JP2010197709A - Voice recognition response method, voice recognition response system and program therefore - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a voice recognition response method and a system etc., promptly and effectively identifying homonyms and words with similar reading. <P>SOLUTION: A voice recognition device 13 recognizes an input voice data, and searches and extracts a corresponding identical or similar word data and its modifier for identification from a storage device. On the basis of this, a voice creation output device 16 creates a fixed form sentence for confirmation of a text format, and outputs it to outside via a voice reproduction output device 18. In response to this, when response of contents denial comes from a speaker, the voice recognition device 13 and a modifier search device 14 extract other identical or similar word data and its modifier again from the storage device, and based on this, the fixed form sentence for confirmation is created and output to outside. When response of the contents affirmation comes from outside on the other hand, the fixed sentence for confirmation which displays that the input voice is recognized is created by the voice creation conversion device 16, and is output to outside via the voice reproduction output device 18. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a speech recognition response method, a speech recognition response system, and the like, and more particularly, to a speech recognition response method, a speech recognition response system, and a program thereof that improve the recognition rate of short utterance words that are easily misrecognized. .

  2. Description of the Related Art Conventionally, known automated voice response systems often input numbers and numeric strings such as telephone numbers and member numbers. And in the case of number recognition, the recognition dictionary is narrowed down to numbers, and similar utterances, for example, 1 and 7 are not mistaken, for example 7 is spoken as “Nana” instead of “Shichi”. I was able to increase the recognition rate.

On the other hand, when an arbitrary alphabetic character or alphanumeric character is used, for example, characters of almost the same utterance such as “9” and “Q” appear, which makes it very difficult to recognize. Regarding English letters, the recognition rate was improved by removing alphabets that are easily mistaken from recognition objects (dictionaries). As a related technology related to speech recognition for this type of short word data, there is one described in Patent Document 1.
In the case of Patent Document 1 relating to speech recognition for short word data, since there are many identical or similar word data, it takes a lot of time for interactive response, and misrecognition The feature is that the similar word table is simplified in order to reduce the number of times of interactive processing.

  On the other hand, regarding the recognition rate improvement related to voice recognition, there is a related art of Patent Document 2 as a related technique for reducing a burden at the time of error correction, and there is a related art of Patent Document 3 as a related technique related to voice recognition.

Among these, the thing of patent document 2 requires the input of the combination (sentence or sentence itself) of several words, and the similarity is determined from the combination between these words. Patent Document 3 relates to counseling in a hospital. A voice analysis process that analyzes input speech to extract speaker's psychological information and a problem plan by extracting problems from the converted character information. In addition, the process of creating a corrective measure proposal is an essential requirement.
The methods disclosed in Patent Documents 2 and 3 are methods that attempt to analyze the content of a sentence or sentence itself of a voice sentence without a response with a speaker.

JP-A-2004-233691 JP2007-256836 JP-A-9-114493

As described above, conventionally, for example, short words such as the numbers “1 (1)” and “7 (shichi)” and the English letters “C (shi)” and “E (ii)” It is known for its many similar utterances and easy misrecognition.
Since this phenomenon is likely to occur even in a person-to-person conversation, there is a limit to the solution for improving the recognition accuracy targeting the recognition engine.

  Moreover, in the voice response system of the above-mentioned patent document 1 known by the related art mentioned above, when it misrecognizes, it repeats several times and repeats many times by the same word. FIG. 10 shows an example of this case.

In the example of FIG. 10, when “1” is input from the speaker (steps S401 to S402), the system easily misidentifies “Michi” (step S403).
In response to this, the system prompts a recurrent voice to the speaker “No” (step S404) (step S405), and “1” is again input by the speaker (step S406). Further, the system misidentifies “Michi” (step S407), and in response to “No” after confirmation to the speaker (step S408), the system outputs a recognition failure as a voice (step S409).

  On the other hand, when there are a plurality of recognition results, there is also a method in which candidates are repeated in order and confirmed. However, when it is difficult to hear the voice synthesis being read back, or when it is the same sound, for example, the number 9 and the letter Q, recognition or confirmation by the device is practically difficult. FIG. 11 shows an example of this case.

The case of FIG. 11 shows a case where the system recognizes “Michi” for the voice input “1” of the speaker (steps S501 to S502) (step S503). Further, in response to the response “No” of the speaker (step S504), the system recognizes “1” (step S505), but indicates that the speaker misidentifies “No” (cannot be heard) ( Step S506). Also in this case, the system outputs the recognition failure as a voice and ends (step S507).
As described above, in the related art described above, the recognition of a short word has a disadvantage that misrecognition easily occurs because there are many similar utterances.

  On the other hand, each of the techniques disclosed in Patent Documents 2 and 3 described above is a technique of trying to analyze the content of a sentence or sentence itself of a speech sentence, and the short word described above. In addition, no solution is disclosed for the recognition and identification of voice data with many similar utterances, and it cannot be dealt with at all.

(Object of invention)
The present invention improves the inconvenience of the related art and enables a speech recognition and response method, speech that can quickly and effectively identify words having the same sound significance and words having similar readings. It is an object of the present invention to provide a recognition response system and a program thereof.

