plain paper", "photo", and "card" are defined as product types. . In addition, "<Product type>
= Photo | card | paper | rough paper ... | bromide. , But the product types are photo, card, paper, rough paper,
... means that it is one of bromide.

The second speech recognition resource 35 is used when performing speech recognition for the second time or later, and is constructed by the speech recognition resource generation unit 34 in, for example, the hard disk device 14 or the like. Specifically, terms to be recognized by speech (including terms of money) are defined. For the second speech recognition resource 35, a general BN
An example described in the notation is shown in FIG.

In FIG. 4, "<SENTENCE> =
Do you check? <Product price> Yen <Product type> as <Sale quantity> sheets? is not it? . , But this is, for example, "5
If the sentence "3 sheets of plain paper of 0 yen" is input by voice, "50" is the product price, "plain paper" is the product type,
This means that “3” is defined as the number of units sold. Also,
There is "<commodity amount> = 1 to 99999.", which means that the commodity amount is any of 1 to 99999.

"<Product type> = plain paper | thermal paper |
bromide. This means that the product type is either plain paper, thermal paper, or bromide. That is, among the original product types to be recognized first as shown in FIG. 3, the products to be recognized for the second and subsequent times are narrowed down to plain paper, thermal paper, and bromide.

Also, "<sales quantity> = 1 to 9999
9. , Which means that the sales quantity is any one of 1 to 99999.

The speech recognition unit 32, the application program unit 33, and the speech recognition resource generation unit 3
4 is, for example, a hard disk drive 14, RO
It is configured by a software program stored in M12 or the like and readable by the CPU 11.

The speech recognition resource generation unit 34 converts the recognition text (the character string such as the product name output from the speech recognition unit 32 by the first speech recognition) passed from the application program unit 33 into the second and subsequent speeches. The second speech recognition resource 35 as a phrase to be recognized
Generate

In the embodiment of the present invention having such a configuration, voice input is performed in the following conversational form (however, the microphone 18 is installed on the operator side, and the voice of the customer is Will not be entered).

For example, in response to an order from a customer, "Well, please give me plain paper, thermal paper and bromide." ? ", The speech recognition unit 32 performs the first speech recognition for the product type based on the first speech recognition resource 31 shown in FIG.

Specifically, "plain paper", "thermal paper",
"Bromide" is recognized and these "plain paper"
The recognition codes corresponding to “thermal paper” and “bromide” and the recognition texts “plain paper”, “thermal paper” and “bromide” are passed to the application program unit 33. Then, the application program unit 33 displays “plain paper”, “thermal paper”, and “bromide” as product types on the display unit 21 based on the recognition code, and passes the recognition text to the speech recognition resource generation unit 34. It is.

Subsequently, the speech recognition resource generation unit 34 narrows down the product types to be recognized in the second and subsequent speech recognitions to the above-mentioned recognized texts, "plain paper", "thermal paper", and "bromide". Then, a second speech recognition resource 35 as shown in FIG. 4 is generated.

In response to the operator (speaker) confirming "is plain paper, thermal paper and bromide?", The customer answers "yes, yes." Asks, "How much should I do?" On the other hand, if the customer answers, "Um, 3 sheets of plain paper of 50 yen, 4 sheets of thermal paper of 60 yen and 10 sheets of bromide of 80 yen," the operator (speaker) "If you check, 5
3 sheets of plain paper of 0 yen, 4 sheets of thermal paper of 60 yen and 10 sheets of bromide of 80 yen. "To the microphone 18. Then, the voice recognition unit 32 outputs the second
The second speech recognition for the product type is performed based on the speech recognition resource 35 of FIG.

In this case, the second speech recognition resource 3
Since the product types of No. 5 are only “plain paper”, “thermal paper”, and “bromide”, the recognition rate of the product type is high because voice recognition is performed from among them. afterwards,
Based on the recognition code of the product amount, the product type, and the sales quantity recognized by voice, the application program unit 33 performs business processes such as registration of the product and calculation of the sales amount of the product.

