JPH1055196A - Speech recognition device and method, information storage medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To make possible precisely and rapidly recognizing plural words and phrases related in content from one continuous voice. SOLUTION: Central words and co-occurrence words existing in a co- occurrence relation are combined to be stored beforehand in a recognition word/phrase dictionary 22, and the central words stored in the recognition word/phrase dictionary 22 are extracted from one continuous input voice, and the co-occurrence words existing in the co-occurrence relation with the extracted central words are read out from the recognition word/phrase dictionary 22 to be extracted from the input voice. Since the co-occurrence words collated with the input voice are narrowed down based on the extraction results of the central words, plural words and phrases are recognized precisely and rapidly from one input voice.

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 and method for recognizing speech, and an information storage medium on which software such as a program is written.

[0002]

2. Description of the Related Art At present, there is a demand for a voice recognition device for recognizing a voice uttered by a human, and various voice recognition methods are being considered. When a human utters only one word, which is a phrase, it is not difficult for the speech recognizer to recognize it, but in a natural human conversation, speech is continuous, and many words contain particle And so on. As a method for recognizing necessary words from continuous speech, word spotting has been devised. In this method, words of preset recognition candidates are extracted from continuous speech and recognized.

A speech recognition apparatus that performs such word spotting has a word dictionary in which readings are stored for each recognition candidate word. And the largest word whose score of the matching exceeds the reference value is recognized from the conversation voice. However, good results cannot be expected even if the word spotting is simply executed, so that it is now general to improve recognition accuracy due to linguistic restrictions.

Such linguistic constraints include those based on syntactic properties such as grammar, and those based on semantic constraints such as case patterns. The former is powerful as a constraint, but it is not suitable for spoken language that is difficult to regularize unlike written language, and its recognition control can be used only within phrases. The latter is suitable for the characteristics of Japanese and the spoken language in which the word order is free, and when used for the word spotting described above, the recognition accuracy can be improved satisfactorily.

[0005]

As described above, in word spotting, only necessary words and phrases can be recognized from continuous speech. In particular, recognition accuracy can be improved by using semantic constraints.

For example, in the speech recognition method of the speech recognition apparatus disclosed in Japanese Patent Application Laid-Open No. 6-102897, a phrase to be recognized is predicted using case relations, and narrowing is performed. However, this cannot deal with relationships other than case relationships, nor can it handle multiple relationships. In other words, the relationship between words included in the continuous speech is not limited to the case relationship, but it is also common that one continuous speech includes a plurality of relationships.

[0007] For example, if the conversation voice is "Please give a box of curry flavor of cup ramen", it is difficult to recognize words even by using case relations. In addition, when the conversation voice is “crying child is also silent”, there are a plurality of relationships of “crying / child” and “child / silent”.

[0008]

According to a first aspect of the present invention, there is provided a voice recognition apparatus for storing a combination of voice input means for receiving a continuous input of a voice to be recognized and a plurality of co-occurring words. And a phrase recognition unit for recognizing a plurality of phrases combined in a co-occurrence relationship from continuous input speech. Therefore, since a plurality of words having a co-occurrence relation are stored in combination in the recognition word dictionary, a plurality of words combined in a co-occurrence relation from the speech to be recognized continuously input to the voice input means are stored. Is recognized by the phrase recognition means. In other words, if predetermined words are combined and set in a co-occurrence relationship, a plurality of words in this co-occurrence relationship cannot be individually recognized from one continuous input voice, Is recognized based on

According to a second aspect of the present invention, in the speech recognition apparatus according to the first aspect, the word / phrase recognizing means reads out and inputs a central word which is one of a pair of words combined in a co-occurrence relation from the recognized word / phrase dictionary. After being extracted from the voice, an additional word, which is the other word having a co-occurrence relationship with the extracted central word, is read from the recognized phrase dictionary and extracted from the input voice. Therefore, when a plurality of words are recognized from one continuous input voice by the phrase recognition means, a central word which is one of the words is first extracted from the input voice, and then a co-occurrence relation with the central word is obtained. One other phrase, the adjunct, is then extracted from the input speech. In other words, as in the case of the conventional word spotting, a large number of central words are collated with the input voice, but the attached words are narrowed down based on the co-occurrence relation and then collated with the input voice.

According to a third aspect of the present invention, in the speech recognition apparatus according to the second aspect, the phrase recognizing means extracts an auxiliary word from a section excluding a section from which a central word of the input speech is extracted. Therefore, when a central word and an auxiliary word having a co-occurrence relation are recognized from one continuous input voice by the phrase recognition means, first, a central word is extracted from the entire input voice, and this central word is extracted. Adjectives are extracted from sections other than the section that has been set. In other words, the central word will be matched against the input voice as in the conventional word spotting, but the attached words will be narrowed down based on the co-occurrence relationship and then matched against the voice section that does not overlap with the central word .

According to a fourth aspect of the present invention, in the speech recognition apparatus according to the second or third aspect, the recognition word dictionary is further provided with information on the order of the combination of the central word and the adjunct word, and the word recognition means is , The central word and the auxiliary word are recognized from the input speech in order. Accordingly, information on the order in which the central word and the auxiliary word occur in the continuous speech is also stored in the recognition phrase dictionary, and when the central word and the auxiliary word occur in the input voice in a predetermined order, they are recognized by the phrase recognition means. Therefore, even when the central word and the auxiliary word are individually recognized from the input voice, they are not recognized unless the order is proper.

According to a fifth aspect of the present invention, in the speech recognition apparatus according to the fourth aspect, the recognition word dictionary also stores an intervening word located in the middle of the combination of the central word and the adjunct word. Recognizes a central word, intervening words, and adjunct words from input speech. Therefore, the intervening word located in the middle of the combination of the central word and the adjunct word is also stored in the recognition phrase dictionary, and the central word, the intervening word and the adjunct word are recognized from the input speech by the phrase recognition means. Even if the reading of the adjunct of the central word and the intervening word is the same as the reading of the combination of the other central words and adjuncts, they are each recognized separately.

According to a sixth aspect of the present invention, in the speech recognition apparatus according to the second or third aspect, the recognition word dictionary is provided with time interval information for a combination of a central word and an adjunct word. Recognizes a central word and an adjunct word from an input voice corresponding to a time interval. Therefore, information on the time interval in which the central word and the adjunct word occur in the continuous input speech is also stored in the recognition phrase dictionary, and the central word and the adjunct word are recognized from the input speech in correspondence with the time interval. Therefore, the central word and the adjunct word are recognized only when they occur in the input speech at appropriate time intervals. Even when the central word and the adjunct word are individually recognized from the input speech, the time interval is Not recognized as improper.

According to a seventh aspect of the present invention, in the speech recognition apparatus according to the second or third aspect, the recognition phrase dictionary stores a combination of a central word and an adjunct word in a hierarchical structure having a plurality of stages, and performs phrase recognition. The means recognizes one central word and a plurality of attached words stepwise from the input speech in accordance with the hierarchical structure. Therefore, the combination of the central word and the adjunct word is stored in the recognition phrase dictionary as a multi-stage hierarchical structure, and one central word and a plurality of adjunct words are recognized by the phrase recognizing means according to the hierarchical structure from the input speech. Recognized step by step. That is, when one central word and one auxiliary word are recognized from a certain input voice, other auxiliary words having this auxiliary word as the central word are also searched from the input voice, and such processing operations are sequentially repeated. Therefore, one central word and a plurality of attached words that are in a co-occurrence relationship in a plurality of stages are sequentially recognized.

According to an eighth aspect of the present invention, in the speech recognition apparatus according to the seventh aspect, the recognition word dictionary is provided with order information for a combination of one central word and a plurality of attached words. The means recognizes one central word and a plurality of attached words in order from the input speech. Therefore, the information on the order of the combination of one central word and a plurality of attached words is also stored in the recognition phrase dictionary, and the one central word and the plurality of attached words correspond in order from the input voice to the phrase recognition means. Therefore, even when one central word and a plurality of attached words are individually recognized from the input speech, they are not recognized unless the respective orders are not proper.

According to a ninth aspect of the present invention, in the speech recognition apparatus according to the seventh or eighth aspect, the recognition phrase dictionary is also provided with information on a depth of a hierarchical structure to a combination of one central word and a plurality of attached words. The word recognition means recognizes one central word and a plurality of attached words from the input speech in accordance with the depth. Therefore, the information of the depth of the hierarchical structure of the combination of one central word and a plurality of attached words is also stored in the recognition phrase dictionary, and one central word and the plurality of attached words correspond to the depth from the input speech. Since it is recognized by the phrase recognition means,
Stepwise recognition of a plurality of adjunct words defining one central word is executed to a predetermined depth.

According to a tenth aspect of the present invention, a plurality of words having a co-occurrence relation are set in combination, a continuous input of a speech to be recognized is received, and co-occurrence is performed from the continuous input speech. Recognize multiple words combined in relationships. Therefore, if predetermined words are combined and set in a co-occurrence relationship, a plurality of words in this co-occurrence relationship cannot be individually recognized from one continuous input voice, and Is recognized based on

In the speech recognition method according to the eleventh aspect, a co-occurrence central word and an adjunct word are set in combination, a continuous input of a speech to be recognized is received, and the continuous input speech is received. A central word prepared from is extracted, and an auxiliary word having a co-occurrence relation with the central word is extracted from the input speech. Therefore, when the central word and the adjunct of the co-occurrence relation are recognized from one continuous input voice, the central word is first extracted from the input speech, and then the auxiliary word having the co-occurrence relation with the central word is recognized. Is then extracted from the input speech. That is,
As with the word word spotting in the related art, a large number of words are collated with the input voice, but the attached words are narrowed down based on the co-occurrence relation and then collated with the input voice.