In order to achieve the above object, in the speech recognition response method according to the present invention,
The voice input device converts the voice of words sent from the outside into voice data and inputs it,
The voice recognition device recognizes the input voice data and searches for the same or similar word data corresponding to the voice data and a modifier for identifying the voice data by using the voice recognition device and the modifier word attached thereto. The device searches and extracts each separately from a preset storage unit, and then the speech generation and output device generates a standard sentence for confirmation of the text format based on the extracted word and its identifying modifier And externally output via the audio playback device,
The voice recognition device and the modifier search device function in the case where a response that denies the content is input from the outside with respect to the externally output confirmation fixed phrase, and the voice recognition device and the modifier search device and other same or approximate word data Re-extracting the modifier for identification from the storage unit, respectively,
Based on the re-extracted each word data and its modifier for identification, the voice generation output device again generates a confirmation fixed phrase, and then externally outputs it through the voice reproduction output device,
The voice generation / conversion device generates a confirmation standard sentence that indicates that the inputted voice has been recognized, which functions when an externally acknowledged response is input to the confirmation standard sentence. However, it is characterized in that it is externally output via an audio reproduction output device.

In order to achieve the above object, in the voice recognition response system according to the present invention,
A speech input device that converts the speech of a word sent from the outside into speech data and inputs it, and a recognition word dictionary as a storage unit in which one or more word data that is the same as or similar to the input speech data is installed in advance A speech recognition device that retrieves and extracts the speech data, a speech generation conversion device that applies word data corresponding to the speech data extracted by the speech recognition device to a pre-stored fixed form sentence, and the speech generation conversion A voice reproduction output device that converts voice data in a sentence format generated and output by the device into voice and outputs it externally;
When an affirmative or negative voice is externally input to the voice input device with respect to sentence-format voice data output from the voice reproduction output device, the voice recognition device copes with this and performs predetermined voice data processing. A voice recognition response system to be executed,
Between the speech recognition device and the speech generation conversion device, a modifier relation dictionary as a storage unit in which modifier word data having a high degree of relation directly related to the word data is stored in advance is searched, and the order of the relation degree is high. Via a modifier search device that extracts modifier data,
The speech generation / conversion device synthesizes the modifier data with a high degree of relevance extracted by the modifier search device with the corresponding word data to generate a fixed phrase with a modifier and use it as a response with an external speaker It is provided with a modifier synthesis output function for outputting.

Furthermore, in order to achieve the above object, in the speech recognition response program according to the present invention,
A word data extraction processing function for recognizing voice data input from the outside via a voice input device and searching a recognition word dictionary preliminarily equipped with the same or similar word data corresponding to the voice data; Based on the extracted word data and the modifier for identifying the modifier, a modifier-related dictionary for searching for and extracting the modifier for identifying the voice data attached to the word data is extracted. Standard text generation processing function for confirmation that generates standard text for confirmation in sentence format, and standard text output control processing function that outputs the generated standard text for confirmation via the audio playback output device for confirmation And
It functions when a response that denies the content from the outside is input to the standard sentence for confirmation output from the outside via the voice input device, and the same or approximated other corresponding to the voice data Re-extract word data and modifiers for identification thereof from the recognition word dictionary and the modifier relation dictionary by operating the word data extraction processing function and the modifier extraction processing function, respectively, and the re-extracted word data A re-extract generation function for generating the confirmation standard sentence again based on the identification modifier, and a standard form for re-outputting the re-extracted and generated standard sentence for confirmation to the outside via an audio reproduction output device Sentence re-output control processing function,
A voice reproduction output device that generates a confirmation standard sentence that indicates that the inputted voice has been recognized by functioning when an externally acknowledged response is input to the confirmation standard sentence A response confirmation processing function is provided for external output via
Each of these processing functions is executed by a computer.

  Since the present invention is configured as described above, according to this, a combination sentence in a limited state of “modifier + word” can be repeated and output to the outside, so that the speaker's response to this is also clearer Therefore, even for words with the same sound significance and words with similar readings, the modifiers function effectively and can be identified efficiently. A voice recognition response method, a voice recognition response system, and a program thereof can be provided.

1 is a block diagram showing a first embodiment of the present invention. FIG. 2A is a diagram showing a sample of words or modifiers stored in a dictionary serving as a storage unit disclosed in the first embodiment of FIG. 1, and FIG. 2A is an example of a dictionary sample of words stored in a recognized word dictionary; FIG. 2B is a chart showing an example of a modifier word dictionary stored in the modifier relation dictionary. It is explanatory drawing which shows the flow of the voice recognition response with a speaker at the time of operation | movement of 1st Embodiment disclosed in FIG. It is explanatory drawing which shows the example of extraction of the word data in the national language dictionary area | region in the recognition word dictionary disclosed in FIG. 1, and the example of a relevance degree as the modifier. It is explanatory drawing which shows the example of description of the word data in the national language dictionary area | region in the recognition word dictionary disclosed in FIG. It is explanatory drawing which shows the usage example of the relevance degree of the modification relation dictionary disclosed in FIG. It is a block diagram which shows 2nd Embodiment of this invention. It is explanatory drawing which shows the example of the function of the speech recognition apparatus disclosed in FIG. It is explanatory drawing which shows the flow of the voice recognition response with a speaker at the time of operation | movement of 2nd Embodiment disclosed in FIG. It is explanatory drawing which shows the flow (in the case of a retry pattern) of the speech recognition response with the speaker at the time of the speech recognition performed by related technology. It is explanatory drawing which shows the flow (when there are multiple word patterns) of the speech recognition response with the speaker at the time of the speech recognition performed by related technology.

[First Embodiment]
Hereinafter, a first embodiment of a speech recognition and response system according to the present invention will be described with reference to FIGS.