As described above, the first speech recognition resource 31 having a large amount of data is used when performing speech recognition for the same phrase for the first time, and the first speech recognition is performed when speech recognition is performed for the second and subsequent times. A second speech recognition resource 35 generated by narrowing down the words of the recognition text obtained as a result.
Is used, the range of words to be speech-recognized is limited, so that the recognition rate of the second and subsequent speech recognition can be improved. As a result, the effect in the case of performing the second or later voice input, that is, the voice input by confirmation can be sufficiently exhibited.

Further, since the second and subsequent speech recognition is performed by the second speech recognition resource 35 in which the words of the first speech recognition resource 31 are narrowed down, the processing of the speech recognition unit 32 required for the second and subsequent speech recognition is performed. The amount can be reduced and the processing time can be reduced.

In the present embodiment, the case where the recognized text from the voice recognition unit 32 is received via the application program unit 33 has been described. However, the present invention is not necessarily limited to this. As shown in FIG. May be directly received from the voice recognition unit 32.

Further, the first speech recognition resource 31 in the present embodiment is divided into a plurality of speech recognition resource groups as shown in FIG. 6, and from among these speech recognition resource groups, A speech recognition resource including the phrase of the recognized text obtained by the speech recognition unit 32 when speech recognition is first performed for the same phrase may be extracted and generated as the second speech recognition resource 35. Even in this case, the range of words to be subjected to voice recognition after the second time is limited, so that the recognition rate can be improved.

In this case, the first speech recognition resource 31
Is a speech recognition resource (1) as shown in FIG.
The speech recognition resources (2), the speech recognition resources (3)... Then, in the first speech recognition, the speech recognition resource generation unit 34
For example, when a voice text of "plain paper" is received from the user, the voice recognition resource (1) including the "plain paper" is set as the second voice recognition resource 35, and the second and subsequent voice recognition is performed by the second voice recognition. This is performed based on the resource 35. As described above, the voice recognition resource generation unit 34 only needs to select a voice recognition resource for performing the second and subsequent voice recognition from the voice recognition resource group of the first voice recognition resource,
Since it is not necessary to newly generate a speech recognition resource, the storage capacity for the resource can be eliminated.

Next, a second embodiment of the present invention will be described with reference to FIGS. Note that a block diagram and a functional block diagram illustrating the configuration of the speech recognition apparatus according to the present embodiment are the same as those illustrated in FIGS. 1 and 2, and thus detailed description thereof will be omitted.

This embodiment is different from the first embodiment in that the first speech recognition resource based on the phrase (recognized text) recognized when the first speech recognition is performed for the same phrase. Instead of generating the second speech recognition resource 35 in which the range of the words stored in 31 is limited, the second speech recognition in which the probability that the words recognized when the first speech recognition is performed is re-recognized is increased. Resource 3
5 is generated.

More specifically, the speech recognition resource generation unit 34 in the present embodiment sets the phrase recognized by performing the first speech recognition as a candidate phrase A and is stored in the first speech recognition resource 31. The term that was used as the candidate word B is 2
In the second and subsequent speech recognition, a second speech recognition resource 35 that can be recognized from either the candidate phrase A or the candidate phrase B is generated.

When this is described in general BN notation, the result is as shown in FIG. The second voice recognition resource 35 shown in FIG. 9 is generated by the voice recognition resource generation unit 34 when “plain paper” is recognized as the product type in the first voice recognition. The difference from the first speech recognition resource 31 shown in FIG.
<Product Type 1> | <Product Type 2> ”is added to define the product type as product type 1 (corresponding to candidate phrase A) or product type 2 (corresponding to candidate phrase B) and product type 1 Is defined as “plain paper”, and the product type 2 is defined as any of the phrases stored in the first speech recognition resource 31. Accordingly, when the speech recognition unit 32 performs speech recognition of the product type using the second speech recognition resource 35, the probability that “plain paper” is recognized increases. That is, since the product type is the product type 1 or the product type 2, the probability that the “type 1” “plain paper” is recognized is 50%, and the product type 2 includes “plain paper”. The overall probability that "plain paper" is recognized is at least 50% or more.

