JP2012247912A - Speech signal processing apparatus - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a technique for precisely extracting necessary information associated with business from a speech signal during a dialog of the business between speakers.SOLUTION: Speech recognition means 13 converts a speed signal during a dialog between speakers into speech recognition information. Conversation unit division means 14 divides the speech recognition information into conversation units. Necessary information clue information imparting means 16 imparts necessary information clue information included in the conversation units by the conversation units. Business determination means 15 determines the business of the dialog between the speakers on the basis of the speech recognition information. Dialog state transition determination means 17 selects a conversation unit including necessary information clue information as a clue to extraction of necessary information corresponding to the business. On the basis of the selected conversation unit and dialog state determination information stored in a conversation state determination information database 25 corresponding to the business, it is determined that the dialog state changes in the order of an information request dialog state, a request acceptance dialog state, an information disclosure dialog state, and an information acceptance dialog state, thereby extracting the necessary information.

Description

  The present invention relates to a technique for extracting necessary information relating to a requirement from a voice signal during dialogue between speakers for the requirement.

When a customer requests a business (service) from a company (for example, when a customer requests an electric power company to start [use of electricity], [stop use of electricity], or [change the contracted capacity of electricity]) The customer calls the corporate service center. At this time, the service center receptionist (operator) listens to the customer about the details of the request requested by the customer and the necessary information necessary for the execution of the requested request. Record on a computer or recording paper. In this case, the receptionist needs to listen to and record the contents of the business and necessary information from the customer, and the burden on the receptionist is large.
Therefore, in order to reduce the burden on the receptionist, there is a demand for the development of a speech signal processing apparatus that extracts necessary information from speech signals during conversation between speakers.
Until now, a technique for extracting (summarizing) necessary information from speech has been proposed. For example, in Patent Document 1, important information (words) during conversation is determined using a characteristic pattern (silence duration, speech level, similarity of speech waveform between speakers, etc.) as a clue by acoustic analysis, A method for speech summarization is disclosed. Patent Document 2 discloses a method of performing morphological analysis, syntax analysis, and semantic analysis on text information that is a speech recognition result, and determining an important part based on these results.
However, such a conventional speech summarization technique cannot accurately extract necessary information related to a requirement from a purpose-oriented dialogue with a clear requirement.

JP 2006-58567 A JP-A-8-212228

  The invention of the present application was devised in view of such points, and provides a technique capable of accurately extracting necessary information relating to a requirement from an audio signal during dialogue between speakers for the requirement. For the purpose.

  The audio signal processing apparatus of the present application extracts necessary information related to the requirement from the audio signal during the dialogue between the speakers for the predetermined requirement. As the “requirement”, typically, a service requested by a customer to a company (for example, [use of electricity], [stop use of electricity], [change of contracted capacity of electricity requested by a customer] ]). The “audio signal” typically corresponds to a signal output from a microphone or the like that converts audio into an electric signal. Since the signal processing in the processing means is performed digitally, a signal obtained by digitizing an analog signal output from a microphone or the like can be used as an “audio signal”.

One invention comprises a storage means, a processing means, an input means, and an output means.
The storage means has a necessary information database and a dialog state determination information database.
In the necessary information database, a necessary information list indicating necessary information related to the business is stored. “Necessary information” is information necessary for executing a requested business (for example, a service requested from a customer to a company), and is determined for each business.
The dialog state determination information database stores dialog state determination information for determining a dialog state between speakers. In the present invention, it is determined that the dialogue state between the speakers has changed in the order of the information request dialogue state in which the necessary information is requested, the information disclosure dialogue state in which the necessary information is disclosed, and the information reception dialogue state in which the necessary information is received. Based on the above, necessary information is determined. For this reason, as the dialogue state determination information, information request clue information for determining whether the conversation state requested the necessary information, information disclosure clue information for determining the conversation state disclosing the necessary information, necessary Information reception clue information for determining that the information is in a dialog state is stored.
The processing means includes voice recognition means, requirement determination means, dialog state transition determination means, and management means.
The voice recognition means converts the voice signal during the dialogue between the speakers input from the input means to voice recognition information including text information. Various known voice recognition methods can be used as a method for converting the voice signal into voice recognition information. For example, a large vocabulary continuous speech recognition method using HMM (Hidden Markov Model) and N-gram (Stochastic Language Model) is used. As the input means, for example, a microphone that converts sound into an electrical signal, an A / D converter that performs A / D conversion processing that converts an analog signal into a digital signal, and the like are used. The A / D conversion process can also be performed by voice recognition means. As the input unit, a reading unit that reads an audio signal or the like stored in a storage medium can be used. Further, in the configuration of the “speech recognition unit that converts a speech signal during dialogue between speakers input from the input unit into speech recognition information” according to the present invention, “speech recognition provided apart from the processing unit” The voice signal is converted into voice recognition information by the means, and the voice recognition information is input from the input means ".
The message determining means determines a message for dialogue between speakers based on the voice recognition information. As a method of “determining a requirement based on speech recognition information”, a method of determining a requirement from a character string included in the speech recognition information is typically used. In this specification, what is expressed using letters, numbers, symbols, and the like is referred to as a “character string”.
The dialogue state transition discriminating means discriminates the situation based on the voice recognition information by discriminating that the dialogue state between the speakers has transitioned in the order of the information request dialogue state, the information disclosure dialogue state, and the information reception dialogue state. Each necessary information shown in the necessary information list corresponding to the requirement determined by the means is extracted. For example, if the required information is [Contractor Name], the person name indicating the contractor's name is extracted. If the required information is [Contractor Address], the place name indicating the contractor's address is extracted. If the necessary information is [contractor telephone number], a number indicating the telephone number of the contractor is extracted. The fact that the dialogue state between the speakers has transitioned in the order of the information request dialogue state, the information disclosure dialogue state, and the information reception dialogue state is, for example, information request clue information stored in the dialogue state discrimination information database in the speech recognition information. This is determined by the fact that information disclosure clue information and information acceptance clue information are included in this order.
The management means outputs the necessary information extracted by the dialog state transition determination means from the output means. As the output means, for example, a display means, a printing means or the like is used. Various output modes can be used as a mode for outputting the necessary information extracted by the dialog state transition determination unit from the output unit.
In the present invention, since necessary information is extracted by determining that the dialog state has transitioned in the order of the information request dialog state, the information disclosure dialog state, and the information reception dialog state, the necessary information can be accurately extracted. it can.

Another invention includes a storage means, a processing means, an input means, and an output means.
The storage means has a necessary information database and a dialog state determination information database. As the necessary information database and the dialog state determination information database, those similar to the one invention described above can be used.
The processing means includes voice recognition means, requirement determination means, dialog state transition determination means, and management means. As the speech recognition means, the dialog state transition determination means, and the management means, the same speech recognition means, dialog state transition determination means, and management means as those of the one invention described above can be used. In the present invention, the business item discrimination means discriminates a conversational requirement between speakers based on the business item identification information input from the input unit. As a method for inputting the item identification information, for example, a method of operating an input key provided in the input unit or a method of selecting an input unit displayed on the input screen provided in the input unit is used. Can do. The requirement identification information is preferably input together with the audio signal, but may be input at a timing different from that of the audio signal.
In the present invention, since necessary information is extracted by determining that the dialog state has transitioned in the order of the information request dialog state, the information disclosure dialog state, and the information reception dialog state, the necessary information can be accurately extracted. it can. Further, in the present invention, since the requirement is determined based on the requirement identification information input from the input means, the requirement for dialogue between the speakers can be easily determined.