(Constitution)
In the speech recognition and response system according to the first embodiment, association information of a short word and a word that modifies the word is prepared in advance. When there are a plurality of short word recognition candidates, the combination of “modifier + short word” is repeated for the outside speaker, and the user is prompted to confirm either “yes” or “no”. When the answer from the speaker is “No”, the second candidate “modifier + short word” combination is repeated.

  In FIG. 1, a speech recognition and response system 10 converts a speech of a word sent from the outside into speech data and inputs it, and one or two or more word data that are the same as or similar to the input speech data. A speech recognition device 13 that searches and extracts a recognition word dictionary 13A as a storage unit installed in advance, and word data corresponding to the speech data extracted by the speech recognition device 13 are stored in advance. A voice generation / conversion device 16 applied to a fixed form sentence, and a voice reproduction output device 18 that converts voice data in a sentence format generated and output by the voice generation / conversion device 16 into voice and externally outputs the voice data.

  Then, from the speaker who recognizes the voice data in the text format output from the voice reproduction output device 18, an affirmative (for example, “Yes”) or negative voice (for example, “No”) of the content is voiced. When input to the input device 11, the above-described speech recognition device 13 is configured to cope with this and execute predetermined speech data processing.

  Between the speech recognition device 13 and the speech generation / conversion device 16, the modifier relation dictionary 14A which is a storage unit in which modifier word data having a high degree of association directly related to the word data described above is stored in advance is searched for A modifier search device 14 is provided for extracting the modifier data in descending order.

  Further, the speech generation / conversion device 16 described above generates a modifier-fixed sentence by synthesizing the highly relevant modifier data extracted by the modifier search device 14 with the corresponding word data described above, Is provided with a modifier synthesizing output function for outputting a response to a speaker existing in

  Here, the speech generation conversion device 16 described above synthesizes the modifier data with a high degree of relevance extracted by the modifier search device 14 and the word data corresponding thereto to generate a sentence with a modifier prefix. It comprises a generating means 16A and a voice converting means 16B that converts the generated fixed phrase with a modifier into voice data and outputs it.

  In addition, in the recognition word dictionary 13A as the storage unit described above, as shown in FIG. 2A, for one word, a plurality of words whose readings are the same, and the one word and its readings. A plurality of words approximated by are stored in advance so as to be searchable. In the recognition word dictionary 13A, as will be described later, a word ID (word identifier), a word notation, a reading, and a part of speech are associated with each other and stored in one line. . Thereby, it is stored so that it can be reliably searched from any.

Further, as shown in FIG. 2 (B), a plurality of modifiers having a high degree of relevance corresponding to the reading of each word are added to the modification relation dictionary 14A serving as the storage unit described above, and the degree of association is added in advance. Stored as possible.
That is, each of the plurality of search words stored in the modification relation dictionary 14A includes, for example, a word ID that is an identification symbol in advance and a word ID of each related word stored in the recognition word dictionary 13A described above. And a numerical value indicating the degree of the degree of relevance, and is stored so as to be searchable from any of them. Here, the number of relevance means that the larger the number, the closer the relevance.

  In the voice recognition device 13 described above, the voice data input to the voice input device 11 as a response from the speaker who has received the voice data of the fixed phrase with modifiers from the voice reproduction output device 18 is output as described above. It is provided with a function for determining whether or not the description content of the fixed phrase with a modifier is an affirmative content or a negative content.

  Further, the modifier search device 14 described above operates based on a negative determination made by executing the availability determination function of the speech recognition device 13 and operates based on this to determine the modifier of the fixed phrase with the modifier of the speech data described above. It has a function of selecting and outputting the part having the next highest value of the relevance value.

  In conjunction with this negative determination, the speech generation conversion device 16 described above, the modifier data extracted by the modifier search device 14 with the next highest degree of association, and the word data related to the negative determination described above, A re-synthesizing output function is provided that forms a fixed phrase with modifiers and re-outputs them.

  On the other hand, when an affirmative determination is made by execution of the availability determination function of the speech recognition device 13, the speech generation conversion device 16 described above operates in conjunction with this based on the affirmative determination by the speech recognition device 13. A recognition standard sentence output function for generating a recognition standard sentence including the word data as a standard sentence indicating that the word data related to the voice data has been recognized, and converting and outputting it is provided.

  In this case, the recognized word data is sent from the speech recognition device 13 to the speech generation / conversion device 16 via a text data storage device 15 described later. At the same time, the recognized word data is stored in the voice data storage device 12 and then sent to a necessary data processing device (not shown) by the recognized word data output control unit 20A. Yes.

In the modifier search device 14 described above, a text data storage device 15 for temporarily storing the word data selected and extracted by the voice identification device 13 and the modifier data extracted by the modifier search device 14 as text data. Is attached.
The speech generation / conversion device 16 described above is configured to execute the above-described modifier synthesis output function for the word data stored in the text data storage device 15 and the modifier data corresponding thereto. ing.

A voice data storage device 12 for storing voice information received by the voice input device 11 is interposed between the voice input device 11 and the voice identification device 13.
A part of the audio data storage device 12 is also interposed between the audio reproduction output device 18 and the audio generation conversion device 16 described above, and is a storage area different from the audio information storage area described above. A function of storing audio data output from the audio generation / conversion device 16 is provided.

  Reference numeral 20 denotes a main control unit that controls each component of the above-described constituent elements so that the function can be executed effectively and smoothly. Further, the voice data storage device 12 is provided with a recognized word data extraction control unit 20A, and the recognized word data is externally output for external processing via the recognized word data extraction control unit 20A. It has become.