Also in the embodiment of the present invention having such a configuration, the voice input is performed in the following conversation format as in the first embodiment. For example, when an operator (speaker) utters “Plain paper?” To the microphone 18 in response to an order “Um, plain paper please” from a customer, the voice recognition unit 32 The first speech recognition for the product type is performed based on the speech recognition resources as shown in FIG. The speech recognition resources shown in FIG. 8 correspond to the first speech recognition resources 31 for terms of the price of the product, terms of the product type, and terms of the number of units sold.

Specifically, “plain paper” is recognized, and a recognition code corresponding to “plain paper” and a recognition text “plain paper” are passed to the application program unit 33. Then, “plain paper” is displayed as a product type on the display unit 21 based on the recognition code by the application program unit 33, and the recognition text is passed to the speech recognition resource generation unit 34. Thereby, the voice recognition resource generation unit 34 generates a new voice recognition resource as shown in FIG. 9 that increases the probability that “plain paper” is recognized for the product type. In addition,
The speech recognition resources shown in FIG. 9 correspond to the second speech recognition resources 35 for terms of the product type, but terms other than the product type, that is, terms of the product price and the number of units sold, are not narrowed down. , Still correspond to the first speech recognition resource 31.

When the customer (the speaker) asks “how much is it?” And the customer answers “50 yen”, the operator (speaker) turns the microphone 18 When the user says "50 yen?"
As a result, the first speech recognition for the price of the commodity is performed based on the speech recognition resources shown in FIG.

Specifically, “50” is recognized, and “5” is recognized.
The recognition code corresponding to “0” and the recognition text “50” are passed to the application program unit 33.
Then, based on the recognition code, “50” is displayed on the display unit 21 as the product price of 50 yen by the application program unit 33, and the recognition text is passed to the speech recognition resource generation unit 34. As a result, the speech recognition resource generation unit 34 generates a new speech recognition resource as shown in FIG. 10 that increases the probability that “50” is recognized for the product price.

That is, as for the product price, “<product price> = <product price 1> | <product price 2>” is added and the product price is changed to product price 1 (corresponding to candidate phrase A) or product price 2
(Corresponding to candidate phrase B), product price 1 is defined as "50", and product price 2 is defined as any one of the phrases (1
９９99999). As a result, when the speech recognition unit 32 performs speech recognition of the commodity price using the speech recognition resources shown in FIG. 10, the probability that "50" is recognized increases. That is, since the product price is the product price 1 or the product price 2, the product price 1 is “50”.
Is 50%, and the commodity price 2 includes "50", so that the overall probability of "50" being recognized is at least 50% or more.

Note that the speech recognition resources shown in FIG. 10 correspond to the second speech recognition resources 35 for terms of the product type and terms of the price of the product, but the terms of the number sold are not narrowed down. , Still correspond to the first speech recognition resource 31.

Next, in response to the question “How many cards?” By the operator (speaker), when the customer answers “Please give 4 cards”, the operator (speaker) receives the microphone 18. "4
Is it a piece? "By the voice recognition unit 32 in FIG.
Based on the speech recognition resource indicated by 0, the first speech recognition for the sales quantity is performed. Specifically, “4” is recognized, and a recognition code corresponding to “4” and a recognition text “4” are passed to the application program unit 33. Then, the application program unit 33 displays “4” on the display unit 21 as four sales units based on the recognition code, and the recognized text is passed to the speech recognition resource generation unit 34. As a result, the speech recognition resource generation unit 34 generates a new speech recognition resource as shown in FIG. 11 that increases the probability that “4” is recognized for the product price.