In one aspect of the invention or another aspect of the other invention, the request state clue information for determining that the request state is a request reception dialog state that has received a request for necessary information is stored in the dialog state determination information database. . Then, the dialog state transition determining means determines based on the voice recognition information that the dialog state between the speakers has transitioned in the order of the information request dialog state, the request reception dialog state, the information disclosure dialog state, and the information reception dialog state. As a result, each necessary information indicated in the necessary information list corresponding to the business determined by the business determining means is extracted. Information stored in the dialog state determination information database is the voice recognition information that the dialog state between the speakers transitions in the order of the information request dialog state, the request reception dialog state, the information disclosure dialog state, and the information reception dialog state. Determination is made by sequentially including request clue information, request acceptance clue information, information disclosure clue information, and information acceptance clue information.
In this mode, since necessary information is extracted by determining that the dialog state has transitioned in the order of information request dialog state, request reception dialog state, information disclosure dialog state, information reception dialog state, the necessary information is accurately Can be extracted well.

In another aspect of the invention, the storage means further includes a conversation unit discrimination information database and a necessary information clue information database, and the processing means further includes a conversation unit division means and a necessary information clue information. It has a giving means.
The conversation unit determination information database stores conversation unit determination information for dividing the speech recognition information into conversation units. “Conversation unit” means a range related to one topic. The conversation unit determination information database typically includes conversation unit determination information including conversation unit start point determination information for determining the start point of the conversation unit and conversation unit end point determination information for determining the end point of the conversation unit. Remembered. In the necessary information clue information database, a necessary information clue information list indicating necessary information clue information that is a clue for extracting necessary information is stored in correspondence with the requirements. Necessary information clue information can be appropriately set corresponding to the necessary information, but at least the necessary information clue information corresponding to the character string extracted as necessary information (typically, characters extracted as necessary information) Necessary information clue information) having the same part of speech as the column is set. For example, when the necessary information is [contractor name], at least necessary information clue information (necessary information clue information having a part of speech [person name]) corresponding to a character string indicating a person name is set. Further, when the necessary information is [contractor address], at least necessary information clue information (necessary information clue information having a part-of-speech [location name]) for the character string indicating the address is set. When the necessary information is [contractor telephone number], at least necessary information clue information (necessary information clue information having a part of speech [numerical number]) for a character string combining numbers is set. The necessary information clue information database and the necessary information database can be configured as one database. For example, necessary information is stored in correspondence with the requirements, and necessary information clue information is stored in correspondence with each necessary information.
The conversation unit dividing means divides the speech recognition information into conversation units based on the conversation unit determination information stored in the conversation unit determination information database. The conversation unit dividing means typically determines the start and end points of the conversation unit based on the speech recognition information and the conversation unit start point determination information and the conversation unit end point determination information stored in the conversation unit determination information database. A conversation unit is defined between the start point of the unit and the end point of the conversation unit following the start point of the conversation unit. For example, a location where the conversation unit start point determination information exists is set as the start point of the conversation unit, and a location where the conversation unit end point determination information exists is set as the end point of the conversation unit. If there is no conversation unit end point determination information between two conversation unit start point determination information, the point immediately before the point where the subsequent conversation unit start point determination information exists is set as the end point of the preceding conversation unit, and the subsequent A location where there is conversation unit start point determination information to be performed is set as the start point of the subsequent conversation unit. In addition, when there is no conversation unit start point determination information between two conversation unit end point determination information, the location where the preceding conversation unit end point determination information exists is set as the end point of the preceding conversation unit, and the preceding conversation unit end point The location immediately after the location where the discrimination information exists is set as the start point of the subsequent conversation unit.
The necessary information clue information giving means gives the necessary information clue information to the conversation unit divided by the conversation unit dividing means. Specifically, if necessary information clue information that is a clue to extract necessary information is included in the conversation unit, the necessary information clue information is given to the conversation unit.
The dialogue state transition determination unit selects a conversation unit including necessary information clue information that is a clue to extract necessary information corresponding to the requirement. As a method of selecting a conversation unit including necessary information clue information as a clue to extract necessary information, a method of selecting a conversation unit including one necessary information clue information corresponding to the necessary information, or a plurality of appropriately selected needs A method of selecting a conversation unit including information clue information can be used. Also, a method of selecting a conversation unit while appropriately changing necessary information clue information included in the conversation unit can be used. When selecting a conversation unit, at least necessary information clue information corresponding to a character string extracted as necessary information (typically, a necessary information clue having the same part of speech as the character string extracted as necessary information). Information unit). For example, when the necessary information is [contractor name], a conversation unit including a character string indicating a person name (a character string having a part of speech [person name]) is selected, and the necessary information is [contractor address]. In some cases, select a conversation unit that contains a character string that indicates an address (a character string that has a part of speech [place name]), and if the required information is [contractor telephone number], a combination of numbers. Select a conversation unit containing a string (a character string with part of speech [numerical]). Then, necessary information is extracted by determining the transition of the conversation state based on the speech recognition information included in the selected conversation unit.
In this embodiment, necessary information can be extracted with higher accuracy by dividing speech recognition information into conversation units. In addition, the necessary information can be efficiently extracted by giving the necessary information clue information to the conversation unit.

Still another invention is a storage medium storing a program or a program for causing a computer to execute the processing of the processing means described above.
By using the program or the storage medium of the present invention, the effects described above can be obtained.

In the audio signal processing device of the present invention, necessary information related to the requirement is extracted from the audio signal during the dialogue between the speakers related to the requirement by determining the transition of the dialogue state. Information can be extracted with high accuracy.
In the present invention,

It is a schematic block diagram of embodiment of this invention. It is a figure which shows an example of a conversation unit discrimination | determination information database. It is a figure which shows an example of a required information list. It is a figure which shows the other example of a required information list. It is a figure which shows the further another example of a required information list. It is a figure which shows an example of a required information clue information list. It is a figure which shows an example of a dialogue state discrimination | determination information database. It is a flowchart explaining operation | movement of embodiment of this invention. It is a figure which shows an example of the speech recognition information which recognized the speech signal at the time of the dialog between speakers. It is a figure which shows an example of the speech recognition information which recognized the speech signal at the time of the dialog between speakers. It is a figure which shows the outline | summary of the method of extracting required information by discriminating the transition of the dialogue state between speakers. It is an example of a method for extracting necessary information by discriminating a transition of a conversation state between speakers. It is another example in which necessary information is extracted by discriminating the transition of dialogue state between speakers. FIG. 10 shows still another example in which necessary information is extracted by discriminating a transition of a conversation state between speakers.

Embodiments of an audio signal processing device according to the present invention will be described below with reference to the drawings.
FIG. 1 shows a schematic configuration diagram of a first embodiment of an audio signal processing device of the present invention. The audio signal processing apparatus according to the first embodiment is requested from an audio signal at the time of dialogue between speakers when requesting a predetermined requirement from one speaker to the other speaker. The present invention can be suitably used as an audio signal processing apparatus that extracts necessary information necessary for performing a requirement. In this specification, “business” means content (for example, service content) requested from one speaker to the other speaker.
The audio signal processing apparatus according to the present embodiment includes processing means 10, storage means 20, input means 30, and output means 40.
As the input means 30, various known input means such as a microphone that converts sound into an audio signal that is an electric signal, a reading means that reads an audio signal stored in a storage medium, an input key, an input screen, and the like are used. it can. “Audio signal” means a signal output from a microphone or the like that converts audio into an electric signal. The “audio signal” includes a signal obtained by digitizing a signal output from a microphone or the like.
As the output means 40, various known output means such as a display means and a printing means can be used.

  The storage means 20 includes an audio signal database 21, a conversation unit determination information database 22, a necessary information database 23, a necessary information clue information database 24, and a dialog state determination information database 25.