According to a twelfth aspect of the present invention, there is provided an information storage medium in which a plurality of words having a co-occurrence relation are combined and stored in an information storage medium in which computer-readable software is written in advance. ,
And a program for recognizing a plurality of phrases combined in a co-occurrence relationship from continuous input speech. Therefore, if the computer reads and operates the software of the information storage medium, the computer functions as a speech recognition device that recognizes words stored in the recognized word dictionary from continuous input speech. At this time, since a plurality of words having a co-occurrence relationship are stored in combination in the recognition word dictionary, a plurality of words combined in a co-occurrence relationship are recognized from continuous input speech. In other words, if predetermined words are combined and set in a co-occurrence relationship, a plurality of words in this co-occurrence relationship cannot be individually recognized from one continuous input voice, Is recognized based on

According to a thirteenth aspect of the present invention, there is provided an information storage medium in which a computer-readable software is written in advance. And a program for reading a central word from the recognized phrase dictionary and extracting it from continuous input speech, and reading and inputting an auxiliary word having a co-occurrence relationship with the extracted central word from the recognized phrase dictionary And a program for extracting from the voice. Therefore, if the computer reads and operates the software of the information storage medium, the computer functions as a speech recognition device that recognizes words stored in the recognized word dictionary from continuous input speech. At this time, when a plurality of words are recognized from one continuous input voice by the phrase recognition means, a central word which is one of the words is first extracted from the input voice, and then a co-occurrence relation with the central word is obtained. Is then extracted from the input speech. In other words, as in the case of the conventional word spotting, a large number of central words are collated with the input voice, but the attached words are narrowed down based on the co-occurrence relation and then collated with the input voice.

[0021]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the present invention will be described below with reference to the drawings. First, as shown in FIGS. 2 and 3, the voice recognition device 1 of the present embodiment has a computer system as a data processing device as hardware. The speech recognition device 1 including the computer system includes a CPU (Centr
al Processing Unit) 2, and the CPU 2 is connected to a ROM (Read Only Memory) by a bus line 3.
4, RAM (Random Access Memory) 5, HD (Hard Disk)
... HDD (HD Drive) 6 with built-in
ppy Disk) 7, FDD (FD Drive) 8, CD (C
ompact Disk) -CD-ROM drive 10 loaded with ROM 9, keyboard 12 connected to mouse 11, display 13, microphone 1 as input device
4, communication I / F (Interface) 15, etc. are connected.

In the speech recognition apparatus 1, a program for causing the CPU 2 to execute various processing operations is set in advance, and software such as this program is stored in, for example, the FD 7 or the CD-ROM 9 which is an information storage medium. It has been written in advance. This software is installed in the HDD 6 as an information storage medium, and is copied to the RAM 5 as an information storage medium and read by the CPU 2 at startup.

As described above, various functions are realized as various means by the CPU 2 reading the software and executing various processing operations. As such various means, the voice recognition device 1 of the present embodiment includes a voice input unit 21, a recognized phrase dictionary 22, a phrase recognition unit 23, a result output unit 24, and the like, as shown in FIG. .
The recognition phrase dictionary 22 includes a word dictionary 25, a co-occurrence dictionary 2
6, word order dictionary 27, said word recognition means 23
Consists of a feature extracting unit 28, a candidate recognizing unit 29, a candidate reading unit 30, a phrase searching unit 31, a result determining unit 32, and the like.

Various means of such a speech recognition device 1 include:
If necessary, hardware such as the display 13 and the microphone 14 is used, and the main component is realized by the operation of the CPU 2 corresponding to software written in the RAM 5 or the like. For this reason, as shown in FIG.
A dictionary storage unit 41 which is software of the recognition phrase dictionary 22, a control program 42 for recognizing a plurality of phrases combined in a co-occurrence relationship from continuous input speech,
Is written.

More specifically, the dictionary storage unit 41 comprises a word dictionary storage unit 43, a co-occurrence dictionary storage unit 44, and a word order dictionary storage unit 45. These storage units 43 to 45 store the dictionaries 25 to 27 Various information is stored. The control program 42 is formed in a module structure and includes a feature extraction module 46, a word matching module 47, a recognition control module 48, and the like.

The feature extraction module 46 comprises a program for analyzing a digital signal of a continuous input speech for each frame and extracting a feature, and the word matching module 47 recognizes the feature of the input speech. The program comprises a program for collating feature amounts which are readings of candidate words and calculating a cumulative similarity as a score. The recognition control module 48 reads various types of information from the dictionaries 25 to 27 and transmits the information to the word matching module 47. From the program for determining the word of the recognition result based on the score output by the word matching module 47, Become.

The recognition phrase dictionary 22 is stored as a data file in the dictionary storage unit 41 of the RAM 5 as described above, and as shown in FIG. Are registered for each of a plurality of words. In the co-occurrence dictionary 26, a plurality of words having a co-occurrence relation are stored here in a combination of one central word and a plurality of attached words. In the word order dictionary 27, information on the meaning of the central word and the auxiliary word is set in order, so that information on the order of the central word and the auxiliary word is stored.

Although the various information in the dictionaries 25 to 27 are expressed as Japanese characters for the sake of simplicity of description, words and the like are composed of identification codes in actual software, and the reading is performed in words. It is set as a feature value of the voice. The feature amount of the reading voice is set, for example, as a state transition model in phoneme units and a phoneme network in word units, and information on the average feature amount and duration is set in each state.

The voice input means 21 receives continuous voice input from the microphone 14 or the like, and converts the input voice into a digital electric signal by A / D (Analog / Digilog).
tal) The phrase recognizing means 23 includes the CPU 2
Reads the control program 42 stored in the RAM 5 and executes the corresponding processing operation, thereby recognizing the co-occurrence central word and adjunct word from the continuous input voice digital signal.

More specifically, the feature extracting means 28 of the phrase recognizing means 23 reads the feature extracting module 46 of the control program 42 and executes the corresponding arithmetic processing to execute the continuous input. The audio digital signal is analyzed for each frame, which is a unit time, and a feature amount is extracted by calculating, for example, an LPC (Linear Predictive Coding) mel-cepstral.

The candidate recognizing means 29 reads out all the central words from the co-occurrence dictionary 26 of the recognized phrase dictionary 22 by the arithmetic processing of the CPU 2 corresponding to the recognition control module 48, and then reads each of them. Read from the word dictionary 25. Further, the word matching module 4
7, the reading of all the central words is compared with the feature amount of the continuous input voice, the similarity of each of them is calculated for each frame, and sequentially accumulated,
One central word whose score as the accumulated similarity exceeds the reference value at the maximum is extracted as a recognition candidate.

The candidate reading means 30 uses the CPU 2 corresponding to the recognition control module 48 to perform all the additional words co-occurring with the central word of the recognition candidate as recognition candidates by the CPU 2. Detected from the co-occurrence dictionary 26, the word / phrase search means 31 performs an arithmetic process of the CPU 2 corresponding to the word matching module 47,
Match all attached words of the recognition candidate to the input speech.

In this processing operation, similarly to the case of the central word, the reading of the adjunct word of the recognition candidate is read from the word dictionary 25 of the recognized phrase dictionary 22, and the score of the collation between the word and the input speech is the maximum. One adjunct that exceeds the value is extracted. However, when the reading of the attached word is compared with the feature amount of the input voice in this way, the central word and the attached word are obtained from the word order dictionary 27 of the recognized phrase dictionary 22 by the arithmetic processing of the CPU 2 corresponding to the recognition control module 48. Is read out, and the reading of the attached word is collated only in a section other than the section in which the central word of the input voice is extracted corresponding to the order.

When the attached word of the recognition candidate is searched for from the input speech as described above, the result determining means 32 calculates the central word of the recognized candidate and the attached word by the arithmetic processing of the CPU 2 corresponding to the recognition control module 48. A word is determined as a recognition result. The result output means 24 outputs the central word and the auxiliary word of the recognition result determined as described above, for example, by character display on the display 13.

In such a configuration, the speech recognition apparatus 1 according to the present embodiment, when a speech of a conversation in which a large number of words are continuous is input to the microphone 14, converts the continuous input speech into the recognized phrase dictionary 22. The stored words are recognized, and the recognition results are displayed on the display 13.
Such a voice recognition method of the voice recognition device 1 will be sequentially described in detail below with reference to FIG.

First, the whole area of the voice continuously input to the microphone 14 is A / D converted into a digital signal, and the characteristic amount is extracted for each frame. Next, the readings of all the central words are read from the recognized phrase dictionary 22, and all the readings are collated with the feature values of the whole input voice. When the cumulative similarity of collation is calculated as a score for all the central words in this way, only one central word having the largest score exceeding the reference value is selected as a recognition candidate.

When the central word of the recognition candidate is selected as described above, all the auxiliary words co-occurring therewith are read out from the recognition phrase dictionary 22 as the recognition candidate, and the same as in the case of the central word described above. Then, the reading of all the adjuncts is compared with the input voice, and one adjunct having the maximum score exceeding the reference value is selected. At this time, the word order dictionary 2 of the recognition phrase dictionary 22
7, information on the order of the central word and the auxiliary word is read out, and the reading of the auxiliary word is collated only in a section other than the section in which the central word of the input voice is extracted in accordance with the order.

When the co-occurrence of the central word and the adjunct word is extracted from one continuous input voice, they are determined as a recognition result and output by the character display on the display 13. If the scores of the collation of all the central words do not exceed the reference value at first, no recognition result is found, and the attached word is stored in the recognized phrase dictionary 22 in a state where the candidate for the central word is detected. If there is no matching word, or if the scores of collation of all attached words do not exceed the reference value, the recognition result is only the central word.

The above-described series of processing operations will be described in detail with reference to FIG. 5. If one continuous input voice is "order is curry taste of cup ramen""Cup ramen, hand-baked rice cracker" is collated, and "cup ramen" with a high score is extracted as a candidate for recognition of the central word. The curriculum flavor, soy sauce flavor is an auxiliary word that co-occurs with this central word “cup ramen”
Is detected from the recognition phrase dictionary 22, and the recognition phrase dictionary 2
2 specifies that an adjunct with the meaning “taste” is located behind the central word “cup ramen”, and therefore, from continuous input voice, the section “no” after “cup ramen” It has curry flavor. " Since the attached words “curry taste, soy sauce taste” are collated only for this voice section, the attached word “curry taste” is recognized.

In the speech recognition method of the speech recognition apparatus 1 according to the present embodiment, the central word and the adjunct words combined in a co-occurrence relation are recognized from one continuous input speech as described above. Is more accurate than when recognizing the words of In particular, many central words need to be collated with the input speech, but only auxiliary words that have a co-occurrence with the central word of the recognition candidate need be collated with the input speech. Processing speed has been improved.

In addition, it is only necessary to collate the attached words only in the section where the section of the central word is excluded from the input voice, and the section in which the attached word is collated is limited to one based on the order with the central word. Further, the processing load is reduced and the processing speed is improved. Further, the information on the order of the attached words with respect to the central word is set based on the information on the type of the attached words, and a plurality of attached words are not individually set.
Storage capacity has also been reduced.

