JP2001013992A - Voice understanding device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To obtain a voice understanding device, which obtains a correct key word even though a necessary key word exists not only in one recognition result candidate but also exists in plural recognition result candidates. SOLUTION: A voice recognition section 101 outputs plural recognition result candidate sentences obtained as a recognition result and scores of each word included in each candidate sentence. A meaning extracting section 104 discriminates whether the key word exists in the plural recognition result candidate sentences outputted from the section 101 or not based on the stored contents of a conversation control section 106 and a meaning expression storage section 107 and outputs the key words existent in these sentences to a meaning output section 105. The section 104 selects a best score outputted by the section 101 for each key word when plural key words, which should not simultaneously exist in one uttering, exist.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech recognition apparatus for recognizing an input speech, and more particularly to a speech understanding apparatus for outputting a meaning intended by a user based on a recognition result of an input speech.

[0002]

2. Description of the Related Art As a speech recognition method used for a device for recognizing or understanding continuously uttered speech, for example, Japanese Patent Laid-Open No. 8-248988 discloses a grammar from a plurality of recognition result candidates obtained by acoustic processing. A speech recognition method has been proposed in which a recognition result having a high probability in terms of meaning or meaning can be obtained, so that a high recognition rate / meaning understanding rate can be obtained as a whole recognition processing.

In this conventional speech recognition method, after a sound processing section outputs a plurality of candidate recognition results, a language processing section gives a grammatical evaluation value to the recognition result candidates, and a grammatical evaluation value is given by the sound processing section. A linear sum obtained by appropriately weighting the obtained acoustic evaluation value and the grammatical evaluation value is defined as a comprehensive evaluation value, and the candidate having the highest overall evaluation value is determined as a recognition result.

[0004]

However, the above-mentioned conventional method has the following problems.

That is, one out of a plurality of recognition result candidates
In order to select one, if the required keyword does not exist only in one recognition result candidate, but exists in each of a plurality of recognition result candidates, the keyword cannot be extracted properly. It is.

For example, it is assumed that the user utters “Please give two tickets for tomorrow's A seat” and the keywords desired as output of the voice understanding device are “tomorrow”, “A seat”, and “two tickets”. I do. At this time, it is assumed that the recognition result in the sound processing unit is as shown in FIG. In FIG. 2, the numbers in parentheses indicate scores.

The keywords "tomorrow" and "seat A" are included in the first-ranked recognition result candidates, but the keywords "two"
Is included only in the second-ranked recognition result candidate, and one of the plurality of recognition result candidates is selected, so that the keyword "2" cannot be extracted. That is, when the required keyword is not present only in one recognition result candidate but is present in each of a plurality of recognition result candidates,
Can not be taken out well.

[0008] Therefore, the present invention has been made in view of the above problems, and an object of the present invention is to output a necessary keyword correctly even when a keyword exists in several recognition result candidate sentences. It is an object of the present invention to provide a voice comprehension device that can perform the same.

[0009]

According to the present invention for achieving the above object, a keyword included in each sentence is selected from a plurality of candidate recognition result middle sentences output from a speech recognition means based on a score corresponding to the keyword. A keyword selecting means is provided.

According to the present invention having such a configuration, even when a keyword is present in a plurality of recognition result candidate sentences, there is an increased possibility that a necessary keyword can be output correctly. Reduces re-entry effort.

[0011]

Embodiments of the present invention will be described. In a preferred embodiment, the speech understanding device of the present invention, in a preferred embodiment, recognizes speech and outputs a plurality of sentences as candidates for a recognition result and a score corresponding to each keyword included in the sentences of the recognition result candidates ( 10
1) and keyword selection means (104) for selecting an appropriate keyword based on the plurality of keywords output from the voice recognition means (101) and the score corresponding to the keywords.

When there are a plurality of keywords that cannot be present simultaneously in one utterance, the keyword selection means selects one keyword based on the score for each keyword output from the speech recognition means (101). .

According to a preferred embodiment of the present invention, a flow of a dialog prepared in advance is stored, and information on what keyword is included in a voice input from the next user is stored in a keyword selecting means. And a dialogue management means (106) for switching a keyword to be accepted for each state of each dialogue.

