JPH11316597A - Voice recognition device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To decrease the number of registered codes needed to code voice- recognized words and phrases. SOLUTION: This device is provided with a voice input part 11 for inputting a speaker's voice, a voice recognition resource 12 which stores a language element code obtained by combining a classification code indicating each classification with an individual code so that it corresponds to each word or phrase, a voice recognition part 13 which recognizes a word or phrase from the voice inputted from the voice input part, extracts the language element code corresponding to each word or phrase from the voice recognition resource when the recognized word or phrase includes a word or phrase to be recognized in advance, and outputs it, and a code reconstitution part 14 which generates a meaning code by rearranging and connecting only individual codes in the predetermined order of classification codes as to language element codes outputted form the voice recognition part.

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 and coding words and phrases from input speech.

[0002]

2. Description of the Related Art As shown in FIG. 11, a conventional voice recognition apparatus has a voice input unit 1 including 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, which is a collection of language element codes defined for the phrase to be recognized;
Recognize words and phrases based on the output from the voice input unit 1,
The speech recognition unit 3 extracts a code corresponding to the word based on the speech recognition resource 2, and an application program 4 uses the code extracted by the speech recognition unit 3 as speech recognition data.

[0003] The speech recognition resource 2 is composed of words to be speech-recognized as shown in FIG. 12 and codes corresponding thereto. A phrase to be recognized as speech is a set of nouns, modifiers, numerals, etc.
Different codes correspond to all such phrases.

[0004] In such an apparatus, when a speaker inputs a phrase composed of a noun, a modifier, a number, etc., from the voice input unit 1, the voice recognition unit 3 collectively inputs a phrase defined in advance. The recognized code corresponding to the phrase has been output to the application program 4 as speech recognition data.

[0005]

However, in such a speech recognition apparatus, since a phrase composed of a noun, a modifier, a numeral, and the like is recognized as a collective phrase, a code is made to correspond to all these phrases ( (Associate) a speech recognition resource 2 had to be created. For this reason,
There is a problem in that the number of phrases to be recognized becomes enormous, and the work of creating the speech recognition resource 2 is extremely troublesome.

Since the number of words and codes registered is enormous, and the registered contents differ depending on the business to which this speech recognition device is to be applied, it must be created for each device. It was a factor that hindered the development of applications for

Accordingly, an object of the present invention is to provide a speech recognition apparatus capable of reducing the number of registration codes required for coding speech-recognized words and phrases.

[0008]

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 language element code storage means for classifying and storing a plurality of words to be recognized and a language element code storage means for storing a language element code obtained by combining an individual code with a code relating to each classification in association with each word, and a speech recognition means Language element code output means for extracting and outputting a code corresponding to each word from the language element code storage means when the extracted word includes a word to be recognized in advance, and a plurality of words output from the language element code output means. Semantic code creation that creates a semantic code by rearranging and concatenating only individual codes in the order based on the code for each classification for the language element code of A speech recognition apparatus characterized by comprising a stage.

According to a second 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, and a plurality of voice recognition means to be recognized in advance. A language element code storage means for storing words in a manner that classifies and stores the words, and stores a language element code obtained by combining a classification code indicating each classification with an individual code corresponding to each word, and a word recognized by the speech recognition means. A language element code output means for extracting and outputting a language element code corresponding to each word from the language element code storage means when including a word to be recognized in advance; and a plurality of languages output from the language element code output means A semantic code creating means for creating a semantic code by rearranging and connecting only the individual codes in the order of the predetermined classification codes for the element codes. A speech recognition apparatus characterized by was e.

According to a third 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, and a plurality of voice recognition means to be recognized in advance. A language element code storage unit that stores words in a manner that classifies and stores words, and stores a language element code obtained by combining an individual code with an order code indicating a sorting order of each classification in association with each word, and is recognized by a speech recognition unit. Language element code output means for extracting and outputting a language element code corresponding to each word from the language element code storage means when the extracted word includes a word to be recognized in advance, and a language element code output means Means for creating a semantic code by arranging and connecting only individual codes in the order of the order code for the plurality of language element codes. It is a speech recognition apparatus according to claim.