Here, each component of this 1st Embodiment mentioned above is demonstrated more concretely.
The audio reproduction output device 18 that reproduces and outputs audio is for reading and reproducing audio data from the audio data storage device 12, and is composed of, for example, a speaker provided with an amplifier.
The audio reproduction output device 18 for reproducing and outputting the audio may be configured to transmit audio data to the speaker of the other party telephone via the telephone network.

  The reproducible audio data sent to the audio reproduction output device 18 is either audio data stored in the audio data storage device 12 in advance or audio data created by the audio data conversion means 16B for converting text into audio data. There may be.

  The voice input device 11 for inputting voice from an external speaker includes a microphone itself or means for converting voice received from the microphone of the other party's telephone via the telephone network into voice data. It has a function of sending data to the audio data storage device 12.

  The speech recognition device 13 has a function of analyzing speech data stored in the speech data storage device 12 by the speech input device 11 and acquiring words and word attributes of a similar utterance from the recognized word dictionary 13A. In this case, when there are a plurality of similar utterances, a function of acquiring the attributes of the plurality of words from the recognized word dictionary 13A after assigning priorities (scores) is provided.

  In the recognition word dictionary 13A, as described above, information related to the word ID, notation, reading, and part of speech is stored in advance. Among these, in this embodiment, the word ID is set to the first number, the notation is the word notation, the reading is the word reading, and the part of speech is the word part of speech.

  The modifier search device 14 that searches for and acquires modifiers corresponding to word data from the modifier relation dictionary 14A described above arranges the words acquired by the speech recognition device 13 in descending order of priority, and modifies each word. It has a function of acquiring a modifier with a high degree of association from the relation dictionary 14A and storing it in the text data storage device 15 in the form of “modifier + short word”.

  Here, in the modification relation dictionary 14A, three pieces of information relating to the word ID, the related word ID, and the degree of association are stored in advance. Among these, in this embodiment, the word ID is set as the word ID of the modifier, the related word ID is set as the word ID of the modified word, and the degree of association is set higher as the two words are synonymous.

  Further, the sentence generation means 16A for generating a sentence for transmitting voice data to the speaker combines a preset standard character string and an atypical character string stored in the text data storage device 15. Speech synthesis. The text generated by the text generation means 16A is sent to the voice reproduction output device 18 via the voice conversion means 16B, and is reproduced and output from the voice reproduction output device 18.

(Operation of the first embodiment)
Next, the operation of the first embodiment will be described.
First, the speech of a recognition word sent from an external speaker is converted into speech data and input via the speech input device 11 (speech data input step).

  Next, the input voice data is stored in the voice data storage device 12 and subsequently recognized by the voice recognition device 13. At the same time, the voice recognition device 13 searches the extracted word word dictionary 13A, which is a preset storage unit, and extracts the same or approximate word data corresponding to the voice data. The above word data is determined in priority and sent to the modifier search device 14 (word data search and extraction step).

  In response to this, the modifier search device 14 operates immediately and searches and extracts the modifier relation dictionary 14A, which is a preset storage unit, corresponding to the word data (modifier search extraction step). Then, the modifier for identification and the extracted word are sent to the text data storage device 15 and stored.

  Subsequently, the words stored in the text data storage device 15 and the modifiers for identification thereof are acquired by the speech generation output device 16. In the voice generation / output device 16, first, the sentence generation unit 16 </ b> A operates to generate a confirmation sentence in the form of a sentence based on the sent word data and its modifier for confirmation (confirmation fixed sentence generation step). ).

  Subsequently, the voice conversion means 16B is operated to convert the confirmation template into voice data and store it in the voice data storage device 12, and then the voice reproduction output device 18 is operated to store the stored confirmation template. The sentence is output as a voice response to an external speaker (response voice output step).

Next, it functions when a response for denying the content from the outside speaker is input via the voice input device 11 to the externally output fixed phrase for confirmation (negative response input step). The recognition device 13 and the modifier search device 14 sequentially re-extract the word data that is the same as or approximate to the word related to the confirmation fixed sentence and the modifier for identification from each storage unit.
Based on the re-extracted word data and its identifying modifier, the voice generation / output device 16 generates a confirmation standard sentence again, and then externally outputs it via the voice reproduction / output device 18. (Confirmed fixed phrase output process).

  This function works when an external speaker receives a positive response to the external output or the first confirmation phrase sentence voice output (acknowledgment input step). The confirmation standard sentence displaying the fact that the voice has been recognized is generated by the voice generation / conversion device 16 and externally output as a voice response to an affirmative response via the voice reproduction output device 18 (confirmation fixed sentence output step) ).

Here, when extracting the above-described word data and its identifying modifier from the storage unit, as described above, first, the word data from one storage unit (recognized word dictionary 13A) in which the word data is stored in advance. Based on this, the modifier for identification is searched and extracted from the other storage unit (modification relation dictionary 14A).
In addition, when extracting the modifier for identification, in this embodiment, preset modifiers having a high degree of association are selected and extracted sequentially.

Next, the details of the operation will be described more specifically with reference to FIG.
Here, it is assumed that the voice input device 11 is configured to target the other party (speaker) who talks via the microphone of the telephone.

First, the conversation start voice data stored in advance in the voice data storage device 12 is transmitted to a speaker (a speaker who makes a call to the system).
For example, the voice reproduction output device 18 sends a message “Please generate a word” to the speaker, and prompts the other party to input by voice (step S101).

