JP2004233852A - System and method for supporting preparation of speech response application - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a system and a method for supporting the preparation of a speech response application that can structure a speech response application by generating an interaction script even when recognition grammar is no available. <P>SOLUTION: It is confirmed whether a secondary script including recognition grammar is available and when the secondary script is not generated, a temporary secondary script for urging a user to speak is generated according to a primary script, and interpreted and executed; and a user's speech is inputted as a speech signal and temporarily stored, at least a language code and a speech recognition engine are selected to analyzes the speech signal and extracts a keyword, and recognition grammar is generated according to the keyword to generate a secondary script in which the recognition grammar is incorporated. When the secondary script is generated, on the other hand, the secondary script is interpreted and executed. <P>COPYRIGHT: (C)2004,JPO&NCIPI

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a dialogue system for performing a task by interacting with an end user. In particular, a script language that describes the dialogue and execution tasks, an interpreter that interprets and executes the dialogue script described by the scripting language, an implementation platform that mediates the interpreter and the user, and a keyword extraction that is necessary for the dialogue script The present invention relates to a grammar generation device.
[0002]
[Prior art]
With the rapid development of computer technology in recent years, many applications utilizing voice-based dialogue have been developed. In these spoken dialogue systems, while repeating the utterance of the user and the utterance of the synthesized speech by the system alternately, necessary information is collected from the user while performing hierarchical branching according to the content of the user's utterance. Some kind of task is executed when important information is obtained.
[0003]
As a script language for describing these response scenarios, VoiceXML Forum (HYPERLINK http://www.voiceexml.org/) http: // www. voicexml. org / ) VoiceXML 1.0 is the mainstream and has already been released on a commercial basis. At present, VoiceXML 2.0 as a successor version is in the development stage at the World Wide Web Consortium (W3C; http://www.w3.org/).
[0004]
In addition, there is a scripting language that was originally formulated by a system integrator that provides voice interaction services. For example, Fujitsu Limited has released a script language for dialog description called "VoiceScript (R)".
[0005]
In order for these voice interaction systems to operate normally, the most important issue is to recognize the content of the utterance made by the input user with higher accuracy. In order to increase recognition accuracy, a method of automatically increasing the vocabulary of the recognition dictionary as in (Patent Document 1) or a method of extracting a keyword and extracting a recognition dictionary corresponding to the keyword as in (Patent Document 2) A method for generating the suffix is considered.
[0006]
Furthermore, when a script language is used, a recognition grammar (grammar) necessary for extracting a necessary command from words uttered by humans is required in order to more accurately recognize the content of the utterance by the user. In this case, if a recognition grammar that matches the utterance content of the user is assumed, the utterance content of the user is analyzed according to the recognition grammar, and the included keywords are extracted. Therefore, in order to operate a speech dialogue system, a dialogue script described in the above script language that describes a dialogue is not enough, and a recognition grammar for primary recognition of a user's utterance is also described. Need to be kept.
[0007]
Regarding the format of the recognition grammar, there are two formats that are considered in the Java Speech Grammar Format (JSGF) and the SpeechRecognition Grammar Specification Version 1.0 (SRGS), which is being developed by the World Wide Web Consortium. GrXML and the like are typical. There are also other formats that have been independently developed by voice recognition providers who have long offered voice recognition applications to the market.
[0008]
Script languages such as VoiceXML and SRGS are languages derived from the XML language which is extended based on the HTML language which is a script for describing web pages. However, compared to the HTML language which was mainly used to modify data, the programming language The color as language is very strong. To write a script using tags defined by each language specification, there is a method using a text editor. ⁺⁺ Just as there is an integrated development environment in which an editing environment and a debugger are integrated with programming languages such as Java and Java (R), some interactive system vendors prepare a dedicated development environment for easily writing scripts.
[0009]
[Patent Document 1]
JP 2002-14693 A
[0010]
[Patent Document 2]
JP-A-11-202890
[0011]
[Problems to be solved by the invention]
However, in the above-described method, it is necessary to prepare an interactive script and a recognition grammar having different formats, but it is not possible to configure an interactive application by itself. That is, it is necessary to prepare a recognition grammar corresponding to each interactive script, and the creator of the interactive application needs to generate the recognition grammar at the same time.