[0011]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, a first embodiment in which the present invention is applied to a business processing device such as an electronic cash register, a POS terminal or the like which performs merchandise sales code registration processing will be described with reference to FIGS. It will be described with reference to FIG.

FIG. 1 is a functional block diagram showing a configuration of a business processing device according to the present embodiment. The business processing device includes a microphone (voice input means) 11a for inputting voice as an analog signal and an A / D converter 11b for converting voice from the microphone 11a into a digital signal. A speech recognition resource (language element code storage unit) 12, which is a set of codes with classification (language element codes) defined below including the classification of words to be recognized, based on an output from a speech input unit 11. Recognize a phrase corresponding to the input speech (speech recognition means), extract a classified code corresponding to the phrase from the speech recognition resource 12, and output (language element code output means)
The speech recognition unit 13 generates a code sequence (meaning code) for transmitting information (meaning of a sentence) such as what product and how many based on the classified code extracted by the speech recognition unit 13. It comprises a code restructuring unit 14 as means for generating a meaning code to be output, and an application program 15 that uses the meaning code from the code restructuring unit 14.

The voice recognition resource 12 is constituted by a storage device such as a hard disk device. Specifically, as shown in FIG. 2, the words to be recognized by speech are classified by word, the classification name, the words in words belonging to each classification, and the same classification code and the same words in each classification. And a classification code corresponding to each word and phrase. For example, for a word "paper" whose classification name is "thing", a classification code of "O" and "P
A classifying code consisting of “O-PA” consisting of an individual code “A” is associated. In addition, for a phrase “large” whose classification name is “size”, a classification code “SL” including a classification code “S” and an individual code “L” is associated.

The speech recognition unit 13, the code reconstruction unit 14, and the application program 15
It comprises a personal computer having a U (central processing unit), a ROM (read only memory), and a RAM (random access memory). The voice recognition unit 13, the code reconstruction unit 14, and the application program 15 are specifically stored in a storage device such as a hard disk device or a memory such as a ROM, and are readable by a CPU of the personal computer. It consists of.

The voice recognition unit 13 defines (corresponds to) voice input based on the output from the voice input unit 11 and voice feature data in the voice recognition resource 12 in advance.
Detecting similarity / approximation with the word (for example, making the speech recognition resource 12 correspond to the same word with speech feature data of a plurality of types of phrases meaning the same word, and input based on this. Then, the speech recognition is performed to identify the uttered phrase), and the speech recognition is performed. A classified code corresponding to the phrase obtained by the speech recognition is extracted from the speech recognition resource 12 and output.

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

The code restructuring unit 14 stores the classified code output from the speech recognizing unit 13 in a memory assigned in a predetermined order of classification, that is, in the order of physical value, size value, and quantity value. Only the individual code of each classification code is stored. As a result, the classified codes are rearranged for each classification, and a semantic code that is linked in a state where the classified codes are deleted is generated.

For example, as shown in FIG. 3A, if the utterance content is "three large papers", the classification code is "S-
L, “O-PA”, and “N-0003” are output in this order. The code reconstructing unit 14 assigns the physical value, size value, and quantity assigned to each classification as shown in FIG. Individual codes "PA", "L", "000" are stored in each value memory.
"PAL000" generated by storing only "3"
"3" is output as a meaning code.

The above application program 15
Based on the semantic code from the code restructuring unit 14, it is constituted by a software program for performing predetermined business processes such as registration of a product sales code and calculation of a price. When the application program 15 utters the product name and the number, the meaning code is output from the code restructuring unit 14.
This meaning code is used to select and display the corresponding product name on a screen such as a display, and to perform subsequent accounting processing such as registration of the product code and calculation of the price.