  The voice signal database 21 stores voice signals that are input via the input means 30 and that are used when a speaker interacts with a message. In this embodiment, as will be described later, necessary information is extracted based on speech recognition information including text information created from a speech signal. Therefore, the voice recognition information database 26 can be used in place of the voice signal database 21 that stores the voice signal. In this case, the voice signal input through the input unit 30 is converted into voice recognition information by the voice recognition unit 12, and the converted voice recognition information is stored in the voice recognition information database 26. Alternatively, the voice signal input via the input means 30 is stored in the voice signal database 21, and the voice recognition information created from the voice signal stored in the voice signal database 21 is stored in the voice recognition information database 26. Also good. Alternatively, the voice recognition information input via the input unit 30 may be stored in the voice recognition information database 26. That is, it is only necessary that the voice recognition information created from the voice signal can be used.

The conversation unit determination information database 22 stores conversation unit determination information for dividing the speech recognition information into conversation units. “Conversation unit” means a range related to one topic.
FIG. 2 shows an example of the conversation unit discrimination information database 22. In the conversation unit determination information database 22 shown in FIG. 2, conversation unit start point determination information indicating the start point of the conversation unit and conversation unit end point determination information indicating the end point of the conversation unit are stored as the conversation unit determination information. . As the conversation unit start point determination information, an expression that is often used at the beginning of a conversation unit is used. For example, the expression “Moshimoshi” at the time of calling, the expression “Good morning” at the beginning greeting, the expressions “At”, “Now”, “Next”, etc. at the time of changing the topic are used. As the conversation unit end point determination information, an expression that is often used at the end of the conversation unit is used. For example, expressions of thanks “Thank you”, expressions of confirmation “Kashikomare” and “Accepted”, expressions of closing greetings “I ’m sorry”, etc. are used.

The necessary information database 23 stores necessary information extracted from speech signals (speech recognition information) during conversation between speakers. Necessary information to be extracted from a speech signal at the time of dialogue between speakers is often different depending on a request from one speaker to the other speaker. For this reason, in the necessary information database 23, a necessary information list indicating necessary information extracted from the voice signal at the time of dialogue between the speakers is stored corresponding to the requirements. Of course, when common necessary information is extracted for a plurality of requirements, a common necessary information list may be stored corresponding to the plurality of requirements.
FIGS. 3 to 5 show necessary information extracted from a voice signal at the time of a dialogue between a customer and a power company receptionist (operator) when a customer requests a service (service) from the customer. An example of the necessary information list is shown. 3 to 5 show a slot number and necessary information corresponding to the slot number (necessary information inserted into the slot having the slot number). The necessary information list shown in FIG. 3 indicates necessary information to be extracted from the audio signal when the customer requests [use of electricity] from the power company to the power company (needs: [use of electricity]). . The necessary information list shown in FIG. 4 indicates necessary information to be extracted from the audio signal when the customer requests the electric power company to [stop electricity] (necessary: [stop electricity use]). . The necessary information list shown in FIG. 5 is required to be extracted from the audio signal when the customer requests the power company to [Change Contracted Electricity Capacity] (Requirement: [Change Contracted Electricity Capacity]). Information is shown.

The necessary information clue information database 24 stores necessary information clue information that serves as a clue when extracting necessary information from voice signals (voice recognition information) during conversation between speakers. In the present embodiment, the necessary information clue information database 24 stores a necessary information clue information list indicating necessary information clue information corresponding to the necessary information corresponding to the requirement. Necessary information clue information can be set as appropriate corresponding to the necessary information (corresponding to the necessary information corresponding to the requirement), but at least the necessary information clue information corresponding to the character string extracted as the necessary information (Typically, necessary information clue information having the same part of speech as the character string extracted as necessary information) is set. For example, when the necessary information is [contractor name], at least necessary information clue information (necessary information clue information having a part of speech [person name]) corresponding to a character string indicating a person name is set. Further, when the necessary information is [contractor address], at least necessary information clue information (necessary information clue information having a part-of-speech [location name]) for the character string indicating the address is set. When the necessary information is [contractor telephone number], at least necessary information clue information (necessary information clue information having a part of speech [numerical number]) for a character string combining numbers is set.
FIG. 6 shows a necessary information clue information list indicating necessary information clue information used as a clue to extract necessary information from an audio signal when a customer requests [stop electricity] (requirement: [stop electricity use]). An example is shown. In FIG. 6, the necessary information clue information [part of speech: XX] has a part of speech type (product type information) of XX (for example, [person name] representing a person's name, [location name] representing an address) Represents a combination of numbers [numerical]), and other necessary information clue information represents a character string.
The necessary information clue information list shown in FIG. 6 includes, for example, necessary information clue information [contractor] that is a clue to extract necessary information [contractor name] corresponding to the necessary information [contractor name]. , [Customer], [name], [part of speech: personal name] are stored. In addition, necessary information [contractor address] corresponding to the necessary information [contractor address] is required information [contractor], [customer], [address], [part of speech: place name] It is remembered. Also, required information [contractor], [customer], [telephone], [mobile phone], necessary information [contractor phone number] corresponding to the necessary information [contractor phone number] [Number] and [Part of speech: Numeral] are stored.
Necessary information clue information is used in a process of selecting a conversation unit when extracting necessary information from a voice signal (voice recognition information) (details will be described later).

The dialog state determination information database 25 stores dialog state determination information for determining a dialog state between speakers. In the present embodiment, as the dialogue state discrimination information, information request clue information 25a for discriminating that it is a dialogue state in which required information is requested (referred to as “information request dialogue state”), and a dialogue in which a request for necessary information has been received. Request acceptance clue information 25b for determining that the state is a state (referred to as "request acceptance dialog state"), information disclosure for determining whether the state is a dialog state disclosing necessary information (referred to as "information disclosure dialog state") The clue information 25c and information acceptance clue information 25d for determining that the dialogue state has received the necessary information (referred to as “information acceptance dialogue state”) are stored.
FIG. 7 shows an example of the dialogue state determination information database 25. The dialogue state determination information database 25 shown in FIG. 7 stores, as information request clue information, an expression “[(required information)] please request disclosure of necessary information”, and the request acceptance clue. Information such as the expression “I understand” that the request for the necessary information has been accepted is stored, and the expressions “person name”, “place name”, “numerical”, etc. that represent the necessary information are stored as information disclosure clue information As the information acceptance clue information, expressions such as “Thank you” and “[(Necessary information)]” indicating that the disclosed necessary information has been received are stored.
Note that [(necessary information)] illustrated in FIG. 7 represents a character string including any one of necessary information extraction information (see FIG. 6) that is a key for extracting necessary information or the necessary information itself. For example, in a dialog requesting necessary information ([Information Request Dialog State]), character strings such as “customer number” and “address of moving address” are used, and a dialog to disclose necessary information ([Information Disclosure Dialogue] Status]), the customer number itself, the address of the new address, etc. are used.

The processing means 10 has a management means 11 and necessary information extraction means 12.
The management unit 11 manages the overall processing of the audio signal processing apparatus. For example, a process of inputting an audio signal via the input unit 30, a process of storing the input audio signal in the audio signal database 21, a process of operating the necessary information extracting unit 12 to extract necessary information from the audio signal, A process for outputting the extracted necessary information from the output means 40 is executed. The process of the management unit 11 can be executed by the necessary information extraction unit 12. In this case, the management means 11 can be omitted.

  The necessary information extracting unit 12 includes a voice recognition unit 13, a conversation unit dividing unit 14, a requirement determining unit 15, a necessary information clue information providing unit 16, and a dialog state transition determining unit 17. In the present embodiment, the processing of each unit is configured to be executed by a common processing device, but the processing of each unit may be configured to be executed by a separate processing device. Each means may be configured to be connected via a communication line such as a LAN, a telephone line, and an Internet line.