[0012]
In addition, it is practically difficult to prepare in advance all recognition grammars corresponding to each interactive script, due to the diversity of the recognition grammar, and due to the physical limitation of the storage capacity, It is also difficult to register the recognition grammar in advance.
[0013]
Further, when there is no recognition grammar, there is a problem that the content of the user's utterance cannot be recognized, and the response by the system becomes an irrelevant response. .
[0014]
The present invention provides a voice response application creation support system and method capable of constructing a voice response application by generating an interactive script even if there is no recognition grammar in order to solve the above problem. The purpose is to:
[0015]
[Means for Solving the Problems]
In order to achieve the above object, a voice response application creation support system according to the present invention is an automatic response application creation support system that supports creation of an automatic response application that automatically responds to user utterance, and includes a recognition grammar. A script interpreter that is an interpreter that checks whether or not a secondary script exists, selects and acquires a primary script or a secondary script according to the presence or absence, and interprets and executes the primary script or the secondary script. A voice input unit that inputs an utterance as a voice signal, a temporary storage unit that temporarily stores the input voice signal, and a platform control unit that selects at least a language code and a voice recognition engine based on the input voice signal, Analyze primary stored voice signal and extract keywords A secondary script generator that generates a recognition grammar based on the primary script and a keyword extracted from the voice signal, and generates a secondary script incorporating the recognition grammar. If no secondary script has been generated, a temporary secondary script for prompting the user to speak is generated based on the primary script, interpreted and executed by the script interpreting unit, and the secondary script is generated. Is characterized in that a secondary script is interpreted and executed by the script interpreter.
[0016]
With this configuration, even when there is no recognition grammar, it is possible to generate an appropriate recognition grammar according to the content of the utterance by the user, and it is possible to reliably configure a dialog as a voice response application.
[0017]
Preferably, the voice response application creation support system according to the present invention further includes a recognition grammar storage unit for storing the recognition grammar. This is because the recognition grammar can be updated and accumulated at any time according to the content of the user's utterance.
[0018]
Further, the voice response application creation support system according to the present invention determines whether or not a secondary script has been generated, and if a secondary script has not been generated, a temporary secondary script has been generated. It is preferable to further include a resource fetcher that passes the next script to the script interpreter.
[0019]
Further, the present invention is characterized by software that executes the function of the voice response application creation support system as described above as a processing step of a computer, and more specifically, an automatic response that automatically corresponds to a user utterance. An automatic response application creation support method that supports creation of an application. The method includes checking whether a secondary script including a recognition grammar exists, and selecting and acquiring a primary script or a secondary script according to the presence / absence of the primary script. Interpreting and executing a script or secondary script, inputting an utterance by a user as an audio signal, temporarily storing the input audio signal, and at least a language code based on the input audio signal. Selecting a speech recognition engine and analyzing a speech signal stored temporarily Generating a recognition grammar based on the keywords extracted from the primary script and the voice signal, and generating a secondary script incorporating the recognition grammar, wherein the secondary script is generated. If not, a temporary secondary script for prompting the user to speak is generated based on the primary script, interpreted and executed, and if a secondary script is generated, the secondary script is generated. The present invention is characterized in that it is an automatic response application creation support method that is interpreted and executed, and a computer-executable program that embodies such a process.
[0020]
With such a configuration, by loading and executing the program on a computer, even when there is no recognition grammar, it is possible to generate an appropriate recognition grammar according to the content of the utterance by the user, and as a voice response application It is possible to realize an automatic response application creation support system capable of reliably forming a conversation.
[0021]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, a voice response application creation support system according to an embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a configuration diagram of a voice response application creation support system according to an embodiment of the present invention.
[0022]
In FIG. 1, first, an utterance by a user, which is prompted by an audio output by a system output as a synthetic voice from an audio output unit 12, is input as an audio signal in an audio input unit 11. Then, the platform control unit 13 switches the language attribute and the recognition engine. That is, in accordance with a voice signal input from the voice input unit 11, a task of switching a language code or selecting an appropriate recognition engine is performed.