The present invention is not limited to the above-described embodiment, but allows various modifications. For example, in the present embodiment, first, a central word having the largest matching score is selected from the input speech as one recognition candidate, and an auxiliary word having a co-occurrence relationship with the selected central word is matched with the input speech to obtain the largest score. As an example of selecting one, multiple core words whose scores exceed the reference value are first extracted as recognition candidates, and all attached words co-occurring with these central words are collated with the input speech Then, each score is calculated, and the combination having the largest total of the scores of the central word and the auxiliary word can be used as the recognition result.

In the present embodiment, a plurality of words having a co-occurrence relation are classified into a central word and an adjunct word, and the central word is first extracted from the input speech, and based on this result, An example of extracting an accessory word from an input voice has been described. However, instead of classifying multiple words that are co-occurring in this way as central words or adjuncts, all words are matched against the input speech at the same time, and the words of the co-occurrence combination that maximizes the total score are It is also possible to use a recognition result.

Further, in the present embodiment, it is assumed that maintenance and information registration are facilitated by separately forming various dictionaries 25 to 27 in the recognition phrase dictionary 22 for each application. It is also possible to form the recognition phrase dictionary 22 as a form in which 25 to 27 are combined into one.

In this embodiment, the speech recognition device 1
As an experimental device using a computer system,
The display of the word recognized from the input voice on the display 13 has been exemplified. However, each part of the speech recognition device 1 as described above is integrated with an ASIC (Application Specific Integrated
It is also possible to manufacture it as a circuit and incorporate it into various products and use it for voice control.

Further, in the present embodiment, it has been exemplified that each unit of the speech recognition apparatus 1 is realized by the operation of the CPU 2 according to a control program stored as software in the RAM 5 or the like. However, it is also possible to manufacture each of these units as unique hardware, and it is also possible to store a part of the unit as software in the RAM 5 or the like and manufacture a part of the unit as hardware. Further, the RAM 5 or the like in which predetermined software is stored and hardware of each unit can be manufactured as firmware, for example.

In the present embodiment, the speech recognition device 1
It is assumed that the software stored in the HDD 6 is copied to the RAM 5 at the time of startup, and the software stored in the RAM 5 is read by the CPU 2 as described above.
With such software stored in the HDD 6, the CP
It is also possible for U2 to use it or to write it in RAM5 in advance.

Further, as described above, software can be written in the FD 7 or CD-ROM 9 which is an information storage medium that can be handled alone, and the software can be installed in the RAM 5 or the like from the FD 7 or the like. It is also possible for the CPU 2 to appropriately read software written in the FD 7 or the like without executing the installation and execute data processing.

It is also possible to realize a control program for realizing each part of the voice recognition apparatus 1 by a combination of a plurality of softwares. In this case, the information storage medium as a single product has a minimum required size. It is only necessary to store the limited software. For example, the voice recognition device 1 on which an operating system is mounted has a CD
In a case where application software is provided by an information storage medium such as the ROM 9, software that realizes each unit of the voice recognition device 1 is realized by a combination of the application software and the operating system. The software can be omitted from the information storage medium of the application software.

In particular, when the speech recognition device 1 of the present invention is manufactured as a business device or the like in which words to be recognized are specified, the contents of the recognition phrase dictionary 22 may be fixedly written in the manufacturing process. However, as described above, the speech recognition device 1
It is preferable that the user can freely register the contents of the recognized phrase dictionary 22 when the application software is sold to general users.

When an information storage medium such as the CD-ROM 9 is manufactured as such a product, in addition to the control program 42 described above, a program for forming the recognition phrase dictionary 22 in the RAM 5 or the like in a predetermined format includes: A program for registering various kinds of information in the recognition phrase dictionary 22 is written in the information storage medium. In this case, these programs become the software of the recognition phrase dictionary 22 in the information storage medium, and the software of the recognition phrase dictionary 22 with the setting of various information is not written in the information storage medium.

Similarly, when manufacturing the speech recognition apparatus 1 as a completed product, various means 21 and 21 for recognizing words are used.
It is also possible to make the parts such as 3, 24, etc. fixedly, and let the user register the setting contents of the recognized phrase dictionary 22 as blank. Furthermore, such a speech recognition device 1
It is also possible to produce a recognition phrase dictionary 22 in which appropriate words are registered for each job as an information storage medium, as an optional component that can be exchangeably mounted on a computer.

The method for supplying the software written in the information storage medium to the computer as described above is not limited to loading the information storage medium directly into the computer. For example, it is also possible to write the above-mentioned software on an information storage medium of a host computer, connect the host computer to a terminal computer via a communication network, and supply the software to the terminal computer by data communication from the host computer.

In this case, it is possible for the terminal computer to execute stand-alone data processing in a state where the software has been downloaded to its own information storage medium, but it is possible to perform real-time data communication with the host computer without downloading the software. It is also possible to perform data processing. In this case, the entire system in which the host computer and the terminal computer are connected by the communication network is the voice recognition device 1 of the present invention.
Would be equivalent to

In the present embodiment, the word dictionary 25 stores the meaning of a word, and the word order dictionary 27 stores auxiliary words as meaning information.
As shown in (1), it is also possible to store information on the type of co-occurrence relation of an accessory word with respect to a central word in the co-occurrence dictionary 26 and the word order dictionary 27. In this case, even if a plurality of adjuncts correspond to one central word in a plurality of co-occurrence relations, the positions of a plurality of adjunct words with respect to the central word can be set by the information of the co-occurrence relation type. It is possible to easily recognize the auxiliary word of the starting relation with good accuracy.

For example, if “cup ramen” is extracted as a candidate for recognition of a central word from an input voice “order is curry flavored cup ramen,” an ancillary word co-occurring with this central word “cup ramen” As for "curry taste,
"Mini" is detected from the recognition phrase dictionary 22. However, here, it is specified that the adjunct of the type "taste" is located ahead of the central word, and the adjunct of the type "size" is the central word. Since it is specified that it is located further behind,
From the continuous input voice, the section of "Order is curry taste" is cut out and the adjunct word of "curry taste" is collated, and the voice section of "Issued" is cut out and the adjunct word of "mini" is collated Is done.

Also, as shown in FIG.
In the co-occurrence dictionary 26 of the second embodiment, an intervening word located in the middle of the combination of the central word and the auxiliary word is also stored.
3 can also recognize the central word, the intervening word, and the attached word from the input speech. In this case, since the central word, the intervening word and the adjunct word are recognized from one input voice, the reading of the adjunct word of a certain central word and the intervening word becomes the reading of the combination of another central word and the adjunct word. Even in the same case, these can be recognized separately from each other.

For example, if the input voice is “order is hand-baked rice cracker with green tea flavor”, both “nori and green tea flavor” as ancillary words are “a hand-roasted rice cracker” which is the central word. From the voice section of “green tea flavor” with the same score. However, if "no" is specified as an intervening word as described above, the adjunct "green tea flavor"
Can only recognize.

Further, as shown in FIG.
In the co-occurrence dictionary 26, information on time intervals is stored together with the combination of the central word and the adjunct word.
3 can also recognize the central word and the adjunct word from the input speech corresponding to the time interval. In this case, the section of the input voice for matching the adjunct can be limited according to the time interval, so that the adjunct can be recognized more quickly and the adjunct that is extremely deviated from the central term is not recognized. It is also possible to prevent the recognition of extra adjuncts.

For example, if “cup ramen” is extracted as a candidate for recognition of a central word from an input voice “Your order is two curry flavors of cup ramen,” a voice section within 20 frames from that voice section Only the adjuncts with the meaning “taste” are compared and “curry taste” is recognized.
Only the speech section within 100 frames is collated with the auxiliary word having the meaning of “quantity”, and “two boxes” is recognized.

As shown in FIG. 10, a combination of a central word and an adjunct word is stored in the co-occurrence dictionary 26 of the recognized phrase dictionary 22 as a hierarchical structure having a plurality of stages. It is also possible to recognize one central word and a plurality of attached words stepwise from the input speech in a hierarchical structure. In this case, when one central word and one auxiliary word are recognized from a certain input voice, other auxiliary words having this auxiliary word as the central word are also searched from the input voice, and such processing operations are sequentially repeated. Therefore, it is possible to sequentially recognize one central word and a plurality of attached words that have a co-occurrence relationship in a plurality of stages.

For example, when the input voice is "Please give me 350 cans of beer", "beer" is first extracted as the central word, and "can" is extracted as the corresponding auxiliary word.
Next, an auxiliary word "350" is extracted with this "can" as the central word, so that three words are finally recognized from one input voice.

Further, when the combination of the central word and the adjunct word has a hierarchical structure of a plurality of stages as described above, FIG.
As shown in (1), information on the order of combinations of one central word and a plurality of attached words is also stored in the recognized phrase dictionary 22, and the phrase recognition means 23 stores one central word and a plurality of attached words in the recognition phrase dictionary 22. It is also possible to recognize from the input voice corresponding to the order. In this case, since one central word and a plurality of attached words are recognized in order from the input speech, a plurality of attached words can be quickly and accurately recognized.

For example, when the input voice is “Please give me 350 cans of beer”, “beer” is first extracted as a central word, and the meaning is obtained from “350 cans of beer” which is a voice section ahead of this. "Can" as an adjunct word of "form" is extracted, and "350" which is an adjunct word of "size" is extracted from "350 no" which is a preceding voice section.

Further, when the combination of the central word and the adjunct word has a hierarchical structure of a plurality of stages as described above, FIG.
As shown in (1), information on the depth of the hierarchical structure of a combination of one central word and a plurality of attached words is also stored in the recognized phrase dictionary 22, and the phrase recognition means 23 stores one central word and a plurality of attached words. It is also possible to recognize words from input speech corresponding to depth. In this case, since the processing operation of searching for a plurality of attached words stepwise from one central word is executed to a predetermined depth, the plurality of attached words can be quickly recognized to a necessary stage.

For example, when the required hierarchical structure is set as “2” and the input voice is “Please give me 350 cans of beer”, the first word “beer” is extracted as the central word and then the first When "can" is extracted as an appendix,
The depth of the hierarchical structure is “1”. Then, when "350" is extracted as a second adjunct word with "can" as the central word, the depth of the hierarchical structure becomes "2", and the step-by-step speech recognition processing operation ends at this point. I do.

[0067]

According to the first aspect of the present invention, there is provided a speech recognition apparatus.
A voice input means for receiving a continuous input of a speech to be recognized, a recognition phrase dictionary in which a plurality of co-occurring phrases are stored in combination, and a plurality of co-occurrence combinations of continuous input voices And the phrase recognition means for recognizing the multiple words can be recognized from one continuous input voice based on a combination of co-occurrence relations. Can be recognized.