The voice recognition means 13 converts the voice signal at the time of dialogue between the speakers, which is input via the input means 30, into voice recognition information including text information. Various known voice recognition methods can be used as a method for converting the voice signal into voice recognition information. For example, a large vocabulary continuous speech recognition method using HMM (Hidden Markov Model) and N-gram (Stochastic Language Model) can be used.
Note that “the audio signal input via the input unit 30” is stored in the audio signal database 21 after being input via the input unit 30 as well as the audio signal itself input via the input unit 30. Also included are audio signals.
In the present embodiment, the speech recognition means 13 outputs speech recognition information to which part of speech information is assigned. For example, when a speech recognition result (speech recognition information) includes a character string having part of speech information (when part of speech information is included in the recognition vocabulary dictionary used by the speech recognition means 13), the part of speech information Is added to the character string. If the speech recognition result (speech recognition information) includes a character string that does not have part-of-speech information, the character string is obtained by performing morphological analysis on the text information included in the speech recognition result. Part of speech information is given to. The part-of-speech information given to the speech recognition information is used when necessary information clue information giving means 16 to be described later gives necessary information clue information to a conversation unit.
Further, the voice recognition means 13 for converting the voice signal into the voice recognition information may not be provided in the processing means 10. For example, the speech signal may be converted into speech recognition information by speech recognition means provided in a computer different from the computer provided with the processing means 10. In this case, speech recognition information output from speech recognition means provided in another computer is input via the input means 30 and stored in the speech recognition information database 26. In the configuration of the present invention, “the speech recognition means recognizes the speech signal at the time of the dialogue between the speakers inputted from the input means and outputs the speech recognition information”, the computer provided with the processing means 10 and Includes a configuration in which voice signals are converted into voice recognition information using voice recognition means provided in another computer, and the converted voice recognition information is input from the input means 30.

  The conversation unit dividing means 14 divides the speech recognition information into conversation units based on the conversation unit determination information stored in the conversation unit determination information database 22. In the present embodiment, the conversation unit determination information database 22 stores conversation unit determination information including conversation unit start point determination information and conversation unit end point determination information. Then, the conversation unit dividing means 14 discriminates the voice recognition information for each conversation unit by discriminating the start point and the end point of the conversation unit based on the conversation unit start point information and the conversation unit end point information stored in the conversation unit discrimination information database. It is divided. For example, a location where the conversation unit start point determination information exists is set as the start point of the conversation unit, and a location where the conversation unit end point determination information exists is set as the end point of the conversation unit. In normal conversation, the conversation unit start point determination information is issued without issuing the conversation unit end point determination information, or the conversation unit start point information is not issued after the conversation unit end point information is issued. May occur. For this reason, when there is no conversation unit end point determination information between two conversation unit start point determination information, the point immediately before the point where the subsequent conversation unit start point determination information exists is set as the end point of the preceding conversation unit, The location where the subsequent conversation unit start point determination information exists is set as the start point of the subsequent conversation unit. In addition, when there is no conversation unit start point determination information between two conversation unit end point determination information, the location where the preceding conversation unit end point determination information exists is set as the end point of the preceding conversation unit, and the preceding conversation unit end point The location immediately after the location where the discrimination information exists is set as the start point of the subsequent conversation unit. Note that when determining the start point and end point of the conversation unit, not only the conversation unit determination information itself but also the change mode of the conversation unit determination information is considered.

  The requirement discriminating means 15 discriminates a requirement for dialogue between speakers. In the present embodiment, the requirement discriminating means 15 discriminates a dialogue requirement between speakers based on the voice recognition information. As a method for discriminating a requirement based on voice recognition information, a method for discriminating which requirement is included in the requirement recognition information is used.

  The necessary information clue information giving means 16 gives necessary information clue information to the conversation units divided by the conversation unit dividing means 14. That is, when the necessary information clue information is included in the conversation unit, the necessary information clue information is given to the conversation unit. As the necessary information clue information, necessary information clue information shown in the necessary information clue information list stored in the necessary information clue information database 24 corresponding to the requirement is used.

  The dialog state transition determining means 17 extracts necessary information by determining the transition of the dialog state based on the voice recognition information. In the present embodiment, the dialog state transition determining unit 17 includes an information request dialog state determining unit 17a, a request receiving dialog state determining unit 17b, an information disclosure dialog state determining unit 17c, and an information receiving dialog state determining unit 17d. . Based on the voice recognition information and the information request clue information 25a stored in the dialog state determination information database 25, the information request dialog state determination unit 17a determines that the dialog state is the information request dialog state. Based on the voice recognition information and the request acceptance clue information 25b stored in the dialogue state discrimination information database 25, the request acceptance dialogue state discriminating means 17b discriminates that the dialogue state is the request acceptance dialogue state. Based on the voice recognition information and the information disclosure clue information 25c stored in the dialog state determination information database 25, the information disclosure dialog state determination unit 17c determines that the dialog state is the information disclosure dialog state. Based on the voice recognition information and the information reception clue information 25d stored in the dialog state determination information database 25, the information reception dialog state determination unit 17d determines that the dialog state is the information reception dialog state. Then, the dialog state transition determination unit 17 extracts necessary information by determining that the dialog state has transitioned in the order of the information request dialog state, the request reception dialog state, the information disclosure dialog state, and the information reception dialog state.

The dialogue state transition is determined by the dialogue state transition discriminating means 17 according to the present embodiment from the voice signal (voice recognition information corresponding to the voice signal) at the time of dialogue between the customer and the receptionist (operator) of the company service center. An outline of the determination method is shown in FIG. In the present embodiment, it is determined that the conversation state has transitioned in the order indicated by the solid line arrows.
First, the receptionist issues information request clue information to the customer. For example, the information request clue information “Please give me [(required information)]” is issued.
When the customer confirms the information requesting clue information issued from the acceptor, the customer issues the request accepting clue information to the acceptor. For example, the request acceptance clue information “I understand” is issued.
Next, the customer issues information disclosure clue information to the receptionist. For example, “[(Necessary information)]” is issued.
When the acceptor confirms the information disclosure clue information issued from the customer, the acceptor issues information acceptance clue information to the customer. For example, “Thank you.”
As a result, it is determined for [[necessary information]] that the dialog state has transitioned in the order of information request dialog state, request reception dialog state, information disclosure dialog state, information reception dialog state, and [(required information) ] Is extracted.