[0023]
In the platform control unit 13, the audio signal itself input from the audio input unit 11 is stored in the temporary storage unit 14. This is because a keyword for generating a recognition grammar is extracted using the stored voice signal.
[0024]
The script interpreter 15 interprets and executes a script prepared in the voice interaction application. Of course, for the interpreted script, the platform control unit 13 switches the language code according to the content of the interpreted script or selects an appropriate synthesized speech to generate a synthesized speech, It is output from the audio output unit 12 as an output.
[0025]
Next, the user input analysis unit 16 analyzes necessary keywords by performing morphological analysis or the like on the voice signal uttered by the user stored in the primary storage unit 14.
[0026]
Then, the secondary script generation unit 17 generates a recognition grammar based on the described script and keywords obtained from the voice signal, and updates the primary script prepared from the beginning to incorporate the recognition grammar. Generates a secondary script. If no recognition grammar exists, a temporary secondary script is generated. The generated secondary script or the temporary secondary script is interpreted by the script interpreting unit 15 and executed, so that the user can perform a voice dialogue. The recognition grammar generated here is stored in the recognition grammar storage unit 19. By storing the generated recognition grammar in the recognition grammar storage unit 19, it becomes possible to refer to the recognition grammar at the time of the next speech input.
[0027]
The confirmation of the presence or absence of the recognition grammar may be performed by the secondary script generation unit 17 when the secondary script is generated, or whether the recognition grammar corresponding to the recognition grammar storage unit 19 is stored. May be confirmed.
[0028]
The resource fetcher 20 interprets a script interpreted and executed by the script interpreter 15 into a temporary secondary script generated based on a primary script described at the time of application creation and a newly generated secondary script. Switch between scripts.
[0029]
Further, the keyword is not limited to the keyword extracted by being analyzed by the user input analysis unit 16, but may be a keyword that is stored in the document database 18 in advance.
[0030]
Next, the flow of processing in the voice response application creation support system according to the embodiment of the present invention having such a configuration will be described. FIG. 2 shows a flowchart of a program for realizing the voice response application creation support system according to the embodiment of the present invention.
[0031]
In FIG. 2, when the platform control unit 13 requests the script interpreting unit 15 to start a dialogue such as an incoming call (step S201), the script interpreting unit 15 determines whether a secondary script including a recognition grammar exists. Confirm (step S202).
[0032]
The method of checking the presence or absence of the recognition grammar is not particularly limited, and may be a method in which the secondary script generation unit 17 determines when the secondary script is generated, or the recognition corresponding to the recognition grammar storage unit 19. It may check whether the grammar is stored or not.
[0033]
If a secondary script including the recognition grammar has already been generated (step S202: Yes), the resource fetcher 20 is requested to retrieve the secondary script (step S203). The secondary script is interpreted and executed (step S204), and an utterance request is issued to the user by voice output by a system using synthesized voice or the like (step S205).
[0034]
If a secondary script including the recognition grammar has not been generated (step S202: No), the resource fetcher 20 is requested to take out the primary script (step S206). When the resource fetcher 20 informs the secondary script generation unit 17 of the fact, the secondary script generation unit 17 retrieves the primary script from the document database 18 and generates a formal temporary secondary script (step S207). . The temporary secondary script does not contain any recognition grammar at this time.
[0035]
When the generated temporary secondary script is passed to the script interpreting unit 15, the script is interpreted and executed (step S208), and an utterance request is issued to the user by voice output by a system using synthesized voice or the like (step S208). S209).
[0036]
Next, the utterance of the user prompted by the utterance request is input as a voice signal (step S210). Then, based on the input voice signal, the language code is switched, an appropriate recognition engine is selected (step S211), and the input voice signal itself is temporarily stored (step S212).
[0037]
Then, a necessary keyword is analyzed by morphological analysis or the like of the voice signal uttered by the user which is temporarily stored (step S213). Further, a recognition grammar is generated based on the described primary script and a keyword extracted from the voice signal, and the primary script is updated to generate a secondary script incorporating the recognition grammar (step S214).