That is, as for the sales quantity, “<sales quantity> = <sales quantity 1> | <sales quantity 2>” is added, and the sales quantity is changed to sales quantity 1 (corresponding to candidate phrase A) or sales quantity 2
(Corresponding to candidate word B), the sales quantity 1 is defined as "4", and the sales quantity 2 is defined as any one of the words (1 to 1) stored in the first speech recognition resource 31.
99999). As a result, when the speech recognition unit 32 performs speech recognition of the number of sales using the speech recognition resources shown in FIG. 11, the probability that “4” is recognized is increased for the same reason as in the case of the commodity price.
Note that the speech recognition resources shown in FIG. 11 are the second words about the product type phrase, the product price phrase, and the sales quantity phrase.
Correspond to the voice recognition resource 35 of FIG.

Then, the operator (speaker) speaks toward the microphone 18 saying, "If you confirm, it is four sheets of plain paper of 50 yen." Then, the speech recognition unit 32 uses the words of the item price, the item type, and the number of units sold as 2 words.
The second speech recognition is performed.

In this case, the speech recognition is performed based on the speech recognition resources shown in FIG. 11 (corresponding to the second speech recognition resources 35 for each phrase), so that the recognition rate increases. After that, the product value, product type,
Based on the sales number recognition code, the application program unit 33 performs business processing such as registration of a product and calculation of the sales amount of the product.

As described above, the speech recognition resource generation unit 34
Generates a second speech recognition resource 35 with the phrase recognized by performing the first speech recognition as the candidate phrase A and the phrase stored in the first speech recognition resource 31 as the candidate phrase B, In the second and subsequent speech recognition, candidate phrase A
And the candidate phrase B, it is possible to increase the probability that the phrase recognized at the time of the first speech recognition will be recognized again. Thereby, the recognition rate of the second and subsequent speech recognition can be improved. Therefore, the effect in the case of performing the second or later voice input, that is, the voice input by confirmation can be sufficiently exhibited.

Next, a third embodiment of the present invention will be described with reference to FIG.
This will be described with reference to FIG. FIG. 1 is a block diagram showing the configuration of the speech recognition apparatus according to the present embodiment.
And the detailed description is omitted. FIG. 12 is a functional block diagram of the speech recognition apparatus according to the present embodiment. The difference from the speech recognition apparatus shown in FIG. The sales amount code indicating the sales amount calculated based on the sales amount code is received, and the term range of the amount stored in the first speech recognition resource 42 used for recognizing the “deposit amount” is determined based on the sales amount code. Limited second speech recognition resource 4
3 is generated. More specifically, a phrase obtained by deleting a phrase having an amount lower than the sales amount indicated by the sales amount code from the phrase of the sales amount of the first speech recognition resource 42 is referred to as a second expression.
Of the speech recognition resource 43.

The first speech recognition resource 42 in the present embodiment is described in a general BN notation as shown in FIG. 13, and based on this, the second speech recognition resource 43 is generated. . In FIG. 13, "<SENTEN
CE> = <deposit amount> JPY This means that, for example, when a sentence “1005 yen will be deposited” is input by voice, “1005” is defined as the deposit amount. Also, “<deposit amount> = 1 to 9999
9. , Which means that the deposit amount is one of 1 to 99999.

In the present embodiment having such a configuration, as described in detail in the first or second embodiment, the commodity recognition unit 32 uses the speech recognition unit 32 based on the speech input from the operator (speaker). Then, the number of products is voice-recognized and a recognition code is output. The sales amount is calculated by the application program unit 33 based on the recognition code, and the sales amount code is passed to the speech recognition resource generation unit 41.

Thus, the speech recognition resource generation unit 41
Generates the second speech recognition resource 43 by using the value of the sales amount after deleting the amount lower than the sales amount indicated by the sales amount code. For example, when the sales amount is 1005 yen, the second speech recognition resource 43 generated by the speech recognition resource generation unit 41 is, as shown in FIG.
It is described as “<deposit amount> = 1005 to 99999.”. That is, 1 to 1004 are deleted from the definition of <deposit amount> in the first voice recognition resource 42. The reason that the sales amount is 1005 yen is because it is impossible to recognize an amount of 1004 yen or less as a deposit amount from a customer.