According to the second aspect of the present invention, the word recognizing means reads out the central word, which is one of a pair of words combined in co-occurrence relation, from the recognized word dictionary and extracts it from the input speech, and then extracts the central word. By reading an auxiliary word, which is the other word co-occurring with the central word, from the recognized phrase dictionary and extracting it from the input speech, it is possible to narrow down the auxiliary words to be matched with the input voice based on the extraction result of the central word. Since it is possible, the burden of the processing operation of recognizing an adjunct word from the input speech can be reduced and the speed can be improved, and the co-occurring central word and the adjunct word can be quickly and accurately recognized with good accuracy. .

According to the third aspect of the present invention, the phrase recognizing means extracts an adjunct word from a section excluding the section from which the central word of the input speech has been extracted, thereby collating the adjunct word based on the result of extracting the central word. Since the section of the input speech to be restricted can be limited, the load of the processing operation of recognizing the adjunct from the input speech can be reduced and the speed can be improved, and the co-occurrence relation between the central term and the adjunct can be improved. High-speed recognition with high accuracy.

According to the fourth aspect of the present invention, the recognition word dictionary is also provided with information on the order of the combination of the central word and the adjunct word, and the word / phrase recognition means associates the central word and the adjunct word in order. By recognizing from the input speech, the section of the input speech for collating the adjunct can be limited to the front or back from the extraction section of the central word, reducing the burden of the processing operation of recognizing the adjunct from the input speech As a result, the central word and the co-occurrence word which are co-occurring can be recognized at high speed with good accuracy.

According to the fifth aspect of the present invention, the recognition word dictionary also stores an intervening word located in the middle of the combination of the central word and the adjunct word. By recognizing from the input voice, for example, even if the reading of the adjunct of a certain central word and the intervening word is the same as the reading of the combination of another central word and the adjunct,
Since these can be recognized separately, the co-occurrence of the central word, intervening word and adjunct word can be recognized with good accuracy.

According to the sixth aspect of the present invention, the recognition phrase dictionary is also provided with information on a time interval for a combination of a central word and an adjunct word, and the phrase recognizing means converts the central word and the adjunct word into a time interval. By correspondingly recognizing from the input voice,
Since the section of the input speech for matching the adjunct can be limited to a range of a predetermined time interval from the central term extraction section, the load of the processing for recognizing the adjunct from the input speech is reduced and the speed is improved. Thus, the co-occurrence of the central word and the auxiliary word can be quickly recognized with good accuracy.

According to the seventh aspect of the present invention, the recognition phrase dictionary stores a combination of a central word and an adjunct word in a hierarchical structure having a plurality of stages, and the phrase recognizing means comprises one central word and a plurality of adjunct words. Are recognized stepwise from the input speech in accordance with the hierarchical structure, so that one central word and a plurality of adjuncts in a co-occurrence relation of a plurality of steps can be sequentially recognized step by step. Based on the phrase extraction results, you can narrow down the next phrase to match with the input voice,
The load on the processing operation of sequentially recognizing a plurality of words in a step-by-step manner from the input voice can be reduced and the speed can be improved, and many words can be recognized from one input voice at high speed with good accuracy.

According to the eighth aspect of the present invention, the recognition word dictionary is also provided with order information for a combination of one central word and a plurality of attached words. By recognizing adjunct words from the input speech in correspondence with each other, if the next phrase is to be collated with the input speech based on the extraction result of one phrase, this collation section is located ahead of the preceding phrase extraction section. Can be restricted to the rear, so that the load of the processing operation of recognizing a plurality of words in a step-by-step manner from the input voice can be reduced and the speed can be improved. It can be recognized at high speed with high accuracy.

According to the ninth aspect of the present invention, in the recognition word dictionary, information on the depth of the hierarchical structure is also given to a combination of one central word and a plurality of attached words.
A process of sequentially recognizing one central word and a plurality of attached words in a multi-stage co-occurrence relationship by recognizing one central word and a plurality of attached words from the input speech in accordance with the depth Since the operation can be performed to a predetermined depth,
Many words and phrases can be recognized from one input voice to a necessary stage.

According to a tenth aspect of the present invention, a plurality of words having a co-occurrence relationship are set in combination, a continuous input of a speech to be recognized is received, and a co-occurrence is obtained from the continuous input speech. By recognizing multiple words combined in a relationship, multiple words are recognized based on a combination of co-occurrence relationships from one continuous input voice, so multiple words can be recognized with good accuracy. Can be recognized at high speed.

In the speech recognition method according to the eleventh aspect, a co-occurrence central word and an adjunct word are set in combination, a continuous input of a speech to be recognized is received, and the continuous input speech is received. From the input speech, the central word and the auxiliary word combined in the co-occurrence relation are one continuous word. It is possible to narrow the attached words that are recognized from the typical input speech and to match with the input speech based on the extraction result of the central word, thereby reducing the burden of processing for recognizing the attached words from the input speech and improving the speed. Thus, the co-occurrence of the central word and the auxiliary word can be quickly recognized with good accuracy.

According to a twelfth aspect of the present invention, there is provided an information storage medium, comprising: a recognition phrase dictionary software in which a plurality of words having a co-occurrence relation are stored in combination; Since a program for recognizing words and phrases has been written, if the computer reads and operates the software of the information storage medium, the computer can share a plurality of words from one continuous input voice. Since recognition can be performed based on a combination of occurrence relationships, a plurality of phrases can be recognized at high speed with good accuracy.

According to a thirteenth aspect of the present invention, there is provided an information storage medium, comprising: a recognition word dictionary software in which a co-occurring central word and an adjunct word are stored in combination; By writing a program for extracting from the input speech and a program for extracting an auxiliary word having a co-occurrence relation with the extracted central word from the recognition phrase dictionary and extracting from the input speech, If the computer reads and operates the software of the information storage medium, the computer can recognize the central word and the adjunct words combined in a co-occurrence relationship from one continuous input voice, Since the auxiliary words to be matched with the input voice can be narrowed down based on the extraction result of the Can be improved, the center word and the accessory words can be recognized at high speed with good accuracy in the co-occurrence relation.

[Brief description of the drawings]

FIG. 1 is a schematic diagram showing a logical structure of a speech recognition device according to an embodiment of the present invention.

FIG. 2 is a block diagram showing a physical structure of the speech recognition device.

FIG. 3 is a perspective view showing an external appearance of the voice recognition device.

FIG. 4 is a schematic diagram showing a logical structure of software written in a RAM serving as an information storage medium.

5A and 5B are schematic diagrams showing storage contents of a recognized phrase dictionary, wherein FIG. 5A is a word dictionary, FIG. 5B is a co-occurrence dictionary, and FIG. 5C is a word order dictionary.

FIG. 6 is a flowchart illustrating a speech recognition method of the speech recognition device.

FIG. 7 is a schematic diagram showing storage contents of a co-occurrence dictionary and a word order dictionary of a recognized phrase dictionary according to a first modified example.

FIG. 8 is a schematic diagram showing storage contents of a co-occurrence dictionary of a recognized phrase dictionary according to a second modified example.

FIG. 9 is a schematic diagram showing storage contents of a co-occurrence dictionary of a recognized phrase dictionary of a third modified example.

FIG. 10 is a schematic diagram showing storage contents of a co-occurrence dictionary of a recognized phrase dictionary according to a fourth modified example.

FIG. 11 is a schematic diagram showing storage contents of a word dictionary and a word order dictionary of a recognized phrase dictionary of a fifth modified example.

FIG. 12 is a schematic diagram showing storage contents of a word order dictionary of a recognized phrase dictionary according to a sixth modified example.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Speech recognition apparatus 2 Computer 4-7, 9 Information storage medium 21 Speech input means 22 Recognized phrase dictionary 23 Phrase recognition means 41, 42 Software 42 Program

────────────────────────────────────────────────── ───

[Procedure amendment]

[Submission date] September 20, 1996

[Procedure amendment 1]

[Document name to be amended] Statement

[Correction target item name] Full text

[Correction method] Change

[Correction contents]

[Document Name] Statement

Patent application title: Speech recognition apparatus and method, information storage medium

[Claims]

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech recognition apparatus and method for recognizing speech, and an information storage medium on which software such as a program is written.

[0002]

2. Description of the Related Art At present, there is a demand for a voice recognition device for recognizing a voice uttered by a human, and various voice recognition methods are being considered. When a human utters only one word, which is a phrase, it is not difficult for the speech recognizer to recognize it, but in a natural human conversation, speech is continuous, and many words contain particle And so on. As a method for recognizing necessary words from continuous speech, word spotting has been devised. In this method, words of preset recognition candidates are extracted from continuous speech and recognized.

A speech recognition apparatus that performs such word spotting has a word dictionary in which readings are stored for each recognition candidate word. And the largest word whose score of the matching exceeds the reference value is recognized from the conversation voice. However, good results cannot be expected even if the word spotting is simply executed, so that it is now general to improve recognition accuracy due to linguistic restrictions.

Such linguistic constraints include those based on syntactic properties such as grammar, and those based on semantic constraints such as case patterns. The former is powerful as a constraint, but it is not suitable for spoken language that is difficult to regularize unlike written language, and its recognition control can be used only within phrases. The latter is suitable for the characteristics of Japanese and the spoken language in which the word order is free, and when used for the word spotting described above, the recognition accuracy can be improved satisfactorily.

[0005]

As described above, in word spotting, only necessary words and phrases can be recognized from continuous speech. In particular, recognition accuracy can be improved by using semantic constraints.

For example, in the speech recognition method of the speech recognition apparatus disclosed in Japanese Patent Application Laid-Open No. 6-102897, a phrase to be recognized is predicted using case relations, and narrowing is performed. However, this cannot deal with relationships other than case relationships, nor can it handle multiple relationships. In other words, the relationship between words included in the continuous speech is not limited to the case relationship, but it is also common that one continuous speech includes a plurality of relationships.

[0007] For example, if the conversation voice is "Please give a box of curry flavor of cup ramen", it is difficult to recognize words even by using case relations. In addition, when the conversation voice is “crying child is also silent”, there are a plurality of relationships of “crying / child” and “child / silent”.