[0038]
With such a configuration, even when no recognition grammar exists, a necessary recognition grammar can be generated based on the voice signal, and thus the voice response application can be normally executed. .
[0039]
Next, specific script processing will be described with reference to specific examples.
[0040]
First, by receiving a telephone call from the user or performing a start operation of a voice response application, a response request is made from the platform control unit 13 to the script interpretation unit 15 which is an interpreter.
[0041]
Then, the script interpreter 15 specifies the primary script as the initial operation script of the voice response application, and instructs the resource fetcher 20 to acquire the primary script. It is assumed that the primary script is a script described in VoiceXML as shown in FIG. 3, for example.
[0042]
However, as is clear from the description of the primary script shown in FIG. 3, the description content of the primary script does not describe the recognition grammar as information for recognizing which user's utterance is accepted. Dialogue is not established and the correctness of the script itself cannot be confirmed. Therefore, the creator of the voice response application needs to separately prepare a recognition grammar in order to confirm the operation of the described primary script.
[0043]
Therefore, the resource fetcher 20 delivers the primary script to the secondary script generation unit 17, and the secondary script generation unit 17 generates a temporary secondary script as shown in FIG. Here, the temporary secondary script means a secondary script in a state before the recognition grammar is incorporated, and is defined as a timing at which a voice signal obtained by a user's utterance is acquired. I do.
[0044]
In the temporary secondary script shown in FIG. 4, as indicated by the underlined portion, constant data used in the primary script is diverted as it is. The other parts are generated by a template generated from the syntax described in the primary script.
[0045]
The template is generated by the following processing. First, a tag that prompts the user for voice input and that actually performs an operation of filling a slot is prepared, and the tag is replaced with a tag for collecting a raw voice signal. In FIG. 4, <initial> and <field> are prepared as tags for prompting the user to input voice and perform an operation of filling a slot, while <record> is used as a tag for collecting a raw voice signal. It is prepared. By comparing FIG. 3 with FIG. 4, it can be seen that the <initial> tag and the <field> tag are replaced with the <record> tag in a one-to-one correspondence.
[0046]
On the other hand, tags <block>, <filled>, and <submit>, which are described only in the temporary secondary script, are generated in a fixed manner using constant values defined in the primary script as data. The <submit> tag also describes the content of processing to be executed in the tag. By executing the tag, a voice input by the user is passed to the user input analysis unit 16.
[0047]
That is, a tag such as a <field> tag in VoiceXML for obtaining a keyword by restricting an input by the user is obtained by collecting a raw audio signal (<record> tag) and generating a tag of the collected audio signal. It is characterized in that it is replaced with a series of operations of acquiring a keyword from the inside (passing it to the user input analysis unit 16 using a <submit> tag to extract the keyword).
[0048]
Note that the <submit> tag also exists in the primary script shown in FIG. 3, but is described to transition to the next task based on the <field> tag. The usage is different from the <submit> tag in the next script.
[0049]
Further, in the present embodiment, such a replacement process is performed by generating a template. However, a method is provided in which a script conversion table is provided in advance and tags are replaced by referring to the script conversion table. Is also good.
[0050]
Next, the script interpreting unit 15 reads, interprets, and executes the generated temporary secondary script to perform a dialogue as shown in FIG. 5 with the user. In FIG. 5, the name of the file in which the audio signal spoken by the user is temporarily stored in the temporary storage unit 13 is displayed in parentheses.
[0051]
Then, after the dialog shown in FIG. 5 is completed and the user inputs and temporarily stores all the requested voice signals, the script interpreting unit 15 outputs the voice signals stored in the temporary storage unit 13 spoken by the user and The corresponding primary script is passed to a user input analyzer 16 in the form of a so-called servlet.
[0052]
In the present embodiment, the recognition grammar generation process is performed after all voice signals have been collected, based on an utterance example (FIG. 5) for generating a temporary secondary script. Instead, the recognition grammar may be generated each time a user inputs a voice.
[0053]
Next, the user input analysis unit 16 includes, for example, a continuous speech recognition module and a morphological analysis module as submodules, and cuts out a phrase serving as a keyword from the content of the speech signal uttered by the user. For example, the file names tmp1. The wav is analyzed as shown in FIG.