As a result, it is possible to prevent erroneous recognition of an impossible deposit amount, and the recognition rate of the deposit amount is improved. For example, assuming that the deposit amounts of 1,000 yen (3 yen) and 3 yen (3 yen) have similar sounds and are easily misrecognized, but if the sales amount is 800 yen, for example,
Since an amount lower than 800 yen is recognized by the eliminated second voice recognition resource, a voice recognition of the deposit amount of 1000 yen is performed.
It is possible to prevent erroneous recognition of 3 yen.

Since the voice recognition of the deposit amount is performed by the second voice recognition resource 43 in which the terms of the amount of money of the first voice recognition resource 42 are narrowed down, the processing of the voice recognition unit 32 required for the voice recognition of the deposit amount is performed. The amount can be reduced,
Processing time can also be reduced.

The third embodiment of the present invention having such a configuration may be applied to a device for inputting only the deposit amount by voice, and may be combined with the first or second embodiment. May be applied.

[0063]

As described in detail above, according to the present invention, when performing speech recognition for the same phrase for the first time, the first speech recognition resource having a large data amount is used, and speech recognition is performed for the second and subsequent times. In the case of performing, since the range of the phrase to be recognized is limited by using the second speech recognition resource narrowed and generated based on the phrase obtained as a result of the first speech recognition, 2 It is possible to improve the recognition rate of the voice recognition after the first time. Further, the processing amount of the voice recognition means required for the second and subsequent voice recognition can be reduced, and the processing time can be reduced.

The phrase recognized by the first speech recognition by the speech recognition resource generating means is set as a candidate phrase A, and the phrase stored in the first speech recognition resource or the first speech including the phrase is also considered. Generating a second speech recognition resource with the word group of the recognition resource as a candidate word B;
In the second and subsequent speech recognitions, by recognizing either of the candidate words A and B, it is possible to increase the probability that the words recognized when the first speech recognition is performed will be recognized again. . Thereby, the recognition rate of the second and subsequent speech recognition can be improved.

When a second speech recognition resource is generated by deleting a phrase having a price lower than the sales amount from the phrase of the first speech recognition resource, a second speech recognition resource is used. By extracting the phrase of the amount of money from the second speech recognition resource, it is possible to prevent the impossible amount of money from being mistakenly recognized, and the recognition rate of the amount of money to be deposited is improved. Further, since the voice recognition of the deposit amount is performed by the second voice recognition resource in which the terms of the amount of money of the first voice recognition resource are narrowed down, the processing amount of the voice recognition means required for the voice recognition of the deposit amount can be reduced. Processing time can be reduced.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a speech recognition device according to a first embodiment of the present invention.

FIG. 2 is a functional block diagram according to the embodiment.

FIG. 3 is a diagram showing an example of a first speech recognition resource according to the embodiment.

FIG. 4 is a diagram showing an example of a second speech recognition resource according to the embodiment.

FIG. 5 is a functional block diagram showing another example of the speech recognition device according to the embodiment;

FIG. 6 is a functional block diagram showing another example of the speech recognition device according to the embodiment.

FIG. 7 is a diagram showing an example of a first speech recognition resource in the speech recognition device shown in FIG. 6;

FIG. 8 is a diagram showing an example of a first speech recognition resource according to the second embodiment of the present invention.

FIG. 9 shows a second example of the product type in the present embodiment.
The figure which shows the example of the speech recognition resource of FIG.

FIG. 10 is a diagram showing an example of a second speech recognition resource for a phrase of a product type and a phrase of a product price in the embodiment.

FIG. 11 is a diagram showing an example of a second speech recognition resource for a phrase of a product type, a phrase of a product price, and a phrase of a sales quantity in the embodiment.