[0008]

According to a first aspect of the present invention, there is provided a voice recognition apparatus for storing a combination of voice input means for receiving a continuous input of a voice to be recognized and a plurality of co-occurring words. And a phrase recognition unit for recognizing a plurality of phrases combined in a co-occurrence relationship from continuous input speech. Therefore, since a plurality of words having a co-occurrence relation are stored in combination in the recognition word dictionary, a plurality of words combined in a co-occurrence relation from the speech to be recognized continuously input to the voice input means are stored. Is recognized by the phrase recognition means. In other words, if predetermined words are combined and set in a co-occurrence relationship, a plurality of words in this co-occurrence relationship cannot be individually recognized from one continuous input voice, Is recognized based on

According to a second aspect of the present invention, in the speech recognition apparatus according to the first aspect, the word / phrase recognizing means reads out and inputs a central word which is one of a pair of words combined in a co-occurrence relation from the recognized word / phrase dictionary. After being extracted from the voice, a co-occurrence word, which is the other word having a co- occurrence relationship with the extracted central word, is read from the recognized phrase dictionary and extracted from the input voice. Therefore, when a plurality of words are recognized from one continuous input voice by the phrase recognition means, a central word which is one of the words is first extracted from the input voice, and then a co-occurrence relation with the central word is obtained. One other phrase, a co-occurrence word, is then extracted from the input speech. In other words, as in the case of conventional word spotting, a large number of central words are collated with the input speech, but co-occurring words are collated with the input speech after being narrowed down based on the co-occurrence relationship.

According to a third aspect of the present invention, in the speech recognition apparatus according to the second aspect, the word / phrase recognizing means extracts a co-occurrence word from a section excluding a section from which a central word of the input speech is extracted. Therefore, when a co-occurrence central word and a co-occurrence word are recognized from one continuous input voice by the phrase recognition means, the central word is first extracted from the entire input voice, and this central word is extracted. A co-occurrence word is extracted from a section other than the section performed. That is, although the center word will be collated with the input speech a number similar to the conventional word spotting, the co-occurrence <br/> words in speech interval that does not overlap the center Language narrowed down based on the co-occurrence relation It will be collated.

According to a fourth aspect of the present invention, in the speech recognition apparatus according to the second or third aspect, the recognition word dictionary is further provided with information on the order of the combination of the central word and the co-occurrence word. Recognizes a central word and a co-occurrence word from the input speech in order. Therefore, information on the order in which the central word and the co-occurring word occur in the continuous speech is also stored in the recognition phrase dictionary. When the central word and the co-occurring word occur in the input speech in a predetermined order, the phrase recognition unit recognizes the central word and the co-occurring word. Therefore, even when the central word and the co-occurring word are individually recognized from the input speech, they are not recognized unless the order is proper.

According to a fifth aspect of the present invention, in the speech recognition apparatus according to the fourth aspect, the recognition word dictionary also stores an auxiliary word located in the middle of the combination of the central word and the co-occurrence word. The word recognition means recognizes a central word, an adjunct word, and a co-occurrence word from the input speech. Therefore, the auxiliary word located in the middle of the combination of the central word and the co-occurrence word is also stored in the recognition phrase dictionary, and the central word, the auxiliary word, and the co-occurrence word are recognized from the input speech by the phrase recognition means. Even if the reading of a co-occurring word of a certain central word and an adjunct word is the same as the reading of a combination of another central word and a co-occurring word, these are recognized separately.

According to a sixth aspect of the present invention, in the speech recognition apparatus according to the second or third aspect, the recognition word dictionary is provided with time interval information for a combination of a central word and a co-occurrence word. The means recognizes the central word and the co-occurrence word from the input voice corresponding to the time interval. Therefore, the center language and co
Recognized by electromotive word and information time interval that occurs a continuous input speech is also stored in the recognition word dictionary, the center word and occurrence word and is compatible from the input speech to the time interval phrase recognition means Therefore, the central word and the co-occurring word are recognized only when they occur at an appropriate time interval in the input voice, and the time interval is not appropriate even when the central word and the co-occurring word are individually recognized from the input voice. Is not recognized.

According to a seventh aspect of the present invention, in the speech recognition apparatus according to the second or third aspect, the recognition word dictionary stores a combination of a central word and a co-occurrence word as a hierarchical structure having a plurality of stages. The recognizing means recognizes one central word and a plurality of co-occurring words stepwise from the input speech in accordance with the hierarchical structure. Therefore, a combination of a central word and a co-occurrence word is stored in the recognition phrase dictionary as a hierarchical structure having a plurality of stages, and one central word and a plurality of co-occurrence words are recognized from the input speech in a hierarchical structure corresponding to the hierarchical structure. Recognized step by step. In other words, one central word and one shared word from a certain input voice
When the electromotive word is recognized, the other centered words the occurrence word
Co-occurrence words are also retrieved from the input speech, and such processing operations are sequentially repeated, so that one central word and a plurality of co-occurrence words having a co- occurrence relationship in a plurality of stages are sequentially recognized.

According to the invention described in claim 8, in the speech recognition apparatus according to claim 7, the recognition word dictionary is further provided with order information for a combination of one central word and a plurality of co-occurring words. The recognizing means recognizes one central word and a plurality of co-occurring words in order from the input speech. Therefore, the information on the order of the combination of one central word and a plurality of co-occurring words is also stored in the recognition phrase dictionary, and the one central word and the plurality of co-occurring words correspond to the phrases in order from the input speech. Since it is recognized by the recognition means, even when one central word and a plurality of co-occurring words are individually recognized from the input voice, it is not recognized that their respective orders are not proper.

According to a ninth aspect of the present invention, in the speech recognition apparatus according to the seventh or eighth aspect, the recognition phrase dictionary also adds information of a hierarchical structure depth to a combination of one central word and a plurality of co-occurring words. The word recognition means recognizes one central word and a plurality of co-occurring words from the input speech in accordance with the depth. Therefore, the information on the depth of the hierarchical structure of the combination of one central word and multiple co-occurring words is also stored in the recognition phrase dictionary, and one central word and multiple co-occurring words correspond to the depth from the input speech. And is recognized by the phrase recognition means,
Stepwise recognition of a plurality of co-occurring words defining one central word is executed to a predetermined depth.

According to a tenth aspect of the present invention, a plurality of words having a co-occurrence relation are set in combination, a continuous input of a speech to be recognized is received, and co-occurrence is performed from the continuous input speech. Recognize multiple words combined in relationships. Therefore, if predetermined words are combined and set in a co-occurrence relationship, a plurality of words in this co-occurrence relationship cannot be individually recognized from one continuous input voice, and Is recognized based on

According to the speech recognition method of the present invention, a co-occurrence central word and a co-occurrence word are set in combination, and a continuous input of a speech to be recognized is received. A prepared central word is extracted from the voice, and co- occurring words having a co- occurrence relationship with the central word are extracted from the input voice. Therefore, co-located if one from a continuous input speech centered word cooccurrence and occurrence word is recognized, the center word is extracted from the first input speech, the co-occurrence relationship with the central word The spoken word is then extracted from the input speech. That is,
As with the central word, a large number of words are collated with the input speech, as in the conventional word spotting, but the co-occurring words are collated with the input speech after being narrowed down based on the co-occurrence relationship.

According to a twelfth aspect of the present invention, there is provided an information storage medium in which a plurality of words having a co-occurrence relation are combined and stored in an information storage medium in which computer-readable software is written in advance. ,
And a program for recognizing a plurality of phrases combined in a co-occurrence relationship from continuous input speech. Therefore, if the computer reads and operates the software of the information storage medium, the computer functions as a speech recognition device that recognizes words stored in the recognized word dictionary from continuous input speech. At this time, since a plurality of words having a co-occurrence relationship are stored in combination in the recognition word dictionary, a plurality of words combined in a co-occurrence relationship are recognized from continuous input speech. In other words, if predetermined words are combined and set in a co-occurrence relationship, a plurality of words in this co-occurrence relationship cannot be individually recognized from one continuous input voice, Is recognized based on

According to a thirteenth aspect of the present invention, there is provided an information storage medium in which a computer readable software is written in advance, and a recognition word stored in combination with a co-occurring central word and a co-occurrence word. Dictionary software, a program for reading a central word from the recognized phrase dictionary and extracting it from continuous input speech, and reading a co-occurring word having a co- occurrence relationship with the extracted central word from the recognized phrase dictionary And a program for extracting from the input voice. Therefore, if the computer reads and operates the software of the information storage medium, the computer functions as a speech recognition device that recognizes words stored in the recognized word dictionary from continuous input speech. At this time, when a plurality of words are recognized from one continuous input voice by the phrase recognition means, a central word which is one of the words is first extracted from the input voice, and then a co-occurrence relation with the central word is obtained. occurrence word is the other word in is extracted from the next input speech. In other words, as in the case of conventional word spotting, a large number of central words are collated with the input speech, but co-occurring words are collated with the input speech after being narrowed down based on the co-occurrence relationship.

[0021]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the present invention will be described below with reference to the drawings. First, as shown in FIGS. 2 and 3, the voice recognition device 1 of the present embodiment has a computer system as a data processing device as hardware. The speech recognition device 1 including the computer system includes a CPU (Centr
al Processing Unit) 2, and the CPU 2 is connected to a ROM (Read Only Memory) by a bus line 3.
4, RAM (Random Access Memory) 5, HD (Hard Disk)
... HDD (HD Drive) 6 with built-in
ppy Disk) 7, FDD (FD Drive) 8, CD (C
ompact Disk) -CD-ROM drive 10 loaded with ROM 9, keyboard 12 connected to mouse 11, display 13, microphone 1 as input device
4, communication I / F (Interface) 15, etc. are connected.

In the speech recognition apparatus 1, a program for causing the CPU 2 to execute various processing operations is set in advance, and software such as this program is stored in, for example, the FD 7 or the CD-ROM 9 which is an information storage medium. It has been written in advance. This software is installed in the HDD 6 as an information storage medium, and is copied to the RAM 5 as an information storage medium and read by the CPU 2 at startup.

As described above, various functions are realized as various means by the CPU 2 reading the software and executing various processing operations. As such various means, the voice recognition device 1 of the present embodiment includes a voice input unit 21, a recognized phrase dictionary 22, a phrase recognition unit 23, a result output unit 24, and the like, as shown in FIG. .
The recognition phrase dictionary 22 includes a word dictionary 25, a co-occurrence dictionary 2
6, word order dictionary 27, said word recognition means 23
Consists of a feature extracting unit 28, a candidate recognizing unit 29, a candidate reading unit 30, a phrase searching unit 31, a result determining unit 32, and the like.