The present invention is a speech comprehension apparatus for recognizing input speech and outputting a plurality of keywords representing the meaning intended by the user based on the recognition result, and stores a recognition grammar used for speech recognition. Grammar storage unit for recognition, a recognition dictionary used for speech recognition and a recognition dictionary storage unit for storing information as to whether or not the word is a keyword, and a keyword that can be input by the user for each state of each dialogue is stored. And (a) performing speech recognition on the speech input by the user from the speech input unit with reference to the recognition grammar storage unit and the recognition dictionary storage unit. Voice recognition means for outputting a plurality of recognition result candidate sentences obtained as and a score of each word included in the candidate sentences,
(B) Based on the storage contents of the semantic expression storage unit, it is determined whether or not there is a keyword in each of the plurality of recognition result candidate sentences output from the voice recognition unit, and the existing keyword is output. At this time, when there are a plurality of keywords that cannot be present simultaneously in one utterance, the one with the best score output by the voice recognition means for each keyword
(A) to (c), meaning extraction means for selecting one of the meanings, and (c) meaning output means for creating and outputting a meaning expression by processing the information output from the meaning extraction means. May be realized by being executed by a computer constituting the speech understanding device.

Also, in the present invention, (d) receiving the information of the output result of the meaning output means, storing the flow of the dialog prepared in advance, and storing the next speech input in accordance with the output of the meaning output means A dialogue management unit that passes information about what keyword is included to the meaning extraction unit, wherein the meaning extraction unit is configured to output the information based on an output from the dialogue management unit and a storage content of the meaning expression storage unit. It may be configured to determine whether or not there is a keyword in each of the plurality of recognition result candidate sentences output from the voice recognition unit, and to output the existing keyword. The processing of the means (d) may be realized by being executed by a computer constituting the speech understanding device.

That is, the present invention can be implemented by reading out a program read from a recording medium or a communication medium storing the above-mentioned program into a computer and executing the program.

[0017]

Embodiments of the present invention will be described in detail with reference to the drawings.

FIG. 1 is a diagram showing the configuration of one embodiment of the present invention. Referring to FIG. 1, a speech understanding apparatus according to an embodiment of the present invention includes a speech recognition unit 101 for recognizing speech input by a user, and a recognition grammar storage unit 102 for storing a recognition grammar used for speech recognition. A recognition dictionary storage unit 103 that stores a recognition dictionary used for speech recognition and information as to whether the word is a keyword, a meaning extraction unit 104 that extracts a keyword included in a recognition result candidate,
Includes a meaning output unit 105 that outputs the obtained keywords, a dialog management unit 106 that manages the flow of the user's dialogue, and a meaning expression storage unit 107 that stores a keyword that can be input by the user for each state of the dialogue. .

The speech recognition unit 101 includes a recognition grammar storage unit 1
02 and the contents stored in the recognition dictionary storage unit 103, the user performs a voice recognition process on the voice input from a voice input unit (not shown) and converted into a digital signal, and a plurality of voices obtained as a recognition result are obtained. Recognition result candidate sentence (text information)
And the score of each word included in each candidate sentence
4 is output.

The score output from the speech recognition unit 101 is, for example, an acoustic score at the time of recognition processing, a language score created by the storage contents of the recognition grammar storage unit 102, or both an acoustic score and a language score. Is considered.

The recognition grammar storage unit 102 stores, for example, CFG (context-free grammar) or Bigr
A grammar such as a statistical language model such as am and Trigram is stored and held, and the stored content is passed to the speech recognition unit 101 as necessary.

The recognition dictionary storage unit 103 stores a recognition dictionary used for voice recognition, and transfers the stored contents to the voice recognition unit 101 as needed.

The meaning extracting unit 104 includes a dialog managing unit 106,
Based on the contents stored in the semantic expression storage unit 107, the speech recognition unit 1
It is determined whether a keyword exists in each of the plurality of recognition result candidate sentences output from No. 01, and the detected keyword is output to the meaning output unit 105.

At this time, when there are a plurality of keywords that cannot be present simultaneously in one utterance, one of the best scores output by the speech recognition unit 101 for each keyword is selected.

The meaning output unit 105 processes the information passed from the meaning extraction unit 104 to create and output a meaning expression.
Further, the output result information is passed to the dialog management unit 106.

The semantic expression output from the meaning output unit 105 includes, for example, a keyword string, data obtained by inputting a keyword obtained in a frame prepared in advance, and the like.