Here, in the present embodiment, a conversation unit dividing unit 14 and a necessary information clue information adding unit 16 are provided in order to increase the accuracy of extracting necessary information and reduce the processing load.
In other words, the speech recognition information is divided into conversation units by the conversation unit dividing means 14. Further, when the necessary information clue information adding means 16 includes the necessary information clue information in the conversation unit, the necessary information clue information is added to the conversation unit.
When the dialog state transition determining unit 17 extracts the necessary information indicated in the necessary information list corresponding to the business determined by the business determining unit 15, the necessary information clue that is a key for extracting the necessary information. Select the conversation unit that contains the information. As a method of selecting a conversation unit including necessary information clue information as a clue to extract necessary information, a method of selecting a conversation unit including one necessary information clue information corresponding to the necessary information, or a plurality of appropriately selected needs A method of selecting a conversation unit including information clue information can be used. Also, a method of selecting a conversation unit while appropriately changing necessary information clue information included in the conversation unit can be used. When selecting a conversation unit, at least necessary information clue information corresponding to a character string extracted as necessary information (typically, a necessary information clue having the same part of speech as the character string extracted as necessary information). Information unit). For example, when the necessary information is [contractor name], at least a conversation unit including a character string indicating a person name (a character string having a part of speech [person name]) is selected, and the necessary information is [contractor address]. ], At least when a conversation unit including a character string indicating an address (a character string having a part of speech [place name]) is selected and the necessary information is [contractor telephone number], at least , A conversation unit including a character string (a character string having a part of speech [numerical]) combined with numbers is selected. Then, necessary information is extracted by determining the transition of the conversation state with respect to the selected conversation unit (or a plurality of conversation units adjacent to the selected conversation unit including the selected conversation unit).
For example, when the necessary information [contractor name] is indicated by the necessary information list corresponding to the requirement, the necessary information clue information giving means 16 includes “contractor”, “customer”, [name] for each conversation. Or the [part of speech: personal name], the necessary information clue information [contractor], [customer], [name], [ Part of speech: Person name] is assigned to the conversation unit. When the dialogue state transition discriminating means 17 extracts necessary information [contractor name], necessary information clue information [contractor], [customer], [name], and [part of speech: personal name] are given ( Necessary information Select the conversation unit (including the "contractor", "customer", "name", and [part of speech: person name]). Then, necessary information [contractor name] is extracted by determining the transition of the conversation state based on the voice recognition information included in the selected conversation unit. In this case, [part of speech: personal name] is extracted. Thereby, required information can be extracted accurately and efficiently. The necessary information clue information adding means 16 adds the part of speech information given to the voice recognition information by the voice recognition means 13 and the necessary information database when giving the necessary information clue information to the conversation unit based on the part of speech information. The part of speech information of the necessary information clue information shown in the necessary information clue information list stored in 23 is compared and necessary information clue information is given.

Next, the operation of the audio signal processing apparatus according to the present embodiment will be described with reference to the flowchart shown in FIG.
In step A1, an audio signal during dialogue between the speakers is input via the input unit 30. The process of step A1 is executed by the management means 11.
In step A2, the voice signal input in step A1 is converted into voice recognition information including text information. Note that the audio signal input through the input unit 30 includes the audio signal stored in the audio signal database 21 after being input through the input unit 30. The process of step A2 is executed by the voice recognition means 13.
In step A3, the speech recognition information created in step A2 is divided into conversation units. The voice recognition information includes voice recognition information stored in the voice recognition information database 26 after being converted from the voice signal. The process of step A3 is executed by the conversation unit dividing means 14.
In step A4, a dialogue requirement between speakers is determined. In the present embodiment, the dialogue requirements between the speakers are determined based on the speech recognition information created in step A2. The process of step A4 is executed by the business matter discriminating means 15.
In step A5, necessary information clue information is given to the conversation unit divided in step A3. In the present embodiment, it is determined whether or not necessary information clue information that is a clue to extract necessary information is included in the conversation unit. If necessary information clue information is included, the necessary information clue information is Granted to the conversation unit. The necessary information is indicated by a necessary information list stored in the necessary information database 24 corresponding to the requirement determined in step A4. The necessary information clue information is shown in the necessary information clue information database 24 corresponding to the necessary information in the necessary information clue information list stored in correspondence with the requirement determined in step A4. The process of step A5 is executed by the necessary information clue information adding means 16.

In step A6, the necessary information list corresponding to the message determined in step A4 stored in the necessary information database 23 is selected.
In step A7, the slot number N is set to “1”.
In step A8, one of the conversation units including necessary information clue information that is a clue to extract necessary information inserted into the slot of slot number N is selected. In step A8, at least a conversation unit including necessary information clue information corresponding to a character string extracted as necessary information is selected. For example, when the necessary information is [contractor name] which is the name of a customer, a conversation unit including at least necessary information clue information [part of speech: person name] indicating a person's name is selected. Preferably, a conversation unit including necessary information clue information [part of speech: personal name], [contractor] (or [customer]), and [name] is selected. Further, when the necessary information is [contractor address] which is a customer address, a conversation unit including at least necessary information clue information [part of speech: place name] indicating the address is selected. Preferably, a conversation unit including necessary information clue information [part of speech: place name], [contractor] (or [customer]), and [address] is selected. When the necessary information is [contractor telephone number], which is the customer's telephone number, a conversation unit including at least necessary information clue information [part of speech: number] indicating a combination of numbers is selected. Preferably, a conversation unit including necessary information clue information [part of speech: number], [contractor] (or [customer]), [phone] (or [mobile phone]), and [number] is selected.
In step A8, as a method for selecting at least a conversation unit including necessary information clue information corresponding to a character string extracted as necessary information, only necessary information clue information corresponding to a character string extracted as necessary information is included. Includes one or more of a method for selecting a conversation unit, necessary information clue information corresponding to a character string extracted as necessary information, and other necessary information clue information indicated in the necessary information clue information list A method for selecting a conversation unit or the like can be used in appropriate combination.

In step A9, information request clue information 25a corresponding to the necessary information inserted in the slot number N of the necessary information list selected in step A6 exists in the speech recognition information included in the conversation unit selected in step A8. It is determined whether or not it is an information request dialogue state regarding necessary information. When it is determined that the information request clue information 25a exists, that is, when it is determined that the information request dialogue state is established, the process proceeds to step A10. When the information request clue information 25a does not exist, that is, when it is not determined that the information request dialogue state is established, necessary information is extracted from the speech recognition information included in the conversation unit selected in step A8. The process proceeds to step A16.
In step A10, whether or not the request recognition clue information 25b corresponding to the necessary information exists in the voice recognition information included in the conversation unit selected in step A8, following the information request clue information 25a corresponding to the necessary information. In other words, it is determined whether or not a transition is made from the information request dialogue state relating to necessary information to the request acceptance dialogue state relating to necessary information. If the information request dialog state has transitioned to the request acceptance dialog state, the process proceeds to step A11, and if not, the process proceeds to step A15.
In step A11, the speech recognition information included in the conversation unit selected in step A8 is followed by information request clue information 25a corresponding to necessary information, request acceptance clue information 25b corresponding to necessary information, and further necessary. Whether there is information disclosure clue information 25c corresponding to the information, that is, whether the information request dialogue state relating to the necessary information has changed to the request acceptance dialogue state relating to the necessary information, and further to the information disclosure dialogue state relating to the necessary information Is determined. When the information request dialogue state is changed to the request acceptance dialogue state and further to the information disclosure dialogue state, the process proceeds to Step A12, and otherwise, the process proceeds to Step A15.
In step A12, the speech recognition information included in the conversation unit selected in step A8 is added to the request acceptance clue information 25b corresponding to the necessary information and the necessary information following the information request clue information 25a corresponding to the necessary information. Whether the corresponding information disclosure clue information 25c and the information acceptance clue information corresponding to the necessary information exist, that is, from the information request dialogue state regarding the necessary information, the request acceptance dialogue state regarding the necessary information, and the information disclosure regarding the necessary information It is determined whether or not a transition is made to a dialog state and further to an information reception dialog state regarding necessary information. When the information request dialogue state is changed to the request acceptance dialogue state, the information disclosure dialogue state, and further to the information acceptance dialogue state, the process proceeds to Step A13, and otherwise, the process proceeds to Step A15.

In step A13, the necessary information is extracted, and the extracted necessary information is inserted into the slot corresponding to the slot number N in the necessary information list selected in step A6. For example, when the necessary information is [contractor name], a character string corresponding to the necessary information clue information [part of speech: person name] indicating the person's name is extracted. When the necessary information is [contractor address], a character string corresponding to the necessary information clue information [part of speech: place name] indicating the address is extracted. If the necessary information is [contractor telephone number], a character string corresponding to the necessary information clue information [part of speech: number] indicating a combination of numbers is extracted.
In step A14, it is determined whether N is equal to the total number of slots in the necessary information list selected in step A6. If N is equal to the total number of slots in the necessary information list, the process is terminated; otherwise, the process proceeds to step A17.