Then, the analyzed contents are sent to the secondary script generation unit 17, and the secondary script generation unit 17 determines keywords necessary for the task and unnecessary keywords, and generates a recognition grammar. FIG. 7 shows an example of a recognition grammar generated in the secondary script generation unit 17 in an XML format.
[0054]
In FIG. 7, among the parts of speech extracted from user utterances, only nouns are accepted as keywords. Also, as shown in FIG. 7, in the present embodiment, two nouns appear in the example of the response to the corresponding <initial> tag in the primary script received at the same time. This is because it is assumed that the utterance corresponds to the tag. In this embodiment, each keyword is automatically associated with the <field> tag. However, by providing an auxiliary script, the user can determine which <field> tag each keyword corresponds to. You may ask to contact.
[0055]
Then, the secondary script generation unit 17 generates a secondary script in which the recognition grammar is embedded as shown in FIG. 8 by updating the primary script. In the underlined part in FIG. 8, by specifying the file name when the recognition grammar file shown in FIG. 7 is stored in the recognition grammar storage unit 19, the recognition grammar to be used when executing the voice response application becomes clear.
[0056]
By doing so, an optimal recognition grammar can be given to a script for which a recognition grammar was not originally prepared, and the script can be executed as a voice response application.
[0057]
Next, a case where the primary script is given in a form as shown in FIG. 9 will be described. In FIG. 9, “drink” and “food” are embedded as representative values of acceptable keywords at the time of writing the primary script, as indicated by the underlined portions. Of course, the keyword list may be stored as another file and referred to. The following processing is similar to the processing other than the keyword extraction processing.
[0058]
That is, by generating a keyword list in advance, keyword extraction processing is unnecessary or simple processing is sufficient, so that the overall processing load can be reduced.
[0059]
Further, as in the temporary secondary script shown in FIG. 10, the uttered voice signal corresponding to the keywords "dlink" and "number" is analyzed by the user input analysis together with the task processing program name described in the primary script. It is also conceivable that the description is passed to the unit 16. In this embodiment, the task processing program name is stored in the variable name TASK_PROCESSOR.
[0060]
By configuring the script in this way, it is also possible to directly execute as a voice response application without generating a secondary script incorporating a recognition grammar.
[0061]
As described above, according to the present embodiment, even when there is no recognition grammar, it is possible to generate an appropriate recognition grammar according to the content of the utterance by the user, and to reliably construct a dialog as a voice response application. It is possible to do.
[0062]
As shown in FIG. 11, the program for realizing the voice response application creation support system according to the embodiment of the present invention is not limited to the portable recording medium 112 such as the CD-ROM 112-1 and the flexible disk 112-2. The program may be stored in any of the other storage device 111 provided at the end of the communication line and the recording medium 114 such as the hard disk or the RAM of the computer 113. When the program is executed, the program is loaded. Runs on memory.
[0063]
Also, as shown in FIG. 11, information on the recognition grammar generated by the voice response application creation support system according to the embodiment of the present invention is a portable type such as a CD-ROM 112-1 or a flexible disk 112-2. Not only the recording medium 112, but also any other storage device 111 provided at the end of the communication line or a recording medium 114 such as a hard disk or a RAM of a computer 113 may be used. It is read by the computer 113 when using such a voice response application creation support system.
[0064]
【The invention's effect】
As described above, according to the voice response application creation support system according to the present invention, even when there is no recognition grammar, it is possible to generate an appropriate recognition grammar according to the content of the utterance by the user, and As a result, it is possible to reliably configure the dialogue.