Various means of such a speech recognition device 1 include:
If necessary, hardware such as the display 13 and the microphone 14 is used, and the main component is realized by the operation of the CPU 2 corresponding to software written in the RAM 5 or the like. For this reason, as shown in FIG.
A dictionary storage unit 41 which is software of the recognition phrase dictionary 22, a control program 42 for recognizing a plurality of phrases combined in a co-occurrence relationship from continuous input speech,
Is written.

More specifically, the dictionary storage unit 41 comprises a word dictionary storage unit 43, a co-occurrence dictionary storage unit 44, and a word order dictionary storage unit 45. These storage units 43 to 45 store the dictionaries 25 to 27 Various information is stored. The control program 42 is formed in a module structure and includes a feature extraction module 46, a word matching module 47, a recognition control module 48, and the like.

The feature extraction module 46 comprises a program for analyzing a digital signal of a continuous input speech for each frame and extracting a feature, and the word matching module 47 recognizes the feature of the input speech. The program comprises a program for collating feature amounts which are readings of candidate words and calculating a cumulative similarity as a score. The recognition control module 48 reads various types of information from the dictionaries 25 to 27 and transmits the information to the word matching module 47. From the program for determining the word of the recognition result based on the score output by the word matching module 47, Become.

The recognition phrase dictionary 22 is stored as a data file in the dictionary storage unit 41 of the RAM 5 as described above, and as shown in FIG. Are registered for each of a plurality of words. In the co-occurrence dictionary 26, a plurality of words having a co-occurrence relationship are stored in the co-occurrence dictionary.
Stored in combination with co-occurrence words. The word order dictionary 27 stores information on the order of the central word and the co-occurring words because the central word and the information on the meaning of the co-occurring words are set in order.

Although the various information in the dictionaries 25 to 27 are expressed as Japanese characters for the sake of simplicity of description, words and the like are composed of identification codes in actual software, and the reading is performed in words. It is set as a feature value of the voice. The feature amount of the reading voice is set, for example, as a state transition model in phoneme units and a phoneme network in word units, and information on the average feature amount and duration is set in each state.

The voice input means 21 receives continuous voice input from the microphone 14 or the like, and converts the input voice into a digital electric signal by A / D (Analog / Digilog).
tal) The phrase recognizing means 23 includes the CPU 2
Reads the control program 42 stored in the RAM 5 and executes a corresponding processing operation, thereby recognizing a co-occurrence center word and a co-occurrence word from a continuous input voice digital signal.

More specifically, the feature extracting means 28 of the phrase recognizing means 23 reads the feature extracting module 46 of the control program 42 and executes the corresponding arithmetic processing to execute the continuous input. The audio digital signal is analyzed for each frame, which is a unit time, and a feature amount is extracted by calculating, for example, an LPC (Linear Predictive Coding) mel-cepstral.

The candidate recognizing means 29 reads out all the central words from the co-occurrence dictionary 26 of the recognized phrase dictionary 22 by the arithmetic processing of the CPU 2 corresponding to the recognition control module 48, and then reads each of them. Read from the word dictionary 25. Further, the word matching module 4
7, the reading of all the central words is compared with the feature amount of the continuous input voice, the similarity of each of them is calculated for each frame, and sequentially accumulated,
One central word whose score as the accumulated similarity exceeds the reference value at the maximum is extracted as a recognition candidate.

The candidate reading means 30 uses the CPU 2 corresponding to the recognition control module 48 to perform all the co-occurrence words having a co- occurrence relationship with the central word of the recognition candidate as recognition candidates by the CPU 2 corresponding to the recognition control module 48. From the co-occurrence dictionary 26, the word / phrase search means 31 performs a calculation process of the CPU 2 corresponding to the word matching module 47,
All the co-occurring words of the recognition candidates are collated with the input speech.

In this processing operation, similarly to the case of the central word, the reading of the co-occurrence word of the recognition candidate is read from the word dictionary 25 of the recognized phrase dictionary 22, and the score of the collation between the word and the input speech is maximum. One co-occurrence word exceeding the reference value is extracted. However, when the reading of the co-occurrence word is compared with the feature amount of the input voice in this way, the CPU 2 corresponding to the recognition control module 48 performs the co- processing with the central word from the word order dictionary 27 of the recognition word dictionary 22. information order with raised words are read out, occurrence word reading is matched only in a section other than the central language of the input speech corresponding to the order is extracted section.

When the co-occurrence word of the recognition candidate is searched from the input voice as described above, the result determination means 32 calculates the central word of the recognition candidate by the arithmetic processing of the CPU 2 corresponding to the recognition control module 48. A co-occurrence word is determined as a recognition result. The result output unit 24 outputs the central word and the co-occurrence word of the recognition result determined as described above, for example, by character display on the display 13.

In such a configuration, the speech recognition apparatus 1 according to the present embodiment, when a speech of a conversation in which a large number of words are continuous is input to the microphone 14, converts the continuous input speech into the recognized phrase dictionary 22. The stored words are recognized, and the recognition results are displayed on the display 13.
Such a voice recognition method of the voice recognition device 1 will be sequentially described in detail below with reference to FIG.

First, the whole area of the voice continuously input to the microphone 14 is A / D converted into a digital signal, and the characteristic amount is extracted for each frame. Next, the readings of all the central words are read from the recognized phrase dictionary 22, and all the readings are collated with the feature values of the whole input voice. When the cumulative similarity of collation is calculated as a score for all the central words in this way, only one central word having the largest score exceeding the reference value is selected as a recognition candidate.

When the central word of the recognition candidate is selected as described above, all co-occurring words having a co- occurrence relation with the selected central word are read out from the recognized phrase dictionary 22 as recognition candidates, and are the same as in the case of the central word described above. Then, the reading of all the co-occurring words and the input voice are collated, and one co-occurring word having the maximum score exceeding the reference value is selected. At this time, the word order dictionary 2 of the recognition phrase dictionary 22
7, information on the order of the central word and the co-occurring word is read, and the reading of the co-occurring word is collated only in a section other than the section in which the central word of the input voice is extracted in accordance with the order.

When a central word and a co-occurrence word having a co-occurrence relation are extracted from one continuous input voice, they are determined as a recognition result and output by character display on the display 13. If the scores of the collation of all the central words do not exceed the reference value at first, no recognition result is found, and the co-occurrence word is stored in the recognized phrase dictionary 22 in a state where the recognition candidate of the central word is detected. If not, or if the scores of collation of all co-occurring words do not exceed the reference value, the recognition result is only the central word.

The above-described series of processing operations will be described in detail with reference to FIG. 5. If one continuous input voice is "order is curry taste of cup ramen""Cup ramen, hand-baked rice cracker" is collated, and "cup ramen" with a high score is extracted as a candidate for recognition of the central word. "Curry flavor, soy sauce flavor" is a co-occurring word that co- occurs with this central word "cup ramen"
Is detected from the recognition phrase dictionary 22, and the recognition phrase dictionary 2
2 specifies that a co-occurrence word whose meaning is “taste” is located behind the central word “cup ramen”, so that a continuous input voice and a subsequent section “coup ramen” It's curry flavor. " Since the co-occurrence word “curry taste, soy sauce taste” is collated only for this voice section, the co-occurrence word “curry taste” is recognized.

In the speech recognition method of the speech recognition apparatus 1 according to the present embodiment, as described above, the central word and the co-occurrence word combined in a co-occurrence relation are recognized from one continuous input speech. The accuracy is better than when these words are individually recognized. In particular, the center word, it is necessary to match the number on the input speech, occurrence word since only it is sufficient to match the input speech that the center word co-occurrence relationships recognition candidate, the processing load is reduced Processing speed has been improved.

[0041] Moreover, occurrence word is limited to one based on it is sufficient only matched to the section which eliminated the central word section, the order of the section is also central word for matching the occurrence word from the input speech Therefore, the processing load is further reduced, and the processing speed is improved. Further, the information on the order of co-occurring words with respect to the central word is set by the information on the type of co-occurring words, and a plurality of co-occurring words are not individually set.
Storage capacity has also been reduced.

The present invention is not limited to the above-described embodiment, but allows various modifications. For example, in the present embodiment, first, a central word having the largest matching score is selected from the input speech as one recognition candidate, and a co- occurring word having a co- occurrence relationship with the selected central word is compared with the input speech to obtain the maximum score. Was selected as an example. First, a plurality of central words whose scores exceeded the reference value were extracted as recognition candidates, and all co-occurring words having a co- occurrence relationship with these central words were input speech. , The respective scores are calculated, and the combination in which the sum of the scores of the central word and the co-occurrence word is the largest can be used as the recognition result.

In the present embodiment, a plurality of co-occurring terms are classified into a central word and a co-occurring word, and the central word is first extracted from the input speech, and then based on the result. Extracting co-occurring words from input speech has been exemplified. However, the center word and co multiple words in this manner cooccurrence
Not classified as causing word, is collated simultaneously input speech all the words, the score of the sum is also possible, as a recognition result a word or phrase in a combination of co-occurrence relation with a maximum.

Further, in the present embodiment, it is assumed that maintenance and information registration are facilitated by separately forming various dictionaries 25 to 27 in the recognition phrase dictionary 22 for each application. It is also possible to form the recognition phrase dictionary 22 as a form in which 25 to 27 are combined into one.

In this embodiment, the speech recognition device 1
As an experimental device using a computer system,
The display of the word recognized from the input voice on the display 13 has been exemplified. However, each part of the speech recognition device 1 as described above is integrated with an ASIC (Application Specific Integrated
It is also possible to manufacture it as a circuit and incorporate it into various products and use it for voice control.

Further, in the present embodiment, it has been exemplified that each unit of the speech recognition apparatus 1 is realized by the operation of the CPU 2 according to a control program stored as software in the RAM 5 or the like. However, it is also possible to manufacture each of these units as unique hardware, and it is also possible to store a part of the unit as software in the RAM 5 or the like and manufacture a part of the unit as hardware. Further, the RAM 5 or the like in which predetermined software is stored and hardware of each unit can be manufactured as firmware, for example.

In the present embodiment, the speech recognition device 1
It is assumed that the software stored in the HDD 6 is copied to the RAM 5 at the time of startup, and the software stored in the RAM 5 is read by the CPU 2 as described above.
With such software stored in the HDD 6, the CP
It is also possible for U2 to use it or to write it in RAM5 in advance.