  On the other hand, if the speaker utters “1”, the utterance sound of “1” of the speaker is taken from the voice input device 11 and converted into voice data and stored in the voice data storage device 12. (Step S102).

  Next, the voice recognition device 13 analyzes the voice data stored in the voice data storage device 12, and the same word data “one (word ID“ 1 ”) is selected from the samples in the recognition word dictionary 13A shown in FIG. ", Reading" 1 ")" and the word data "way (word ID" 2 ", reading" Michi ")" are extracted. In this case, the priority is in the order of “Road” and “One”.

  Subsequently, the modifier search device 14 refers to the modifier relation dictionary 14A, and “road (word ID“ 5 ”)”, which is most related to “road (word ID:“ 2 ”)”, “one ( Select and acquire “one (word word ID“ 4 ”)” that is most relevant to the word ID “1”).

If each related word is the same word, a word having a high degree of relevance is acquired next.
For example, as shown in FIG. 6, when the recognition word candidates “ichi” and “shichi” are related by “numbers”, it is possible to distinguish between “ichi” and “shichi” even if they are repeated. Can not. For this reason, the related word having the next highest relevance value is selected.

In this case, the modifier search device 14 selects “road (word ID“ 5 ”)” having the highest relationship with “road (word ID“ 2 ”)” and stores it in the text data storage device 15.
On the other hand, “shichi” is incorporated such that “one” and “nanana” with the next highest degree of association are selected because “numbers” with the highest degree of association match (see FIG. 6).

Here, an example of the construction of an effective modification relation dictionary will be disclosed.
For example, in the case of a meaningful word, as shown in FIG. 4 to FIG. 5, a related word prepared in advance is extracted using a national language dictionary or a synonym dictionary. FIG. 4 shows an example of such a case. First, a case where a plurality of words “number” and “one” are extracted from the Japanese language dictionary shown in FIG. S201). In this case, the extracted words “number” and “one” are examined in the national language dictionary (FIG. 4: step S202). When a plurality of words are found, whether or not they have the same meaning is specified and the degree of association is determined (FIG. 4: step S203).

  On the other hand, in the case of a meaningless character, for example, a Chinese character, the reading or radical may be registered in advance as a modifier. For example, when the kanji “low” is recognized and there are candidates for “ro” and “ro”, “do you want to be“ low ”in“ Hagara ”?” ““ Low ”in“ Ozora ” Can you repeat it? Also, if you have a single hiragana character, you can register modifiers such as “Are you sure of“ Aiueo ”?” Or “Are you sure of“ Ka ”? Good.

Restore the description.
Next, the sentence generation unit 16A applies the text data from the text data storage device 15 to the fixed sentence, and generates a sentence “Are you sure you want to“ Michi ”in“ Road ”?”.

Next, the voice conversion means 16B is activated, and the text generated by the text generation means 16A is converted into voice data and stored in the voice data storage device 12.
Finally, the voice reproduction output device 18 is activated to output the voice data “Are you sure you want to“ Michi ”of“ Road ”?” Stored in the voice data storage device 12, and to the speaker via the handset. Transmit (step S103).

Following the processing of step S103, when a utterance “No” is sent from the speaker, the voice input device 11 converts the utterance “No” of the speaker into voice data, and stores it in the voice data storage device 12. Store (step S104).
Subsequently, the voice recognition device 13 analyzes the voice data stored in the voice data storage device 12 and recognizes this “No”.

Next, the modifier search device 14 is activated in response to the recognition of “no” by the speech recognition device 13, and is most related to “one (word ID 1)” having the next highest priority after “road”. The high “one (word ID 4)” is stored in the text data storage device 15.
Subsequent to the storing operation of the modifier searching device 14, the sentence generating means 16A is operated, and the text data from the text data storing device 15 is applied to the fixed sentence, so that “Are you sure? Is generated.

Next, the voice conversion means 16B inputs the text generated by the text generation means 16A, converts it into voice data, and stores it in the voice data storage device 12.
Finally, the voice reproduction output device 18 outputs the voice data “Are you sure of“ one ”?” Stored in the voice data storage device 12 and sends it to the speaker via the receiver. (Step S105).

Following the processing of step S105, when a utterance “Yes” is sent from the speaker, the voice input device 11 converts the utterance “Yes” of the speaker into voice data and stores it in the voice data storage device 12. Store (step S106).
Subsequently, the voice recognition device 13 analyzes the voice data stored in the voice data storage device 12 and recognizes and identifies this “yes”.

  Next, the modifier search device 14 receives the recognition result of the speech recognition device 13, determines “one (word ID 1)” as the final recognition result, and stores this text in the text data storage device 15. Subsequently, the sentence generation unit 16A operates to apply the text data from the text data storage device 15 to the standard sentence and generate a sentence “I recognized“ 1 ””.

Next, the voice conversion unit 16B converts the sentence generated by the sentence generation unit 16A into voice data and stores the voice data in the voice data storage device 12.
Finally, the voice reproduction output device 18 operates to output the voice data “Ichi” has been recognized stored in the voice data storage device 12, and transmits it to the speaker via the receiver (step S107). ).

  The voice data storage device 12 is provided with a recognition word data extraction control unit 20A, and the recognition word data is externally output for external processing via the recognition word data extraction control unit 20A. It has become.

  Here, in each operation of the above-described embodiment, the execution content executed by each component in each processing step may be programmed and executed by a computer.