In the embodiment of the present invention having such a configuration, for example, when the user of this apparatus utters "three large papers" as shown in FIG. At 11, it is converted into a digital signal and supplied to the voice recognition unit 13. Then, the speech recognition unit 13 refers to the speech recognition resource 12 and detects the similarity / approximation between the input speech and a phrase defined in advance in the speech recognition resource, as shown in FIG. For "Large", "S
-L ", and" OP "for" paper "
"A" and "N-00" for "3"
03 "are sequentially output. These classified codes are passed to the next code restructuring unit 14, and the individual codes "P" are stored in the memory for the physical value, size value, and quantity value assigned to each classification as shown in FIG.
A, “L” and “0003” are stored and “PAL” is stored.
0003 ”is generated and passed to the application program 15. This meaning code has a meaning of “three large papers” for the application program 15. In other words, "big",
The words “paper” and “three” are independently recognized one by one, converted into associated classification codes, and combined with the classification codes to form semantic codes.

As described above, the words and phrases to be recognized are classified on a word-by-word basis, and a code with classification including a classification code and an individual code is stored corresponding to (defined) each word (word). Alone, semantic codes can be generated by the number of products of the registered numbers of words belonging to each classification.

Therefore, when the speech recognition resource 12 is created, it is not necessary to register codes for one sentence one by one as in the prior art. May be registered (stored). Thus, the number of codes stored in the speech recognition resource 12 can be significantly reduced. As a result, the speech recognition resource 12 can be easily created, and the retrieval speed can be increased, so that the encoding speed of the speech information can be increased.

Also, since the speech recognition resources 12 are stored for each classification, addition, deletion,
Work such as correction becomes easy.

When a sentence having the same meaning is uttered,
Even if the utterance order and wording of the words that compose it are different,
Since the meaning codes are created by being rearranged in a predetermined classification order and passed to the application program 15, even if the words constituting the sentence have different utterance orders / phrases, they can be recognized as sentences having the same meaning. .

Next, referring to FIGS. 4 and 5, a second embodiment in which the present invention is applied to a business processing device such as an electronic cash register, a POS terminal or the like for performing a merchandise sales code registration process will be described. explain. The same parts as those in the first embodiment are denoted by the same reference numerals, and detailed description is omitted. Further, the functional block diagram of the business processing device according to the present embodiment is the same as that shown in FIG. 1, and therefore, detailed description thereof will be omitted.

Speech recognition resource 1 in the present embodiment
2 differs from the first embodiment in that “amount” is registered (stored) instead of “size” as a classification name as shown in FIG.

In the embodiment of the present invention having such a configuration, for example, when the user of the present apparatus utters “two 50-yen boxes” as shown in FIG. It is converted into a digital signal by the input unit 11 and
Supplied to Then, the speech recognition unit 13 refers to the speech recognition resource 12 and detects the similarity / approximation between the input speech and a phrase defined in advance in the speech recognition resource, as shown in FIG. A classification code "K-0050" for "50 yen", a classification code "O-BO" for "box", and a classification "N-0002" for "2" The attached codes are output in order. These classified codes are passed to the next code restructuring unit 14 and stored in the memories of the physical value, the monetary value, and the quantity value assigned to each classification as shown in FIG. , “0050”, and “0002” are stored, and a meaning code “BO00500002” is generated and passed to the application program 15. This meaning code has a meaning to the application program 15 of “two 50-yen boxes”. In other words, the words "50 yen", "box", and "two" are individually recognized one by one, converted into corresponding (associated) classification codes, and combined with the classification codes. It becomes a meaning code.

As described above, the words and phrases to be recognized are classified on a word-by-word basis, and a code with classification including a classification code and an individual code is stored in association with (defined) each word (word). Only in the same manner as in the first embodiment, it is possible to generate the meaning codes by the number of products of the registered numbers of the words belonging to each classification. Therefore, the same effect as in the first embodiment can be obtained.

Further, in the present embodiment, since the price information is registered as the classification name, by referring to the "money" of the meaning code generated from the classification-added code, not only the characteristics of the product, but also the The price information of the product can be obtained.