The dialogue management unit 106 stores a flow of the dialogue prepared in advance, and, in accordance with the output of the meaning output unit 105, information on what kind of keyword is included in the voice input from the next user. To the meaning extraction unit 104.

The semantic expression storage unit 107 stores a keyword that can be input by the user for each state of each conversation, and outputs the stored contents to the meaning extracting unit 104 as needed.

Next, an embodiment of the present invention will be described with reference to specific data. The dialog management unit 106
The user interaction flow is managed by the processing flow shown in FIG.

First, the user enters the state <reservation input> (20
In 1), enter the type of seat, date and time, and the number of seats. Then, in state <confirmation of recognition result> (202), check whether there is any problem in the recognition result. If there is no problem, the process ends. If there is, input again in the state <reservation input> (201).

Also, it is assumed that keywords as shown in FIG. 4 are registered in the semantic expression storage unit 107.

In the state <reservation input>, "today", "tomorrow", "tomorrow", "seat A"... Are used as keywords, and in the state <recognition result confirmation>, "yes" and "good"
... is a keyword.

In FIG. 4, for example, "A seat", "B seat",
"S seat" shall belong to the same attribute <seat type>
It is assumed that the attribute having the same attribute exists only once in one utterance.

It is assumed that the user voice-inputs "Please give two tickets for seat A tomorrow."

At this time, the recognition result in the voice recognition unit 101 is as shown in FIG. 2, and it is assumed that the result is passed to the meaning extraction unit 104.

In FIG. 2, the numbers in parentheses indicate the scores output by the voice recognition unit 101.

The meaning extracting unit 104 includes a dialog managing unit 106,
A keyword existing in the recognition result candidate sentence is checked from the storage contents of the semantic expression storage unit 107.

At this time, since the state is <reservation confirmation>,
It can be seen that “tomorrow”, “A seat”, “2 seats”, “B seat”, and “5 seats” are keywords included in the recognition result candidate sentence.

Here, since the meaning extraction unit 104 has the same attribute for "seat A" and "seat B" and "5 sheets" and "2 sheets", the score output by the speech recognition unit 101 is good. And delete "Seat B" and "5", "Tomorrow"
"Seat A" and "two sheets" are output as meaning expressions, and this information is sent to the dialog management unit 106. The dialog management unit 106 sets the state of the dialog as <confirmation of recognition result> based on the transmitted information.

[0040]

As described above, according to the present invention,
The following effects are obtained.

The first effect is to improve the performance of understanding input speech.

The second effect is to reduce the trouble of correcting erroneous recognition and re-inputting speech.

The reason is that not only the first recognition result candidate but also the keywords included in all the recognition result candidate sentences are extracted, so that the possibility that the keyword is correctly selected is improved.

[Brief description of the drawings]

FIG. 1 is a diagram showing a configuration of an embodiment of the present invention.

FIG. 2 is a diagram illustrating a specific example of a recognition result candidate according to an embodiment of the present invention.

FIG. 3 is a flowchart showing one specific example of a dialog flow in one embodiment of the present invention.

FIG. 4 is a diagram showing one specific example of keyword storage contents in one embodiment of the present invention.

[Explanation of symbols]

 Reference Signs List 101 speech recognition unit 102 recognition grammar storage unit 103 recognition dictionary storage unit 104 meaning extraction unit 105 meaning output unit 106 dialog management unit 107 meaning expression storage unit

Claims (8)