[Effects of First Embodiment]
Since the first embodiment is configured and operates as described above, according to this, when recognizing a word, a combination of “modifier + word” is repeated for an external speaker, and “Yes” is selected. Or, if the answer from the speaker is “no”, repeat the combination of “modifier + short word” of another candidate (second candidate). Since the limited combination sentence of “modifier + word” is read back and output externally, the speaker's response to this is also clearer. Even for words whose readings are similar, the modifier can function effectively, and the speaker can efficiently identify the modifier.

  That is, according to the first embodiment, it is possible to greatly reduce the possibility of misrecognizing a word, or to recite the result of a misrecognition by a voice response system, that is, a mistake on the speaker side that occurs during a repetition In spite of this, it is possible to prevent the possibility that the speaker is mistaken when the speaker is correct. In this case, according to the first embodiment, the higher the relevance and uniqueness of a modifier word registered in the modifier relation dictionary, the higher the recognition effect and the shorter the processing time. It has become.

[Second Embodiment]
Next, a second embodiment of the present invention will be described with reference to FIGS.
Here, the same code | symbol shall be used about the same component as the speech recognition response system in 1st Embodiment mentioned above.

The second embodiment is characterized in that syntax analysis means is provided in the voice data input stage of the voice recognition apparatus 13 in the first embodiment described above shown in FIG.
Specifically, it functions between the voice data storage means 12 and the voice recognition device 13 shown in FIG. 1 described above and functions when the above-mentioned external response is a sentence-type response including a modifier. It is equipped with syntax analysis means 23A that analyzes the syntax of the text format of the content and divides it into specific speech data, its modifier data, and terms that connect them.

  Further, the second embodiment is characterized in that, instead of the speech recognition device 13 disclosed in FIG. 1, a speech recognition device 23 having a function capable of operating in correspondence with the syntax analysis means 23A described above is employed. Have.

The voice recognition device 23 has a function equivalent to that of the voice recognition device 13 disclosed in FIG. 1 described above, and also recognizes specific voice data and its modifier data sent from the syntax analysis means 23A. And a word data limited extraction function for searching and extracting word data corresponding to the voice data based on the recognized modifier and the specific voice data limited by the modifier.
Other configurations are the same as those of the first embodiment described above.

Next, the operation of the second embodiment will be described.
In the first embodiment described above, when the response from the outside (speaker) to the confirmation fixed phrase output from the voice reproduction output device 18 is a sentence format response including a modifier, the voice input device 11 Is inputted, converted into voice data, and stored in the voice data storage device 12.

  Next, when the response data in sentence format including this modifier is stored in the voice data storage device 12, the syntax analysis unit 23A immediately functions to analyze the syntax of the sentence format of the response content, and the specific voice data And the modifier data and the term that connects them are executed.

  Subsequently, the word data limited extraction function of the speech recognition device 23 functions based on the analyzed and specified speech data and its modifier data, and stores word data corresponding to the specific speech data in the storage unit ( The recognition word dictionary 13A) is searched and specific voice data is extracted.

This will be described more specifically with reference to FIG.
In the second embodiment, first, for example, when a sentence “Please say a word” is output from the audio reproduction output device 18 (step S301), in response to this, for example, “Kyukyu” "Is spoken and input (step S302). In this case, each part functions from the audio reproduction output device 18 in the same manner as in the first embodiment described above, and “Are you sure you want to use“ Q of the alphabet ”?” Is output. That is, the combination sentence “modifier + word” is repeated (output) (step S303).

  On the other hand, as a reply from the speaker, not only “Yes” and “No” but also a combination of “modifier” or “modifier + word” such as “numeral” or “numeral cue” is input. (Step S304), the parsing means 23A immediately functions as described above, and the syntax analysis of “word (noun)” + “no” + “word”, “word (adjective, adjective verb)” + “word” is performed. And two word recognitions are performed.

  Then, the modifier search device 14 selects, for example, a word in which two words recognized by the voice recognition device 23 are registered as a modifier relationship (word ID and related word ID) as a recognition result, and receives this. The voice reproduction output device again outputs “Are you sure you want to use the number 9” generated by the voice generation conversion means 1 (step S305)? That is, a combination sentence of “modifier + word” is repeated (output).

  On the other hand, when the voice “Yes” is input from the speaker (step S306), the entire system responds to ““ Kyu ”from the voice reproduction output device as in the case of the first embodiment described above. "Recognized" is output (step S307), and a series of word recognition operations are completed.

In this way, even if there are a plurality of candidates, it is possible to narrow down the recognition result to one only by receiving a reply from the speaker once.
Here, in the speech recognition device 23, when a combination of “modifier + word” is input, only “word” is extracted, and the word data processing is performed as in the case of the first embodiment described below. You may comprise so that (audio | voice data processing) may be performed.
Other configurations and the effects thereof are the same as those of the first embodiment described above.

  In the first embodiment of the present invention, as described above, modifiers corresponding to short words with similar utterances are registered in advance in the form of relationships as shown in FIGS. When repeating the word recognition result, information is automatically added in the form of “modifier” + “word”, and confirmation is requested from the speaker. For this reason, word recognition errors can be effectively reduced.

  Further, in each of the above embodiments, as shown in FIG. 2B, a modification relation dictionary 14A is provided as a storage unit, and a plurality of modifiers are prepared for one word (modified word), and the degree of association is set. Set (the higher the number, the higher the relevance). Furthermore, when acquiring the word modifier from the modifier relation dictionary 14A, the word having the highest degree of association is selected. At the same time, when there are a plurality of candidate word recognition results and the modifiers of the candidates match, a method of acquiring the modifier with the next highest degree of association from the modifier-related dictionary 14A is adopted. For this reason, it is possible to reduce word recognition errors more effectively.