Next, referring to FIGS. 6 and 7, a third embodiment in which the present invention is applied to a business processing device such as an electronic cash register, a POS terminal or the like which performs a merchandise sales code registration process will be described. explain. The same parts as those in the first embodiment are denoted by the same reference numerals, and detailed description is omitted. Further, the functional block diagram of the business processing device according to the present embodiment is the same as that shown in FIG. 1, and therefore, detailed description thereof will be omitted.

Speech recognition resource 1 in the present embodiment
2 differs from the first embodiment in that “color” and “amount” are added as classification names as shown in FIG.

In the embodiment of the present invention having such a configuration, for example, as shown in FIG. 7A, when the user of this apparatus utters "one blue and large box of 100 yen",
This voice is converted into a digital signal by the voice input unit 11 and supplied to the voice recognition unit 13. Then, the voice recognition unit 13
Refers to the speech recognition resource 12 to detect the similarity / approximation between the input speech and the phrase defined in advance in the speech recognition resource, and as shown in FIG.
For "yen", a classification code "K-0100",
Classified code "CB" for "blue", Classified code "SL" for "Large", Classified code "O-BO" for "Box" , "1" are sequentially output with classification codes of "N-0001". These classified codes are stored in the following code restructuring unit 1.
4, the individual codes "BO", "L", "B",
Only “0100” and “0001” are stored and “BOL”
A meaning code “B01000001” is generated and passed to the application program 15. This meaning code is, for the application program 15,
It means "one blue and big box of 100 yen". That is, "100 yen", "blue", "large",
The words "box" and "one" are independently recognized one by one, converted to corresponding (associated) classification codes, and combined with the classification codes to form semantic codes.

As described above, the words and phrases to be recognized are classified on a word-by-word basis, and the classified code including the classification code and the individual code is stored in association with (defined) each word (word). Only in the same manner as in the first embodiment, it is possible to generate semantic codes by the number of products of the registered numbers of words belonging to each classification. Therefore, the same effect as in the first embodiment can be obtained.

In particular, the speech recognition resource 12 according to the present embodiment has four types of phrases belonging to the classification "object",
There are three types of terms belonging to the category "size", four types of terms belonging to the category "color", three types of terms belonging to the category "money", and three types of terms belonging to the category "quantity". Since the code is registered, a total of 4 ×
Although 3 × 4 × 3 × 3 = 432 kinds of semantic codes can be handled, when such 432 kinds of semantic codes are handled, the number of 432 codes must be registered in the related art. On the other hand, in the present embodiment, only 17
The speech recognition resource 12 can be created only by registering the type classification code in the speech recognition resource. That is, the number of codes stored in the speech recognition resource 12 can be further reduced.

Next, a fourth embodiment in which the present invention is applied to the business processing apparatus as described above will be described with reference to FIGS.
0 will be described. The same parts as those in the first embodiment are denoted by the same reference numerals, and detailed description is omitted.

As shown in FIG. 8, the business processing apparatus according to the present embodiment includes a voice input unit 11 and a voice recognition resource 1.
2 ', a speech recognition unit 13, a code restructuring unit 14, an application program 15 using the semantic code from the code restructuring unit 14, and a language element code before the semantic code is created by the code restructuring unit 14. It comprises a reordering unit 21 for reordering.

The speech recognition unit 13, the code restructuring unit 14, the application program 15,
And the rearranging unit 21 includes a CPU (Central Processing Unit) / RO
It is composed of a personal computer having M (read only memory) and RAM (random access memory). The speech recognition unit 13, the code reconstruction unit 14, the application program 15, and the rearrangement unit 21 are specifically stored in a storage device such as a hard disk device or a memory such as a ROM, and read by the CPU of the personal computer. It consists of possible software programs.

Speech recognition resource 1 in the present embodiment
2 'differs from the case of the first embodiment in that FIG.
And language element codes corresponding to these words and phrases to be recognized. This language element code is composed of a combination of a key value indicating the order of classification and a code (individual code) indicating an actual code number. Here, similarly to the first embodiment, words to be recognized by speech are classified into “product name” and “quantity”, and the same classification is associated with the same key value and the code thereof. .