FIG. 12 is a functional block diagram according to a third embodiment of the present invention.

FIG. 13 is a diagram showing an example of a first speech recognition resource according to the embodiment.

FIG. 14 is a diagram showing an example of a second speech recognition resource according to the embodiment.

FIG. 15 is a functional block diagram of a conventional voice recognition device.

FIG. 16 is a diagram showing an example of a speech recognition resource in a conventional speech recognition device.

[Explanation of symbols]

 11 CPU 17 Voice input units 31 and 42 First voice recognition resource 32 Voice recognition unit 33 Application program unit 34 Voice recognition resource generation unit 35 and 43 Second voice recognition resource 41 Voice recognition resource Generator

Claims (6)

[Claims] 1. A voice input unit for inputting a voice of a speaker, a voice recognition unit for recognizing a phrase from a voice input from the voice input unit, and a first storing a plurality of phrases to be recognized in advance. A speech recognition resource, a phrase extraction unit that extracts a phrase recognized by the speech recognition unit from the first speech recognition resource, and the first speech recognition resource based on the phrase extracted by the phrase extraction unit. And a speech recognition resource generating means for generating a second speech recognition resource by narrowing down the words stored in the second speech recognition resource. A speech recognition device for extracting words and phrases. 2. The speech recognition resource generation unit generates a second speech recognition resource as a phrase to be recognized in advance, using a phrase extracted from the first speech recognition resource by the phrase extraction unit. The speech recognition device according to claim 1, wherein 3. The first speech recognition resource stores a plurality of phrases to be recognized in advance for each phrase group, and the speech recognition resource generation unit includes: 2. A speech recognition apparatus according to claim 1, wherein said phrase extraction means generates a speech recognition resource of a phrase group including the phrase extracted from said first speech recognition resource as a second speech recognition resource. 4. A voice input means for inputting a voice of a speaker, a voice recognition means for recognizing a word from a voice input from the voice input means, and a first memory for storing a plurality of words to be recognized in advance. And a phrase extraction unit for extracting a phrase from which the phrase recognized by the speech recognition unit is to be recognized in advance, from the first speech recognition resource, and the first speech recognition resource by the phrase extraction unit. And a speech recognition resource generating means for generating a second speech recognition resource having a phrase stored in the first speech recognition resource as a candidate phrase B, while the phrase extracted from When recognizing a phrase re-input from the voice input unit, the phrase is extracted from one of the candidate phrases A and B of the second speech recognition resource. Voice recognition device. 5. A voice input means for inputting a voice of a speaker, a voice recognition means for recognizing a phrase from a voice input from the voice input means, and a plurality of words to be recognized in advance for each word group. The stored first speech recognition resource, a phrase extraction unit for extracting a phrase recognized by the speech recognition unit from the first speech recognition resource, and a phrase extracted from the first speech recognition resource by the phrase extraction unit. A second speech recognition resource is generated in which the extracted phrase is set as a candidate phrase A, and a group of phrases including the phrase extracted by the phrase extraction unit is set as a candidate phrase B among the phrases of the first speech recognition resource. And a speech recognition resource generating unit that performs speech recognition of a word re-input from the speech input unit, and selects one of the candidate words A and B of the second speech recognition resource. A speech recognition device characterized by extracting a word from a phrase. 6. A voice input means for inputting a voice of a speaker, a voice recognition means for recognizing a phrase from a voice input from the voice input means, and a price term as a plurality of words to be recognized in advance. A first speech recognition resource to be stored, a word extraction unit for extracting a word recognized by the speech recognition unit from the first speech recognition resource,
Voice recognition resource generating means for generating, as a second voice recognition resource, a phrase obtained by deleting a phrase having a lower price than the sales amount of the product from the phrase of the price of the voice recognition resource. When inputting, the speech recognition device extracts a phrase of a price from the second speech recognition resource.