Further, as described above, software can be written in the FD 7 or CD-ROM 9 which is an information storage medium that can be handled alone, and the software can be installed in the RAM 5 or the like from the FD 7 or the like. It is also possible for the CPU 2 to appropriately read software written in the FD 7 or the like without executing the installation and execute data processing.

It is also possible to realize a control program for realizing each part of the voice recognition apparatus 1 by a combination of a plurality of softwares. In this case, the information storage medium as a single product has a minimum required size. It is only necessary to store the limited software. For example, the voice recognition device 1 on which an operating system is mounted has a CD
In a case where application software is provided by an information storage medium such as the ROM 9, software that realizes each unit of the voice recognition device 1 is realized by a combination of the application software and the operating system. The software can be omitted from the information storage medium of the application software.

In particular, when the speech recognition device 1 of the present invention is manufactured as a business device or the like in which words to be recognized are specified, the contents of the recognition phrase dictionary 22 may be fixedly written in the manufacturing process. However, as described above, the speech recognition device 1
It is preferable that the user can freely register the contents of the recognized phrase dictionary 22 when the application software is sold to general users.

When an information storage medium such as the CD-ROM 9 is manufactured as such a product, in addition to the control program 42 described above, a program for forming the recognition phrase dictionary 22 in the RAM 5 or the like in a predetermined format includes: A program for registering various kinds of information in the recognition phrase dictionary 22 is written in the information storage medium. In this case, these programs become the software of the recognition phrase dictionary 22 in the information storage medium, and the software of the recognition phrase dictionary 22 with the setting of various information is not written in the information storage medium.

Similarly, when manufacturing the speech recognition apparatus 1 as a completed product, various means 21 and 21 for recognizing words are used.
It is also possible to make the parts such as 3, 24, etc. fixedly, and let the user register the setting contents of the recognized phrase dictionary 22 as blank. Furthermore, such a speech recognition device 1
It is also possible to produce a recognition phrase dictionary 22 in which appropriate words are registered for each job as an information storage medium, as an optional component that can be exchangeably mounted on a computer.

The method for supplying the software written in the information storage medium to the computer as described above is not limited to loading the information storage medium directly into the computer. For example, it is also possible to write the above-mentioned software on an information storage medium of a host computer, connect the host computer to a terminal computer via a communication network, and supply the software to the terminal computer by data communication from the host computer.

In this case, it is possible for the terminal computer to execute stand-alone data processing in a state where the software has been downloaded to its own information storage medium, but it is possible to perform real-time data communication with the host computer without downloading the software. It is also possible to perform data processing. In this case, the entire system in which the host computer and the terminal computer are connected by the communication network is the voice recognition device 1 of the present invention.
Would be equivalent to

In this embodiment, the word dictionary 25 stores the meanings of words, and the word order dictionary 27 stores co-occurrence words as meaning information.
As shown in (1), it is also possible to store information on the type of co-occurrence relation of a co- occurrence word with the central word in the co- occurrence dictionary 26 and the word order dictionary 27. In this case, even if a plurality of co-occurring words correspond to one central word in a plurality of co-occurring relations, the positions of the plurality of co-occurring words with respect to the central word can be set by the information of the type of the co-occurring relation. the co-occurrence relationship cooccurrence words can be easily recognized with good accuracy.

For example, if “cup ramen” is extracted as a candidate for recognition of a central word from an input voice “Your order is curry flavored cup ramen,” a co- occurrence co- occurring with this central word “cup ramen” The word is "curry taste,
Mini "is detected from the recognition word dictionary 22. However, where the kind of" taste "co-occurrence word is defined to be located forward of the central words, the type is" occurrence word size "is Since it is specified that it is located after the central word,
From the continuous input voice, the section "Order is curry taste" is cut out and the co-occurrence word "curry taste" is collated, and the voice section of "is" is cut out and co-occurrence word "mini" Are collated.

Also, as shown in FIG.
In the co-occurrence dictionary 26 of FIG. 2, an auxiliary word located in the middle of the combination of the central word and the co-occurrence word is also stored.
3 can also recognize the central word, the adjunct word, and the co-occurrence word from the input speech. In this case, since the central word, the adjunct word and the co-occurrence word are recognized from one input voice, the reading of the co-occurrence word of a certain central word and the adjunct word is not co- occurred with other central words . even if the same reading in combination with electromotive words, it is possible to recognize them each separately.

For example, if the input voice is “order is hand-baked rice cracker green tea flavor”, both of the co-occurrence words “Nori and green tea flavor” are used for the central word “hand-baked rice cracker”. Recognition is performed with the same score from the voice section of “no green tea flavor”. However, if it is stipulated "in" as an accessory words, as described above, "green tea flavor" as the co-occurrence word
Can only recognize.

Further, as shown in FIG.
In the co-occurrence dictionary 26, information on the time interval is stored together with the combination of the central word and the co-occurrence word.
3 can also recognize the central word and the co-occurrence word from the input voice corresponding to the time interval. In this case, since the section of the input speech to match the occurrence word can be limited to correspond to the time intervals, it can recognize the occurrence word faster, occurrence word was extremely away from the center word is not recognized Therefore, it is possible to prevent recognition of inappropriate co-occurrence words.

For example, if “cup ramen” is extracted as a candidate for recognition of a central word from an input voice “Your order is two curry flavors of cup ramen,” a voice section within 20 frames from that voice section Only the co-occurrence word with the meaning “taste” is compared and “curry taste” is recognized,
Only co-occurrence words having a meaning of “quantity” are collated only in a voice section within 100 frames, and “two boxes” are recognized.

As shown in FIG. 10, a combination of a central word and a co-occurrence word is stored in the co-occurrence dictionary 26 of the recognized word dictionary 22 as a hierarchical structure having a plurality of stages. It is also possible to recognize one central word and a plurality of co-occurring words stepwise from the input speech in a hierarchical structure. In this case, one central word and one
When occurrence word and is recognized, the other co-occurrence word centered words the occurrence word is also retrieved from the input speech, such a process operation is repeated sequentially, in co-occurrence relationship a plurality of stages One central word and a plurality of co-occurring words can be sequentially recognized.

For example, when the input voice is “Please give me 350 cans of beer”, “beer” is first extracted as the central word, and “can” is extracted as the corresponding co-occurrence word.
Next, since the co-occurrence word "350" is extracted with this "can" as the central word, three words are finally recognized from one input voice.

Further, when the combination of the central word and the co-occurring word has a hierarchical structure of a plurality of stages as described above, FIG.
As shown in (1), information on the order of combinations of one central word and a plurality of co-occurring words is also stored in the recognized phrase dictionary 22, and the phrase recognizing means 23 stores one central word and a plurality of co-occurring words. Can be recognized from the input voice in order. In this case, since one central word and a plurality of co-occurring words are recognized in order from the input speech, a plurality of co-occurring words can be recognized with good accuracy and at high speed.

For example, when the input voice is “Please give me 350 cans of beer”, “beer” is first extracted as a central word, and the meaning is obtained from “350 cans of beer” which is a voice section ahead of this. "Can" which is a co-occurrence word of "form" is extracted, and "350" which is a co-occurrence word of "size" is extracted from "350 of" which is a preceding voice section.

Further, when the combination of the central word and the co-occurring word has a hierarchical structure of a plurality of stages as described above, FIG.
As shown in (1), information on the depth of the hierarchical structure of a combination of one central word and a plurality of co-occurring words is also stored in the recognized phrase dictionary 22, and the phrase recognition means 23 It is also possible to recognize a co-occurrence word from an input voice in accordance with the depth. In this case, since a processing operation of searching for a plurality of co-occurring words stepwise from one central word is executed to a predetermined depth, it is possible to quickly recognize a plurality of co-occurring words to a necessary stage. Can be.

For example, when the required hierarchical structure is set as “2” and the input voice is “Please give me 350 cans of beer”, the first word “beer” is extracted as the central word and then the first When “can” is extracted as a co-occurrence word,
The depth of the hierarchical structure is “1”. Then, when "350" is extracted as the second co-occurrence word with this "can" as the central word, the depth of the hierarchical structure becomes "2". finish.

[0067]

According to the first aspect of the present invention, there is provided a speech recognition apparatus.
A voice input means for receiving a continuous input of a speech to be recognized, a recognition phrase dictionary in which a plurality of co-occurring phrases are stored in combination, and a plurality of co-occurrence combinations of continuous input voices And the phrase recognition means for recognizing the multiple words can be recognized from one continuous input voice based on a combination of co-occurrence relations. Can be recognized.

According to the second aspect of the present invention, the word recognizing means reads out the central word, which is one of a pair of words combined in co-occurrence relation, from the recognized word dictionary and extracts it from the input speech, and then extracts the central word. central word and by extraction reads from the input speech from the other recognition word dictionary the occurrence word is a word in the co-occurrence relation, Filter occurrence word to be collated with the input speech based on the extracted result of the central word Can reduce the burden of processing operations for recognizing co-occurring words from input speech and improve the speed, and can quickly and quickly recognize co-occurring central words and co-occurring words with good accuracy. can do.

[0069] In the present invention of claim 3, wherein, the phrase recognition means, by extracting the occurrence word from eliminating the section a section of extracting center language of the input speech, occurrence word based on the extraction result of the central word it is possible to limit the period of the input speech to be collated, the burden of recognizing processing operation occurrence word from the input speech speed can be improved by co-occurrence a central word in the co-occurrence relation Words can be quickly recognized with good accuracy.

[0070] In the present invention of claim 4, wherein the recognition word dictionary, also the order of the information to the combination of the central word and occurrence word are applied, the phrase recognition means, in turn a central word and occurrence word by recognizing the input speech corresponding to, it is possible to limit the forward or backward from the extraction section of the central words the section of the input speech to match the co-occurrence word, the co-occurrence <br/> word from the input speech The speed of the recognition operation can be reduced by reducing the load of the recognition operation, and the co-occurrence center word and the co-occurrence word can be quickly and accurately recognized with good accuracy.

According to the fifth aspect of the present invention, the recognition phrase dictionary is a combination of a central word and a co-occurrence word .
Generic words are also stored, and the phrase recognition means recognizes a central word, an adjunct word, and a co-occurrence word from the input voice, for example, a co-occurrence word of a certain central word and an adjunct word. Is the same as the reading of the combination of other central words and co-occurring words,
Since these can be recognized separately, the central word, the auxiliary word, and the co-occurrence word which are in a co-occurrence relationship can be recognized with good accuracy.