[Claims] 1. A speech understanding device for recognizing input speech and outputting a plurality of keywords representing a meaning intended by a user from a result of the recognition, comprising: A voice recognition unit that outputs a score corresponding to each keyword included in the recognition candidate sentence; a plurality of the keywords output from the voice recognition unit and a keyword that selects an appropriate keyword based on the score corresponding to the keyword A speech understanding device, comprising: selecting means. 2. The method according to claim 1, wherein said keyword selecting means selects one keyword based on a score for each keyword output from said voice recognizing means when there are a plurality of keywords which cannot be present simultaneously in one utterance. The speech understanding device according to claim 1, wherein 3. The speech comprehension apparatus according to claim 1, further comprising dialogue management means for switching a keyword to be accepted for each dialogue state. 4. A recognition grammar storage unit for storing a recognition grammar used for speech recognition, a recognition dictionary storage unit for storing a recognition dictionary used for speech recognition and information as to whether or not the word is a keyword. A semantic expression storage unit for storing a keyword that can be input by the user in association with attribute information for each state of each dialogue; a recognition grammar storage unit and the recognition dictionary for a voice input from a voice input unit A voice recognition unit that performs a voice recognition process with reference to a storage unit and outputs a plurality of recognition result candidate sentences obtained as a recognition result of the input voice and a score of each word included in the candidate sentences; Based on the storage contents of the semantic expression storage unit, determine whether there is a keyword in each of the plurality of recognition result candidate sentences output from the speech recognition means, and output a detected keyword, a meaning extraction means, And a meaning output means for processing the information output from the meaning extraction means to create and output a meaning expression. 5. A method according to claim 1, wherein said semantic extraction means includes a plurality of keywords which cannot be present simultaneously in one utterance in said recognition result candidate sentence. 5. The speech understanding device according to claim 4, wherein one of the best ones is selected. 6. Receiving information on the output result of said meaning output means, storing a flow of a dialog prepared in advance, and according to the output of said meaning output means, what keyword is included in the next voice input. Dialogue means for passing information to the meaning extraction means, the meaning extraction means being output from the speech recognition means based on the output from the dialogue management means and the contents stored in the meaning expression storage unit. 5. The speech understanding device according to claim 4, wherein it is determined whether or not there is a keyword in each of the plurality of recognition result candidate sentences, and the existing keyword is output. 7. A speech comprehension apparatus for performing, by a computer, a process of recognizing an input speech and outputting a plurality of keywords representing a meaning intended by a user based on a result of the recognition. A recognition grammar storage unit for storing grammar, a recognition dictionary for use in speech recognition and a recognition dictionary storage unit for storing information as to whether or not the word is a keyword, and a keyword that can be input by the user for each state of each dialogue And (a) performing voice recognition on the voice input by the user from the voice input unit with reference to the recognition grammar storage unit and the recognition dictionary storage unit. Voice recognition processing for outputting a plurality of recognition result candidate sentences obtained as a recognition result of the input speech and scores of each word included in the candidate sentences; and (b) storage contents of the semantic expression storage unit. Based on the above, it is determined whether or not there is a keyword in each of the plurality of recognition result candidate sentences output from the voice recognition process, and the detected keyword is output. At this time, the keyword may be present simultaneously in one utterance. If there are a plurality of missing keywords, a meaning extraction process for selecting one having the best score output by the voice recognition process for each keyword; and (c) processing information output from the meaning extraction process. A recording medium which stores a program for causing the computer to execute each of the above-described processes (a) to (c) of creating and outputting a semantic expression. 8. A speech comprehension apparatus for performing, by a computer, a process of recognizing input speech and outputting a plurality of keywords representing a meaning intended by a user based on a result of the recognition. A recognition grammar storage unit for storing grammar, a recognition dictionary for use in speech recognition and a recognition dictionary storage unit for storing information as to whether or not the word is a keyword, and a keyword that can be input by the user for each state of each dialogue (A) referring to the recognition grammar storage unit and the recognition dictionary storage unit for the voice input by the user from the voice input unit. Speech recognition processing for performing speech recognition and outputting a plurality of recognition result candidate sentences obtained as a result of recognition of the input speech and a score of each word included in the candidate sentences; Based on the content stored in the current storage unit, it is determined whether or not there is a keyword in each of the plurality of recognition result candidate sentences output from the voice recognition process, and the detected keyword is output. If there are a plurality of keywords that cannot be present at the same time during the utterance, a semantic extraction process of selecting one having the best score output by the speech recognition process for each keyword; and (c) a semantic extraction process. Processing the output information to create and output a semantic expression; and (d) receiving the output result information of the semantic output processing and receiving;
A dialog management process of storing a flow of a dialog prepared in advance, and passing information on what keyword is included in the next voice input to the meaning extraction process in accordance with the output of the meaning output means. The semantic extraction processing is based on the output from the dialog management processing and the storage contents of the semantic expression storage unit, and determines whether or not there is a keyword in each of the plurality of recognition result candidate sentences output from the speech recognition means. A storage medium for storing a program for outputting the detected keyword, and causing the computer to execute each of the processes (a) to (d).