[Brief description of the drawings]
FIG. 1 is a configuration diagram of a voice response application creation support system according to an embodiment of the present invention;
FIG. 2 is a flowchart of processing of a voice response application creation support system according to the embodiment of the present invention;
FIG. 3 is an exemplary diagram of a primary script in the voice response application creation support system according to the embodiment of the present invention;
FIG. 4 is an exemplary diagram of a temporary secondary script in the voice response application creation support system according to the embodiment of the present invention;
FIG. 5 is an exemplary diagram of a voice response in the voice response application creation support system according to the embodiment of the present invention;
FIG. 6 is an exemplary diagram of morphological analysis in the voice response application creation support system according to the embodiment of the present invention;
FIG. 7 is an exemplary diagram of a recognition grammar in the voice response application creation support system according to the embodiment of the present invention;
FIG. 8 is an exemplary diagram of a secondary script incorporating a recognition grammar in the voice response application creation support system according to the embodiment of the present invention;
FIG. 9 is another exemplary diagram of the primary script in the voice response application creation support system according to the embodiment of the present invention;
FIG. 10 is another exemplary diagram of a temporary secondary script in the voice response application creation support system according to the embodiment of the present invention;
FIG. 11 is an exemplary diagram of a computer environment.
[Explanation of symbols]
11 Voice input section
12 Audio output unit
13 Platform control unit
14 Temporary storage
15 Script interpreter
16 User input analysis unit
17 Secondary script generator
18 Document Database
19 Recognition grammar storage
20 Resource Fetcher
111 Destination storage device
112 Portable recording media such as CD-ROM and flexible disk
112-1 CD-ROM
112-2 Flexible disk
113 Computer
114 Recording media such as RAM / hard disk on computer

Claims (5)

An automatic response application creation support system that supports creation of an automatic response application that automatically responds to user utterances,
Check the presence or absence of a secondary script containing a recognition grammar, select and obtain a primary script or the secondary script according to the presence or absence, and interpret and execute the primary script or the secondary script with an interpreter. A script interpreter,
An audio input unit for inputting an utterance by the user as an audio signal;
A temporary storage unit for temporarily storing the input audio signal,
A platform control unit that selects at least a language code and a voice recognition engine based on the input voice signal;
A user input analysis unit for analyzing the voice signal temporarily stored and extracting a keyword;
A secondary script generation unit that generates a recognition grammar based on the keyword extracted from the primary script and the audio signal, and generates a secondary script incorporating the recognition grammar,
If the secondary script has not been generated, based on the primary script, to generate a temporary secondary script to prompt the utterance of the user, interpreted and executed by the script interpretation unit,
If the secondary script has been generated, the script interpreting unit interprets the secondary script and executes the script. 2. The system according to claim 1, further comprising a recognition grammar storage unit for storing the recognition grammar. It is determined whether or not the secondary script has been generated. If the secondary script has not been generated, the temporary secondary script is passed to the script interpreting unit. 3. The automatic response application creation support system according to claim 1, further comprising a resource fetcher. An automatic response application creation support method that supports creation of an automatic response application that automatically responds to a user utterance,
Confirming the presence or absence of a secondary script including a recognition grammar, selecting and acquiring a primary script or the secondary script according to the presence or absence, interpreting and executing the primary script or the secondary script, and ,
Inputting the utterance by the user as an audio signal;
Temporarily storing the input audio signal;
Selecting at least a language code and a speech recognition engine based on the input speech signal;
Analyzing the primary stored voice signal to extract keywords;
Generating a recognition grammar based on the keyword extracted from the primary script and the audio signal, and generating a secondary script incorporating the recognition grammar,
If the secondary script has not been generated, based on the primary script, to generate and interpret a temporary secondary script for prompting the user to utter and execute it,
If the secondary script has been generated, the secondary script is interpreted and executed. A computer-executable program embodying an automatic response application creation support method for supporting creation of an automatic response application automatically corresponding to a user utterance,
Confirming the presence or absence of a secondary script including a recognition grammar, selecting and acquiring a primary script or the secondary script according to the presence or absence, interpreting and executing the primary script or the secondary script, and ,
Inputting the utterance by the user as an audio signal;
Temporarily storing the input audio signal;
Selecting at least a language code and a speech recognition engine based on the input speech signal;
Analyzing the primary stored voice signal to extract a keyword;
Generating a recognition grammar based on the keywords extracted from the primary script and the audio signal, and generating a secondary script incorporating the recognition grammar,
If the secondary script has not been generated, based on the primary script, to generate and interpret a temporary secondary script for prompting the user to utter and execute it,
A computer-executable program for interpreting and executing the secondary script when the secondary script has been generated.

Images