  Further, in the second embodiment, not only “Yes” and “No” but also a utterance in the form of “modifier” or “modifier + word” is accepted in response to the word recognition result for the speaker. Since the speech recognition device 13 recognizes two words, ie, a short word and a modifier, which are parsed by the syntax analysis unit 23A, the modifier recognition device 14 recognizes the two words recognized by the speech recognition device 13. Is selected as a modification relationship (word ID and related word ID). For this reason, it is possible to reduce word recognition errors more effectively.

  As described above, in each of the above embodiments, the identification processing can be effectively performed on the input voice data of a short word, and in particular, a combination sentence in a limited state of “modifier + word”. The speaker's response to this can be made clearer, so that modifiers can be applied to words with similar meaning and words with similar readings. It is possible to obtain a speech recognition response method, a speech recognition response system, and a program thereof that are unique and capable of effectively identifying and efficiently identifying the same.

  The speech recognition and response system according to the present invention effectively uses speech recognition of short words, for example, numbers, English letters, hiragana, a single kanji character, homonyms, etc., in many fields that require recognition. be able to.

DESCRIPTION OF SYMBOLS 10 Speech recognition response system 11 Voice input device 12 Voice data storage device 13, 23 Speech recognition device 13A Recognition word dictionary as a storage unit 14 Modifier search device 14A Modification relation dictionary as a storage unit 15 Text data storage device 16 Speech generation conversion Device 16A Text generation means 16B Voice conversion means (voice data conversion means)
18 Voice reproduction output means 20 Main control section 20A Recognition word data output control section 23A Syntax analysis means

Claims (18)