[0072] In the present invention of claim 6, wherein the recognition word dictionary, information of the time interval to a combination of the central word and occurrence word also been granted, the phrase recognition means, the center word and occurrence word and the time By recognizing from the input voice corresponding to the interval,
Since the section of the input speech for collating co- occurred words can be limited to a range of a predetermined time interval from the extraction section of the central word, the load of the processing operation of recognizing co- occurred words from the input speech is reduced and the speed is reduced. Thus, the co-occurrence word and the co-occurrence word can be quickly recognized with good accuracy.

[0073] In the present invention of claim 7, wherein the recognition word dictionary, a combination of a central word and occurrence word is stored as a hierarchical structure of a plurality of stages, the phrase recognition means, one of the central words and more co by stepwise recognized from the input speech corresponding to the hierarchical structure and cause words, it is possible to sequentially recognize stepwise and one central word and a plurality of co-occurrence word in the co-occurrence relationship a plurality of stages , You can narrow down the next phrase to match against the input voice based on the result of extracting one phrase,
The load on the processing operation of sequentially recognizing a plurality of words in a step-by-step manner from the input voice can be reduced and the speed can be improved, and many words can be recognized from one input voice at high speed with good accuracy.

According to the eighth aspect of the present invention, the recognition word dictionary is also provided with order information for a combination of one central word and a plurality of co-occurring words. When the next phrase is collated with the input voice based on the result of extracting one phrase by recognizing the co-occurrence words of the input speech in order, Since it is possible to restrict the forward or backward direction, it is possible to reduce the burden of the processing operation of sequentially recognizing a plurality of phrases from the input voice and improve the speed, and it is possible to improve a plurality of phrases from one input voice. High-speed recognition can be performed with good accuracy.

According to the ninth aspect of the present invention, in the recognition word dictionary, information on the depth of the hierarchical structure is also given to a combination of one central word and a plurality of co-occurrence words.
Recognizing one central word and multiple co-occurring words from the input speech corresponding to depth, and sequentially recognizing one central word and multiple co-occurring words in a multi-stage co-occurrence relationship Can be performed to a predetermined depth,
Many words and phrases can be recognized from one input voice to a necessary stage.

According to a tenth aspect of the present invention, a plurality of words having a co-occurrence relationship are set in combination, a continuous input of a speech to be recognized is received, and a co-occurrence is obtained from the continuous input speech. By recognizing multiple words combined in a relationship, multiple words are recognized based on a combination of co-occurrence relationships from one continuous input voice, so multiple words can be recognized with good accuracy. Can be recognized at high speed.

In the speech recognition method according to the present invention, a central word and a co-occurrence word having a co-occurrence relation are set in combination, and a continuous input of a speech to be recognized is received. By extracting a central word prepared from the voice and extracting a co-occurrence word having a co- occurrence relation with the central word from the input voice, the central word and the co-occurrence word combined in the co-occurrence relation are extracted. Since co-occurring words that are recognized from one continuous input voice and are matched with the input speech based on the extraction result of the central word can be narrowed down, the burden of processing operations for recognizing co-occurring words from the input voice can be reduced. Speed can be improved, and a central word and a co-occurring word in a co-occurring relationship can be quickly recognized with good accuracy.

According to a twelfth aspect of the present invention, there is provided an information storage medium, comprising: a recognition phrase dictionary software in which a plurality of words having a co-occurrence relation are stored in combination; Since a program for recognizing words and phrases has been written, if the computer reads and operates the software of the information storage medium, the computer can share a plurality of words from one continuous input voice. Since recognition can be performed based on a combination of occurrence relationships, a plurality of phrases can be recognized at high speed with good accuracy.

According to a thirteenth aspect of the present invention, there is provided an information storage medium, comprising: a recognition word dictionary software for storing a co-occurring central word and a co-occurring word in combination; a program for extracting the Do input speech, co-located on the extracted central word co-occurrence relationships
A program for extracting from the input speech by reading an electromotive Language recognition word dictionary, by is written, be operated is read by the software of the information storage medium into the computer, the computer cooccurrence Since the central word and co-occurrence word combined in the relation can be recognized from one continuous input voice, and the co-occurrence word to be collated with the input voice based on the extraction result of the central word can be narrowed down, The load on the processing operation of recognizing a co-occurring word from the input speech can be reduced and the speed can be improved, and the central word and the co-occurring word that have a co-occurring relationship can be quickly and accurately recognized.

[Brief description of the drawings]

FIG. 1 is a schematic diagram showing a logical structure of a speech recognition device according to an embodiment of the present invention.

FIG. 2 is a block diagram showing a physical structure of the speech recognition device.

FIG. 3 is a perspective view showing an external appearance of the voice recognition device.

FIG. 4 is a schematic diagram showing a logical structure of software written in a RAM serving as an information storage medium.

5A and 5B are schematic diagrams showing storage contents of a recognized phrase dictionary, wherein FIG. 5A is a word dictionary, FIG. 5B is a co-occurrence dictionary, and FIG. 5C is a word order dictionary.

FIG. 6 is a flowchart illustrating a speech recognition method of the speech recognition device.

FIG. 7 is a schematic diagram showing storage contents of a co-occurrence dictionary and a word order dictionary of a recognized phrase dictionary according to a first modified example.

FIG. 8 is a schematic diagram showing storage contents of a co-occurrence dictionary of a recognized phrase dictionary according to a second modified example.

FIG. 9 is a schematic diagram showing storage contents of a co-occurrence dictionary of a recognized phrase dictionary of a third modified example.

FIG. 10 is a schematic diagram showing storage contents of a co-occurrence dictionary of a recognized phrase dictionary according to a fourth modified example.

FIG. 11 is a schematic diagram showing storage contents of a word dictionary and a word order dictionary of a recognized phrase dictionary of a fifth modified example.

FIG. 12 is a schematic diagram showing storage contents of a word order dictionary of a recognized phrase dictionary according to a sixth modified example.

[Description of Signs] 1 Voice Recognition Device 2 Computer 4-7, 9 Information Storage Medium 21 Voice Input Means 22 Recognized Word Dictionary 23 Word Recognition Means 41, 42 Software 42 Program

[Procedure amendment 2]

[Document name to be amended] Drawing

[Correction target item name] Fig. 5

[Correction method] Change

[Correction contents]

FIG. 5

[Procedure amendment 3]

[Document name to be amended] Drawing

[Correction target item name] Fig. 6

[Correction method] Change

[Correction contents]

FIG. 6

[Procedure amendment 4]

[Document name to be amended] Drawing

[Correction target item name] Fig. 7

[Correction method] Change

[Correction contents]

FIG. 7

[Procedure amendment 5]

[Document name to be amended] Drawing

[Correction target item name] Fig. 8

[Correction method] Change

[Correction contents]

FIG. 8

[Procedure amendment 6]

[Document name to be amended] Drawing

[Correction target item name] Fig. 9

[Correction method] Change

[Correction contents]

FIG. 9

[Procedure amendment 7]

[Document name to be amended] Drawing

[Correction target item name] FIG.

[Correction method] Change

[Correction contents]

FIG. 10

Claims (13)

[Claims] 1. A speech input means for receiving a continuous input of a speech to be recognized, a recognition phrase dictionary in which a plurality of co-occurring phrases are stored in combination, and a co-occurrence relationship from the continuous input speech And a phrase recognizing means for recognizing a plurality of phrases combined with each other. 2. The phrase recognition means reads a central word, which is one of a pair of phrases combined in co-occurrence relation, from a recognized phrase dictionary and extracts it from an input voice, and then co-occurs with the extracted central word. 2. The speech recognition apparatus according to claim 1, wherein an attached word, which is the other related phrase, is read from the recognized phrase dictionary and extracted from the input speech. 3. The speech recognition apparatus according to claim 2, wherein the word recognition means extracts an auxiliary word from a section excluding a section from which a central word of the input speech has been extracted. 4. The recognition phrase dictionary is also provided with order information for a combination of a central word and an adjunct word, and the phrase recognizing means recognizes the central word and the adjunct word from the input speech in correspondence with the order. The speech recognition device according to claim 2 or 3, wherein: 5. The recognition phrase dictionary also stores intervening words located in the middle of a combination of a central word and an adjunct word, and the phrase recognizing means recognizes the central word, the intervening word and the adjunct word from the input speech. The voice recognition device according to claim 4, wherein the voice recognition is performed. 6. The recognition phrase dictionary is also provided with information on a time interval for a combination of a central word and an adjunct word, and the phrase recognizing means converts the central word and the adjunct word from an input voice in accordance with the time interval. 4. The voice recognition device according to claim 2, wherein the voice recognition is performed. 7. The recognition phrase dictionary stores a combination of a central word and an adjunct word in a hierarchical structure having a plurality of stages.
4. The speech recognition apparatus according to claim 2, wherein the phrase recognition means recognizes one central word and a plurality of attached words stepwise from the input speech corresponding to a hierarchical structure. 8. The recognition phrase dictionary is also provided with order information for a combination of one central word and a plurality of attached words,
8. The speech recognition apparatus according to claim 7, wherein the phrase recognition unit recognizes one central word and a plurality of attached words in order from the input speech. 9. The recognition phrase dictionary is provided with information on the depth of a hierarchical structure for a combination of one central word and a plurality of attached words, and the phrase recognition means includes one central word and a plurality of attached words. 9. The speech recognition apparatus according to claim 7, wherein the speech recognition unit recognizes the input speech from the input speech in accordance with the depth. 10. A plurality of words having a co-occurrence relationship are set in combination, a continuous input of a speech to be recognized is received, and a plurality of words combined in a co-occurrence relationship from the continuous input speech. A voice recognition method characterized by recognizing a voice. 11. A co-occurrence central word and an adjunct word are set in combination, a continuous input of a speech to be recognized is received, and a prepared central word is extracted from the continuous input voice. And a supplementary word having a co-occurrence relationship with the central word is extracted from the input speech. 12. An information storage medium in which software readable by a computer is written in advance, and a software for a recognition phrase dictionary in which a plurality of words having a co-occurrence relation are combined and stored, and a continuous input voice. An information storage medium, in which a program for recognizing a plurality of phrases combined in a starting relationship is written. 13. An information storage medium in which software that is readable by a computer is pre-written, wherein a software for a recognition phrase dictionary in which co-occurring central words and adjunct words are stored in combination, A program for reading from the recognition phrase dictionary and extracting it from the continuous input speech, and a program for reading from the recognition phrase dictionary an auxiliary word having a co-occurrence relationship with the extracted central word from the input speech. , An information storage medium characterized by being written therein.