In step A15, all the information request clue information 25a corresponding to the necessary information to be inserted into the slot number N in the necessary information list selected in step A6 is determined from the speech recognition information included in the conversation unit selected in step A8. It is determined whether or not. If all the information request clue information 25a is determined, the process proceeds to step A16. If all the information request clue information 25a has not been determined, the process returns to step A9 to determine the information request clue information 25a existing in the remaining speech recognition information.
In step A16, it is determined whether or not all the conversation units including the necessary information clue information to be used as a clue to extract the necessary information inserted into the slot of slot number N in the necessary information list selected in step A6 have been selected. If not all the conversation units including necessary information clue information that is a clue to extract necessary information to be inserted into the slot with slot number N have been selected, it is necessary to return to step A8 and insert into the slot with slot number N. Other conversation units including necessary information clue information as a clue to extract information are selected. When all conversation units including necessary information clue information that is a clue to extract necessary information to be inserted into the slot of slot number N are selected, the process proceeds to step A14.
In step A17, “1” is added to the slot number N, and then the process returns to step A8 to extract necessary information to be inserted into the slot of the next slot number N (= N + 1) in the necessary information list selected in step A6. Repeat the process.
The processing in steps A6 to A17 is executed by the dialog state transition determination unit 17.

In the first embodiment, necessary information is extracted by determining that the dialog state has transitioned in the order of the information request dialog state, the request reception dialog state, the information disclosure dialog state, and the information reception dialog state. Here, in a normal dialogue, the information request dialogue state may change to the information disclosure dialogue state. That is, there may be no request acceptance dialog state.
In the second embodiment of the audio signal processing apparatus of the present invention, the necessary information is extracted by determining that the dialogue state has transitioned in the order of the information request dialogue state, the information disclosure dialogue state, and the information reception dialogue state. A form is demonstrated. Similarly to the audio signal processing apparatus of the first embodiment, the audio signal processing apparatus of the second embodiment also has dialogue between speakers when requesting a request from one speaker to the other speaker. The present invention can be suitably used as an audio signal processing apparatus that extracts necessary information necessary for performing a requested request from an audio signal at the time.
The audio signal processing apparatus according to the second embodiment includes a processing means 10, a storage means 20, an input means 30, and an output means 40, similarly to the audio signal processing apparatus according to the first embodiment. However, in the second embodiment, the necessary information is extracted by determining that the dialog state has transitioned in the order of the information request dialog state, the information disclosure dialog state, and the information reception dialog state. The request acceptance clue information 25b of the dialog state determination information database 25 and the request reception dialog state determination unit 17b of the dialog state transition determination unit 17 are omitted. Other configurations are the same as those of the audio signal processing apparatus according to the first embodiment.

From the voice signal (voice recognition information corresponding to the voice signal) at the time of dialogue between the customer and the receptionist (operator) of the company service center, the dialogue state transition discriminating means 17 of the second embodiment An overview of the method for discriminating fibers is shown in FIG. In the second embodiment, it is determined that the conversation state has transitioned in the order indicated by the dashed arrows.
First, the receptionist issues information request clue information to the customer. For example, the information request clue information “Please give me [(required information)]” is issued.
When the customer confirms the information requesting clue information issued from the acceptor, the customer issues information disclosure clue information to the acceptor. For example, “[(Necessary information)]” is issued.
When the acceptor confirms the information disclosure clue information issued from the customer, the acceptor issues information acceptance clue information to the customer. For example, “Thank you.”
As a result, it is determined that the dialogue state has transitioned in the order of the information request dialogue state, the information disclosure dialogue state, and the information reception dialogue state with respect to [(necessary information)], and [(necessary information)] is extracted.

Next, the operation of the audio signal processing apparatus according to the second embodiment will be described with reference to the flowchart shown in FIG. In the audio signal processing apparatus according to the second embodiment, step A10 is omitted because it is not determined that the dialogue state is the request acceptance dialogue state.
That is, in step A9, when the information request clue information 25a is determined from the voice recognition information included in the selected conversation unit, the process proceeds to step A11 as indicated by the broken line arrow.
Other processes are the same as those of the audio signal processing apparatus according to the first embodiment.

Next, the operation of the present embodiment will be specifically described for a case where a customer requests an electric power company to stop using electricity.
FIG. 9 and FIG. 10 show a voice signal including text information as a voice signal at the time of a dialogue between a customer and a receptionist (operator) of a service center of an electric power company when requesting [use of electricity]. An example is shown in which documents C1 to C49 are created after conversion into recognition information.

First, the dialogue requirements are determined based on the voice recognition information. In this case, for example, the character string “stop” included in the document C3 “I want to stop the electricity” is a necessary information clue information list corresponding to the requirement [stop electricity use] (FIG. 6). Since it matches the necessary information clue information [stop] in the reference), it is determined that the requirement for the conversation (contents of service requested from the customer to the power company) is [stop use of electricity]. Various methods other than this can be used as a method of discriminating the dialogue requirements from the speech recognition information.
Next, the speech recognition information (speech recognition information included in the documents C1 to C49) is stored in the conversation unit end point determination information and the conversation unit end point determination place stored as the conversation unit determination information in the conversation unit determination information database 22. Based on the conversation units S1 to S15. In FIG. 9 and FIG. 10, the speech recognition information surrounded by a solid line is conversation unit start point determination information, and the voice recognition information surrounded by a broken line is conversation unit end point determination information. For example, “Good morning” in document C1 (expression of the first greeting), “Now” in document C6, “In” in document C8, “Next” in document C11 (expression to change the topic), etc. Discriminated as discriminating information, “Cumshot” in document C5 (expression of confirmation), “Thank you” in document C10 (expression of thanks), “I am sorry” (expression of greeting at the end) in document C48, etc. Is determined as the conversation end point determination information.

In addition, if the divided conversation unit includes necessary information clue information that serves as a clue to extract necessary information indicated in the necessary information list corresponding to the determined requirement, the necessary information clue information is included in the conversation. It is given to the unit. For example, when the requirement is [use of electricity], a clue to extract necessary information shown in the necessary information list corresponding to the requirement [use of electricity] shown in FIG. Necessary information clue information is given. FIG. 5 shows an example of necessary information clue information corresponding to necessary information in the requirement [stop using electricity].
In FIG. 9, for example, necessary information clue information [reception center] corresponding to necessary information [recipient name] based on the character strings “XX reception center” and “ΔΔ” included in the conversation unit S1, [Part of speech: personal name] is assigned to the conversation unit S1. Further, based on the character strings “stop electricity” and “stop electricity”, the necessary information clue information [stop] and [stop] corresponding to the necessary information [use stop date] and [use stop time] are conversation units S1. Has been granted. Further, necessary information clue information [customer number], [part of speech] used as a clue to extract necessary information [customer number] based on the character strings “customer number” and “00 11 22 33” included in the conversation unit S3. : Numeral] is assigned to the conversation unit S3. Further, based on the character strings “contractor”, “name”, and “□□□□” included in the conversation unit S4, necessary information clue information [contract] that is a clue to extract necessary information [contractor name]. [Person], [name], [part of speech: personal name] are assigned to the conversation unit S4.
In FIG. 10, for example, necessary information [usage stop date], [use date] based on the character strings “X month”, “Y day”, “Tuesday”, “12:00” included in the conversation unit S8. Necessary information clue information “Month”, “Day”, “Time”, “Day”, and [Part of speech: Numeral], which are clues for extracting [stop time], are assigned to the conversation unit S8. Further, based on the character strings “X month”, “Y day”, “Tuesday”, “12:00”, “stop” included in the conversation unit S9, necessary information [usage stop date], [use stop time] Necessary information clue information “month”, “day”, “hour”, “day of the week”, “stop”, and “part of speech: number” are assigned to the conversation unit S9. Further, based on the character strings “moving”, “address”, and “DD city EE4-5-6 FF apartment 202” included in the conversation unit S12, a clue for extracting necessary information [contractor relocation address] Necessary information clue information [moving], [address], [part of speech: place name] is given.