For example, a language element code consisting of a key value “1” and a code “01” is associated with the phrase “product A”, and a key value “2” and a code A language element code consisting of "01" is associated. In this way, by classifying words and associating key values indicating the order of each classification, by rearranging the language element codes in the order of the key values, the same code is obtained regardless of the output order of the language element codes.

It should be noted that, although not shown, the words to be recognized by speech may be stored in advance in association with the voice feature data of the standard speaker (unspecified speaker-compatible type). May be associated with the uttered voice feature data (specific speaker-compatible type), as in the first embodiment.

Further, when code reconstructing section 14 in the present embodiment receives two language element codes from speech recognition section 13, it passes them to rearranging section 21. Then, the rearranging unit 21 rearranges the received language element codes in the order of the key values, and returns the language element codes to the code reconstructing unit 14 in the rearranged order. Thereby, the code restructuring unit 14
Deletes the key value from each language element code and concatenates it to make it a four-digit meaning code and passes it to the application program 15. As a result, the same semantic code is output from the code restructuring unit 14 even when the utterance order of the product name and the number is different, as in the first embodiment.

In the embodiment of the present invention having such a configuration, for example, when the user of the present apparatus utters "3 products A" as shown in FIG. At 11, it is converted into a digital signal and supplied to the voice recognition unit 13. Then, the speech recognition unit 13 refers to the speech recognition resource 12 ′, detects the similarity / approximation between the input speech and a phrase defined in advance in the speech recognition resource, and for “product A”, The language element code “101” is output, and the language element code “203” is output for “3” in that order. These language element codes are passed to the next code restructuring unit 14, and are rearranged by the rearranging unit 21 according to the key values defined in the language element codes. That is, since the key value of "commodity A" is "1", it is arranged at the head, and the key value of "three" is "2", and is arranged second.

After the rearrangement is performed, these language element codes are returned to the code reconstructing unit 14, where each key value is deleted, and concatenated to generate a meaning code "0103". 15 is passed. This meaning code corresponds to application program 1
For 5, it has a meaning of “three products A”.

On the other hand, the user of the apparatus shown in FIG.
As shown in (b), when "3 products A" is uttered, the voice is supplied to the voice recognition unit 13 via the voice input unit 11 as described above, and the voice recognition unit 13 converts the voice into "3". The language element code “203” is output to the user, and the language element code “101” is output to the “product A” in that order. These language element codes are passed to the next code restructuring unit 14, and are rearranged by the rearranging unit 21 according to the key values defined in the language element codes. That is, since the key value of "commodity A" is "1", it is arranged at the head, and the key value of "three" is "2", and is arranged second.

After the rearrangement is performed, these language element codes are returned to the code reconstructing unit 14, where each key value is deleted and concatenated to generate a meaning code "0103". It is passed to the program 15. This meaning code corresponds to application program 1
For 5, it has a meaning of “three products A”. In other words, the words "commodity A" and "three" are individually recognized one by one, converted into associated language element codes, and combined with the language element codes to form a semantic code.

As described above, the words and phrases to be recognized are classified in units of words, and the language element codes including the order codes and the individual codes are stored so as to correspond (define) to the respective words (words). Only in the same manner as in the first embodiment, it is possible to generate the meaning codes by the number of products of the registered numbers of the words belonging to each classification. Therefore, the same effect as in the first embodiment can be obtained.

When a sentence having the same meaning is uttered,
Even if the utterance order and wording of the words that compose it are different,
Since the meaning code “0103” having the same meaning is created and can be passed to the application program 15, the speech recognition resource 1 is set in the same manner as in the first embodiment.
The number of codes stored in 2 'can be greatly reduced, and even if the words forming the sentence have different utterance orders / phrases, they can be recognized as sentences having the same meaning.