The voice input device converts the voice of words sent from the outside into voice data and inputs it,
The voice recognition device recognizes the input voice data and searches for the same or similar word data corresponding to the voice data and a modifier for identifying the voice data by using the voice recognition device and the modifier word attached thereto. The device is separately searched and extracted from the preset storage unit,
Based on the extracted word and its modifier for identification, the speech generation and output device generates a standard sentence for confirmation in a sentence format and outputs it externally via the speech playback device,
The voice recognition device and the modifier search device function in the case where a response that denies the content is input from the outside with respect to the externally output confirmation fixed phrase, and the voice recognition device and the modifier search device and other same or approximate word data Re-extract the modifier for identification from the storage unit,
Based on the re-extracted word data and the modifier for identification, the speech generation output device again generates a confirmation standard sentence, and then externally outputs via the speech reproduction output device,
The voice generation / conversion device generates a confirmation standard sentence that indicates that the inputted voice has been recognized by functioning when a response is received from the outside to the confirmation standard sentence. And a voice recognition response method characterized in that it is externally output via a voice reproduction output device. The speech recognition response method according to claim 1,
When extracting the word data and its identifying modifier from the storage unit, the word data is first extracted from the storage unit in which the word data is stored in advance, and based on this, the identifying modifier is stored. A speech recognition response method, wherein a modifier for identification corresponding to the word data is searched and extracted from the stored storage unit. The speech recognition response method according to claim 2,
A voice recognition response method characterized in that, when extracting the modifier for identification, a preset modifier having a high degree of relevance is sequentially selected and extracted. The speech recognition response method according to claim 1, 2, or 3,
When the response from the outside to the confirmation standard sentence is a response in a sentence format including a modifier, the syntax analysis means equipped in advance immediately functions to analyze the sentence format syntax of the response content. Execute the operation to divide specific audio data, its modifier data, and the term that connects them,
Subsequently, based on this analyzed and specified voice data and its modifier data, the voice recognition device searches the storage unit for word data corresponding to the specific voice data and extracts a word data limited extraction function. A speech recognition response method characterized by being executed. A speech input device that converts the speech of a word sent from the outside into speech data and inputs it, and a recognition word dictionary as a storage unit in which one or more word data that is the same as or similar to the input speech data is installed in advance A speech recognition device that retrieves and extracts the speech data, a speech generation conversion device that applies word data corresponding to the speech data extracted by the speech recognition device to a pre-stored fixed form sentence, and the speech generation conversion A voice reproduction output device that converts voice data in a sentence format generated and output by the device into voice and outputs it externally;
When an affirmative or negative voice is externally input to the voice input device with respect to sentence-format voice data output from the voice reproduction output device, the voice recognition device copes with this and performs predetermined voice data processing. A voice recognition response system to be executed,
Between the speech recognition device and the speech generation conversion device, a modifier relation dictionary as a storage unit in which modifier word data having a high degree of relation directly related to the word data is stored in advance is searched, and the order of the relation degree is high. Via a modifier search device that extracts modifier data,
The speech generation / conversion device synthesizes the modifier data with a high degree of relevance extracted by the modifier search device with the corresponding word data to generate a fixed phrase with a modifier and use it as a response with an external speaker A speech recognition response system characterized by having a modifier synthesis output function for outputting. The voice recognition response system according to claim 5,
The speech generation / conversion device synthesizes the modifier data having a high degree of relevance extracted by the modifier search device and the word data corresponding to the modifier data, and a sentence generator that generates a modifier fixed phrase, A speech recognition and response system, comprising: speech conversion means for converting the generated fixed phrase with modifiers into speech data and outputting the speech data. The recognition response system according to claim 5,
In the recognition word dictionary, a plurality of words whose readings are the same as one word, and a plurality of words whose readings are similar to the one word are stored in advance so as to be searchable,
A speech recognition response system, wherein a plurality of modifiers having a high degree of association corresponding to the reading of each word are stored in the modification relation dictionary so as to be searchable with a degree of association beforehand. The voice recognition response system according to claim 5,
A plurality of search words stored in the recognition word dictionary are preliminarily assigned with a word ID as an identification symbol, a reading thereof, and a part of speech indicating a noun, a verb, etc. A voice recognition response system characterized by being stored. The voice recognition response system according to claim 5,
For each of the plurality of search words stored in the modification relation dictionary, a word ID that is an identification symbol in advance, a word ID of each related word stored in the recognition word dictionary, and a degree of the degree of association A voice recognition response system characterized in that a numerical value representing the above is attached and is stored so as to be searchable from any of them. The voice recognition response system according to claim 7,
The speech recognition device determines whether the speech data externally input to the speech input device after outputting the speech data of the fixed phrase with a modifier from the speech reproduction output device confirms the description of the fixed phrase with the modifier It has a function to determine whether it is content to be
The modifier search device operates based on a negative determination by the availability determination function of the voice recognition device, and the modifier portion of the fixed phrase with a modifier in the speech data is the next highest value after the relevance value. It has a function to select and output,
The speech generation and conversion device has a resynthesizing output function for synthesizing the modifier data with the next highest degree of relevance extracted by the modifier search device and the extracted word data and re-outputting them as modifier fixed phrases. Voice recognition response system characterized by having. The speech recognition response system according to claim 10,
The speech generation / conversion device operates based on an affirmative determination by the speech recognition device, generates a recognition standard sentence including the word data as a standard sentence indicating that the word data related to the voice data has been recognized, and converts and outputs the recognition standard sentence. A speech recognition and response system characterized by having a recognition fixed sentence output function. The speech recognition response system according to claim 5 or 10,
Functions when the external response to the confirmation standard sentence is a sentence format response including a modifier, and analyzes the syntax of the sentence format of the response content to the voice data input stage of the speech recognition device Equipped with a syntax analysis means that divides speech data, its modifier data, and the terms that connect them,
The voice recognition device recognizes specific voice data and its modifier data sent from the syntax analysis means, and searches word data corresponding to the voice data based on the plurality of recognized data. A voice recognition response system characterized by having a limited word data extraction function for extraction. The speech recognition response system according to claim 5 or 10,
The modifier search device is provided with a text data storage device that stores the word data selected and extracted by the speech identification device and the modifier data extracted by the modifier search device as text data,
The speech recognition / response system, wherein the speech generation / conversion device is configured to execute the modifier synthesis function on word data and modifier data stored in the text data storage device. The speech recognition response system according to claim 5 or 10,
An audio data storage device that stores audio information received by the audio input device is interposed between the audio input device and the audio identification device, and
A speech recognition / response system, wherein a speech data storage device for storing speech data output from the speech generation / conversion device is interposed between the speech reproduction / output device and the speech generation / conversion device. A word data extraction processing function for recognizing voice data input from the outside via a voice input device and searching a recognition word dictionary preliminarily equipped with the same or similar word data corresponding to the voice data; Based on the extracted word data and the modifier for identifying the modifier, a modifier-related dictionary for searching for and extracting the modifier for identifying the voice data attached to the word data is extracted. Standard text generation processing function for confirmation that generates standard text for confirmation in sentence format, and standard text output control processing function that outputs the generated standard text for confirmation via the audio playback output device for confirmation And
It functions when a response that denies the content from the outside is input to the standard sentence for confirmation output from the outside via the voice input device, and the same or approximated other corresponding to the voice data The word data and the modifier for identification thereof are re-extracted from the recognized word dictionary and the modifier relation dictionary by operating the word data extraction processing function and the modifier extraction processing function, respectively, and the re-extracted word A re-extraction generation function for generating the confirmation fixed sentence again based on the data and its modifier for identification;
A fixed sentence re-output control processing function for re-outputting the re-extracted and generated fixed phrases for confirmation to the outside via the audio reproduction output device,
A voice reproduction output device that generates a confirmation standard sentence that indicates that the inputted voice has been recognized by functioning when an externally acknowledged response is input to the confirmation standard sentence A response confirmation processing function is provided for external output via
A speech recognition response program characterized by causing a computer to execute each of these processing functions. The speech recognition response program according to claim 15,
When executing the word data extraction processing function and the modification word extraction processing function, first, word data is extracted from a storage unit in which word data and a modification word for identification thereof are stored in advance, and the extracted word data A speech recognition response program characterized in that a modifier for identification is searched and extracted based on this, and is executed by the computer. The speech recognition response program according to claim 16,
When executing the modifier extraction processing function, a speech recognition response program characterized by sequentially selecting and extracting preset modifiers having a high degree of relevance, and causing the computer to execute the extraction processing. . The speech recognition response program according to claim 15, 16 or 17,
When the response to the confirmation standard sentence is a word response in a sentence format including a modifier, the parsing processing function for analyzing the sentence format of the response content, the modification specified by the syntax analysis A multi-word recognition processing function for recognizing a plurality of word data including words,
A limited word data extraction processing function for searching and extracting word data corresponding to the voice data based on a plurality of word data including the modifier,
A speech recognition response program characterized by causing the computer to execute each of these processing functions.

Images