Then, by determining the transition of the dialogue state based on the voice recognition information, the necessary information indicated in the necessary information list corresponding to the determined requirement is extracted. For example, it is determined that the information request dialogue state, the request acceptance dialogue state, the information disclosure dialogue state, the information acceptance dialogue state are changed in this order, or the information request dialogue state, the information disclosure dialogue state, the information acceptance dialogue state are changed in this order. To extract necessary information. At this time, a conversation unit including necessary information clue information that is a clue to extract necessary information is selected.
A specific example of processing for extracting necessary information by determining the transition of the dialogue state will be described with reference to FIGS. 12 and 13 are examples of determining that the dialog state has transitioned in the order of the information request dialog state, the information disclosure dialog state, and the information reception dialog state. FIG. 14 illustrates the dialog state as the information request dialog state, It is an example which discriminate | determines that it changed in order of a request reception dialogue state, an information disclosure dialogue state, and an information reception dialogue state.

FIG. 12 shows a process of extracting necessary information [customer number] for the dialogue requirement [electricity stoppage]. In FIG. 12, necessary information clue information [customer number], [part of speech: numerology] as a conversation unit including necessary information clue information for extracting necessary information [customer number] of slot number B002 shown in FIG. ] Has been selected.
The document C8 indicating the content of the dialogue from the reception person to the customer includes an expression “Please enter the customer number from the left” indicating the request for the necessary information [customer number]. As a result, it is determined from the document C8 that the information request dialogue state has requested the necessary information [customer number].
The document C9 indicating the content of the dialogue from the customer to the receiver subsequent to the document C8 includes a character string “00 11 22 33” corresponding to the character string indicating the necessary information [customer number]. As a result, it is determined from the document C9 that the information disclosure dialogue state discloses the necessary information [customer number].
The document C10 indicating the dialogue contents from the receiver to the customer following the document C9 includes an expression “00 11 22 33, thank you” indicating confirmation of the necessary information [customer number]. Thus, it is determined from the document C10 that the information reception dialogue state in which the necessary information [customer number] is received is established.
As described above, regarding the necessary information [customer number], in the conversation unit S3, it is determined that the dialogue state has transitioned in the order of the information request dialogue state, the information disclosure dialogue state, and the information reception dialogue state. The character string “00112233” corresponding to the necessary information clue information [part of speech: number] indicating the necessary information [customer number] is extracted.

FIG. 13 shows a process of extracting necessary information [contractor name] for the dialogue requirement [electricity stoppage]. In FIG. 13, necessary information clue information [contractor], [name] is used as a conversation unit including necessary information clue information that is a clue to extract the necessary information [contractor name] of the slot number B003 shown in FIG. The conversation unit S4 to which [part of speech: personal name] is assigned is selected.
The document C11 indicating the content of the dialogue from the acceptor to the customer includes the expression “Please give the name of the contractor” indicating the request for the necessary information [contractor name]. As a result, it is determined from the document C11 that the information request dialogue state has requested the necessary information [contractor name].
The document C12 indicating the content of the dialogue from the customer to the receiver subsequent to the document C11 includes a character string “□□□□” corresponding to the character string indicating the necessary information [contractor name]. Thus, it is determined from the document C12 that the information disclosure dialogue state is disclosed in which the necessary information [contractor name] is disclosed.
The document C13 showing the content of the dialogue from the receiver to the customer following the document C12 includes the expression “Thank you, □□□□,” indicating confirmation of the necessary information [contractor name]. . Thus, it is determined from the document C13 that the information reception dialogue state in which the necessary information [contractor name] is received is established.
As described above, regarding the necessary information [contractor name], it is determined in the conversation unit S4 that the dialog state has transitioned in the order of the information request dialog state, the information disclosure dialog state, and the information reception dialog state. As the “name”, the character string “□□□□” corresponding to the necessary information clue information [part of speech: personal name] indicating the necessary information [contractor name] is extracted.

FIG. 14 shows a process of extracting necessary information [contractor's transfer destination telephone number] for the dialogue requirement [electricity stoppage]. In FIG. 14, the necessary information clue information [moving], [moving], [conversion unit including necessary information clue information that is a clue to extract the necessary information [contractor transfer destination telephone number] of the slot number B011 shown in FIG. The conversation unit S13 to which "telephone" (or [mobile phone]), [number], and [part of speech: number] are assigned is selected.
The document C37 indicating the content of the dialogue from the reception person to the customer includes an expression “Please give me the phone number of the destination” indicating the request for the necessary information [contractor transfer destination phone number]. As a result, it is determined from the document C37 that the information request dialogue state in which the required information [contractor transfer destination telephone number] is requested.
The document C38 indicating the content of the dialogue from the customer to the receiver subsequent to the document C37 includes an expression “I understand” indicating that the request for the necessary information [contractor transfer destination telephone number] has been received. . As a result, it is determined from the document C38 that the request acceptance dialogue state in which the request for the necessary information [contractor transfer destination telephone number] has been accepted is made.
The document C41 indicating the content of the dialogue from the customer to the receiver subsequent to the document C39 includes a character string “090 0000 1111” corresponding to the character string indicating the necessary information [contractor transfer destination telephone number]. . Accordingly, it is determined from the document C41 that the information disclosure dialogue state is disclosed in which the necessary information [contractor transfer destination telephone number] is disclosed.
The document C42 indicating the content of the dialogue from the receiver to the customer following the document C41 includes an expression “090 0000 1111, thank you” indicating confirmation of necessary information [contractor's transfer destination telephone number]. ing. Thus, it is determined from the document C42 that the information reception dialogue state in which the necessary information [contractor transfer destination telephone number] has been received is determined.
As described above, regarding the necessary information [contractor transfer destination telephone number], it is determined in the conversation unit S13 that the dialog state has changed in the order of the information request dialog state, the request reception dialog state, the information disclosure dialog state, and the information reception dialog state. As the necessary information [contractor relocation destination telephone number], the character string “090001111” corresponding to the necessary information clue information [part of speech: number] indicating the necessary information [contractor relocation destination telephone number] is extracted.

  For example, in the conversation unit S14, the necessary information [usage date], [usage time], [contractor name], [contractor address] and necessary information clue information [month], [day], [Time], [Day of the week], [Stop], [Part of speech: Numeral], [Part of speech: Place name] are given, but the conversation state transition described above does not occur in the conversation unit S14. For this reason, the dialog state transition determining means 17 cannot determine the above-described dialog state transition from the speech recognition information included in the conversation unit S14. That is, necessary information is not extracted from the conversation unit S14.

  As described above, in the above-described embodiment, the dialogue state is the information request dialogue state regarding the necessary information for the requirement from the voice signal (voice recognition information created from the voice signal) during the dialogue between the speakers. By determining that the request acceptance dialogue state, the information disclosure dialogue state, and the information acceptance dialogue state have been changed in order, or by determining that the information request dialogue state, the information disclosure dialogue state, and the information acceptance dialogue state have been changed in this order. Necessary information is extracted by. As a result, information necessary for the conversation between the speakers is obtained from the voice signal during the conversation between the speakers (for example, necessary to execute the request requested from one speaker to the other speaker). Information) can be extracted with high accuracy. In particular, in the present embodiment described above, the necessary information is extracted by dividing the speech recognition information into conversation units and discriminating the transition of the conversation state within the conversation units (tracing the conversation progress). Information can be detected more accurately. In addition, if the conversation unit includes necessary information clue information that is a clue to extract necessary information, the necessary information clue information is assigned to the conversation unit and the necessary information is extracted when the necessary information is extracted. Necessary information used as a clue to be extracted Since the conversation unit containing the clue information is selected, the necessary information can be extracted more accurately and the processing load for extracting the necessary information can be reduced. Can do.