[0048]

As described above, according to the present invention, when creating a speech recognition resource (language element code storage means), it is necessary to register codes for one sentence one by one as in the prior art. Instead, for each class, a code with classification obtained by combining a classification code with an individual code for a phrase belonging to the classification may be registered. As a result, the number of codes registered (stored) in the speech recognition resource can be greatly reduced, so that the creation of the speech recognition resource is facilitated and the search speed is also increased, so that the speed of coding the speech information is increased. Become.

Further, since the speech recognition resources are stored for each classification, operations such as addition, deletion, and modification of classification-added codes are facilitated.

Further, when uttering a meaningful sentence,
Even if the utterance order and wording of the words constituting the sentence are different, the same meaning code is created, so that the sentence having the same meaning can be recognized.

[Brief description of the drawings]

FIG. 1 is a functional block diagram showing a configuration of a business processing device according to a first embodiment of the present invention.

FIG. 2 is a diagram showing a configuration of a speech recognition resource shown in FIG.

FIG. 3 is a diagram illustrating an operation of the embodiment.

FIG. 4 is a diagram showing a configuration of a speech recognition resource of a business processing device according to a second embodiment of the present invention.

FIG. 5 is a diagram illustrating the operation of the present embodiment.

FIG. 6 is a diagram showing a configuration of a speech recognition resource of a business processing device according to a third embodiment of the present invention.

FIG. 7 is a diagram illustrating the operation of the present embodiment.

FIG. 8 is a functional block diagram showing a configuration of a business processing device according to a fourth embodiment of the present invention.

FIG. 9 is a diagram showing a configuration of a speech recognition resource shown in FIG. 8;

FIG. 10 is a diagram illustrating the operation of the present embodiment.

FIG. 11 is a functional block diagram showing a configuration of a business processing device to which a conventional voice recognition device is applied.

FIG. 12 is a diagram showing a configuration of a speech recognition resource shown in FIG. 11;

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 11 ... Speech input part 11a ... Microphone 11b ... A / D converter 12 ... Speech recognition resource 13 ... Speech recognition part 14 ... Code reconstruction part 15 ... Application program 21 ... Sort part

Claims (3)

[Claims] 1. A speech input means for inputting a voice of a speaker, a speech recognition means for recognizing a phrase from a speech input from the speech input means, and a plurality of phrases to be recognized in advance are classified and stored. And a language element code storage means for storing a language element code obtained by combining a code for each classification with an individual code in association with each word, and a word for which the word recognized by the voice recognition means is to be recognized in advance. When including, a language element code output means for extracting and outputting a code corresponding to each word from the language element code storage means, and a plurality of language element codes output from the language element code output means, a code relating to each classification. And semantic code creating means for creating a semantic code by rearranging and connecting only individual codes in an order based on Voice recognition device. 2. A speech input means for inputting a voice of a speaker, a speech recognition means for recognizing a phrase from speech inputted from the speech input means, and a plurality of phrases to be recognized are classified and stored. Language element code storage means for storing a language element code obtained by combining a classification code indicating each classification with an individual code in association with each word, and a word recognized by the voice recognition means should be recognized in advance. When including a phrase, a language element code output means for extracting and outputting a language element code corresponding to each word from the language element code storage means, and a plurality of language element codes output from the language element code output means, A semantic code creating means for creating a semantic code by rearranging and connecting only individual codes in a predetermined classification code order. Characteristic speech recognition device. 3. Speech input means for inputting a speaker's speech, speech recognition means for recognizing a phrase from speech input from the speech input means, and classifying and storing a plurality of phrases to be recognized in advance. Language element code storage means for storing a language element code in which an individual code is combined with an order code indicating a rearrangement order of each classification in association with each word, and a word recognized by the speech recognition means in advance. A language element code output means for extracting and outputting a language element code corresponding to each word from the language element code storage means when including a word to be recognized; and a plurality of languages output from the language element code output means. About element code,
A speech recognition device comprising: a meaning code creating unit that creates a meaning code by arranging and connecting only individual codes in the order of the order codes.