  In the present invention, the processing means described above (management means constituting the processing means, voice recognition means, conversation unit dividing means, requirement determining means, necessary information clue information providing means, dialog state transition determining means, etc.) It is also possible to configure a program for executing the process or a storage medium storing the program.

The present invention is not limited to the configuration of the embodiment described above, and various changes, additions, and deletions are possible.
Although the requirement determination unit 15 determines the requirement for dialogue based on the voice recognition information, various other methods can be used as a method for determining the requirement for dialogue. For example, by operating an input key provided on the input means 30 or by selecting an input part of a display screen provided on the input means 30, a condition indicating a condition is input from the input means 30. A method of inputting identification information can be used. In this case, in step A <b> 4 shown in FIG. 8, the matter determination unit 15 determines a conversation item based on the item identification information input via the input unit 30. By using such a requirement discriminating means 15, processing for discriminating a requirement is facilitated.
The conversation unit discrimination information used for discriminating the conversation unit can be set as appropriate according to the dialogue requirements.
Necessary information to be extracted from the audio signal can be set as appropriate according to the dialogue requirements.
Necessary information clue information as a clue to extract necessary information can be set as appropriate according to the necessary information.
The speech recognition information is divided into conversation units, and the necessary information clue information that is a clue to extract necessary information is assigned to the conversation unit, but the process of dividing into conversation units and the process of giving necessary information clue information to the conversation unit Can be omitted. Even in this case, necessary information can be extracted with high accuracy because necessary information is extracted by discriminating the transition of the dialogue state.
Dialog state determination information used to determine the dialog state can be set as appropriate according to the requirements of the dialog.
The conversation requirement can be set as appropriate according to the content of the requirement requested from one speaker to the other speaker.
As the input means 30 and the output means 40, an input means and an output means provided in a remote terminal device that can be connected to the processing means 10 via a communication line can be used.

DESCRIPTION OF SYMBOLS 10 Processing means 11 Management means 12 Necessary information extraction means 13 Speech recognition means 14 Conversation unit division means 15 Requirements determination means 16 Necessary information clue information addition means 17 Dialog state transition determination means 17a Information request dialog state determination means 17b Request acceptance dialog state Discriminating unit 17c Information disclosure dialogue state discriminating unit 17d Information receiving dialogue state discriminating unit 20 Storage unit 21 Audio signal database 22 Conversation unit discriminating information database 23 Necessary information database 24 Necessary information clue information database 25 Dialogue state discriminating information database 25a Information request clue information 25b Request acceptance clue information 25c Information disclosure clue information 25d Information acceptance clue information 26 Speech recognition information database 30 Input means 40 Output means

Claims (6)

A speech signal processing apparatus that extracts necessary information related to the requirement from a speech signal at the time of dialogue between speakers for the requirement,
A storage unit, a processing unit, an input unit, and an output unit;
The storage means includes a necessary information database and a dialog state determination information database,
In the necessary information database, a necessary information list indicating necessary information relating to the requirement is stored corresponding to the requirement,
In the dialog state determination information database, information request clue information for determining whether the dialog state requested the necessary information, information disclosure clue information for determining the dialog state disclosing the necessary information, and necessary information Information reception clue information for determining whether the conversation is accepted is stored.
The processing means includes voice recognition means, requirement determination means, dialogue state transition determination means, and management means,
The speech recognition means recognizes a speech signal at the time of dialogue between speakers inputted from the input means and outputs speech recognition information.
The requirement determining means determines a requirement for dialogue between speakers based on the speech recognition information output from the speech recognition means,
The dialog state transition determination means is based on the voice recognition information output from the voice recognition means, information request clue information, information disclosure clue information, and information acceptance clue information stored in the dialog state determination information database. By determining that the conversation state between the speakers has transitioned in the order of the conversation state in which the necessary information is requested, the conversation state in which the necessary information is disclosed, and the conversation state in which the necessary information has been received, it is determined by the requirement determining unit. The necessary information shown in the necessary information list stored in the necessary information database corresponding to the requirements is extracted,
The audio signal processing apparatus, wherein the management unit outputs the necessary information extracted by the dialog state transition determination unit from the output unit. A speech signal processing apparatus that extracts necessary information related to the requirement from a speech signal at the time of dialogue between speakers for the requirement,
A storage unit, a processing unit, an input unit, and an output unit;
The storage means includes a necessary information database and a dialog state determination information database,
In the necessary information database, a necessary information list indicating necessary information relating to the requirement is stored corresponding to the requirement,
In the dialog state determination information database, information request clue information for determining whether the dialog state requested the necessary information, information disclosure clue information for determining the dialog state disclosing the necessary information, and necessary information Information reception clue information for determining whether the conversation is accepted is stored.
The processing means includes voice recognition means, requirement determination means, dialogue state transition determination means, and management means,
The speech recognition means recognizes a speech signal at the time of dialogue between speakers inputted from the input means and outputs speech recognition information.
The message determination means determines a message for dialogue between the speakers based on the message identification information input from the input means,
The dialog state transition determination means is based on the voice recognition information output from the voice recognition means, information request clue information, information disclosure clue information, and information acceptance clue information stored in the dialog state determination information database. By determining that the conversation state between the speakers has transitioned in the order of the conversation state in which the necessary information is requested, the conversation state in which the necessary information is disclosed, and the conversation state in which the necessary information has been received, it is determined by the requirement determining unit. The necessary information shown in the necessary information list stored in the necessary information database corresponding to the requirements is extracted,
The audio signal processing apparatus, wherein the management unit outputs the necessary information extracted by the dialog state transition determination unit from the output unit. The audio signal processing device according to claim 1 or 2,
The dialog state determination information database further stores request acceptance clue information for determining that the required state information is in a dialog state.
The dialog state transition determination means includes voice recognition information output from the voice recognition means, information request clue information, request acceptance clue information, information disclosure clue information, and information acceptance clue stored in the dialog state determination information database. Based on the information, the conversation state between the speakers transitioned in the order of the conversation state that requested the necessary information, the conversation state that accepted the request for the necessary information, the conversation state that disclosed the necessary information, and the conversation state that accepted the necessary information. And extracting necessary information indicated by a necessary information list stored in the necessary information database corresponding to the business determined by the business determination unit based on the determination result An audio signal processing device. The audio signal processing device according to any one of claims 1 to 3,
The storage means further includes a conversation unit determination information database and a necessary information clue information database,
In the conversation unit determination information database, conversation unit determination information for dividing voice recognition information into conversation units is stored,
In the necessary information clue information database, a necessary information clue information list indicating necessary information clue information serving as a clue for extracting necessary information is stored in correspondence with requirements.
The processing means further includes a conversation unit dividing means and a necessary information clue information giving means,
The conversation unit dividing unit divides the speech recognition information output from the speech recognition unit into conversation units based on the conversation unit determination information stored in the conversation unit determination information database.
The necessary information clue information providing means is the necessary information stored in the necessary information clue information database corresponding to the message determined by the message determining means for the conversation unit divided by the conversation unit dividing means. When necessary information clue information shown in the clue information list is included, the necessary information clue information is given to the conversation unit,
The dialog state transition determining means selects a conversation unit including necessary information clue information that is a clue to extract necessary information, and extracts the necessary information based on voice recognition information included in the selected conversation unit. An audio signal processing device.   The program for making a computer perform the process of the said process means in any one of Claims 1-4.   A storage medium in which a program for causing a computer to execute the processing of the processing means according to claim 1 is recorded.

Images