JP4749438B2 - Phonetic character conversion device, phonetic character conversion method, and phonetic character conversion program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To improve the accuracy of voice recognition. <P>SOLUTION: A voice conversion part 121 converts two voice information pieces showing the same information input by two persons to generate two character information pieces. A character information comparing part 122 compares two character information pieces to extract a mismatch part. A mismatch part deciding part 123 decides which character is adopted depending on which character of the mismatch part is correctly converted from the voice information to the character information. A character information generating pat 124 generates character information corresponding to two voice information pieces by replacing a mismatch part of one of two character information pieces with a character decided to be adopted. <P>COPYRIGHT: (C)2010,JPO&amp;INPIT

Description

  The present invention relates to a technique for converting voice information into character information, for example.

2. Description of the Related Art Conventionally, there is a voice character conversion device (voice recognition device) that converts input voice information into character information and outputs it when voice information is input. Also, there is an apparatus that identifies a user by comparing a pattern of input voice information with a feature pattern of a user's voice (see Patent Document 1).
Moreover, there is an example in which the above-mentioned phonetic character conversion device or a device for specifying a user is applied to a call center system (see Patent Documents 2 and 3).
JP 2002-279245 A JP 2006-126966 A JP 2002-9955 A

Conventional speech recognition technology has low speech recognition accuracy. Therefore, for example, when creating an electronic document such as a contract, it is difficult to input each information (personal information, etc.) using voice information and convert the input voice information into character information to create a document.
An object of the present invention is to increase the accuracy of voice recognition, for example. Another object is to create an electronic document such as a contract by inputting voice information between an operator and a user, for example, at a call center.

The phonetic character conversion device according to the present invention is, for example,
A first voice information input unit for inputting the first voice information input by the first terminal and storing the first voice information in a storage device;
A second voice information input unit for inputting the second voice information input by the second terminal and storing the second voice information in a storage device;
The first voice information input by the first voice information input unit is converted into first character information by the processing device, and the second voice information input by the second voice information input unit is converted into the second character information. A voice conversion unit that converts the text information into a character information by a processing device;
A character information comparison unit that compares the first character information converted by the voice conversion unit with the second character information and extracts a mismatched portion by a processing device;
The processing device determines the character information of the inconsistent portion extracted by comparison by the character information comparing unit as the character information of the inconsistent portion of the first character information and the second character information by a predetermined method. A non-matching part determination unit to
A character information generation unit that replaces the mismatched portion of the first character information or the second character information with the character information determined by the mismatched portion determination unit, and generates character information by a processing device. And

The non-matching portion determination unit has a first accuracy indicating the accuracy with which the conversion from the first speech information to the first character information is correct, and the conversion from the second speech information into the second character information is correct. When the first accuracy is high, the character information of the inconsistent portion is determined as the first character information when compared with the second accuracy indicating the accuracy, and when the second accuracy is high. Is determined as the second character information.

The inconsistent part determination unit determines the character information of the inconsistent part for each of the characters included in the first character information and the second character information, either the first character information or the second character information. The character information is determined.

The phonetic character conversion device further includes:
For each user who uses the first terminal and the second terminal, a language model storage unit that stores a language model for converting speech information into character information in a storage device,
The speech conversion unit converts the first speech information into first character information based on a first language model stored by the language model storage unit for a first user who uses the first terminal. In addition to the conversion, the language model storage unit converts the second speech information into second character information based on the second language model stored for the second user who uses the second terminal. It is characterized by that.

The phonetic character conversion device further includes:
Based on the character information generated by the character information generation unit and the first voice information, the language model of the first user who uses the first terminal is updated by the processing device, and the character information generation unit A language model updating unit is provided that updates a language model of a second user who uses the second terminal by a processing device based on the generated character information and the second voice information.

The phonetic character conversion method according to the present invention is, for example,
A first audio information input step in which the processing device inputs the first audio information input by the first terminal;
A second audio information input step in which the processing device inputs the second audio information input by the second terminal;
The processing device converts the first voice information input in the first voice information input step into first character information, and also converts the second voice information input in the second voice information input step into the second voice information. A voice conversion step to convert the text information into
A character information comparison step in which the processing device compares the first character information converted in the voice conversion step with the second character information and extracts a mismatched portion;
The processing device converts the character information of the inconsistent portion extracted by comparison in the character information comparing step into character information of the inconsistent portion of the first character information and the second character information by a predetermined method. A non-matching part determination step to be determined;
A processing apparatus comprising: a character information generation step of generating character information by replacing the mismatched portion of the first character information or the second character information with the character information determined in the mismatched portion determination step. Features.

The phonetic character conversion program according to the present invention is, for example,
A first audio information input process for inputting the first audio information input by the first terminal;
A second audio information input process for inputting the second audio information input by the second terminal;
The first voice information input in the first voice information input process is converted into first character information, and the second voice information input in the second voice information input process is converted into second character information. Voice conversion processing to convert,
A character information comparison process for comparing the first character information converted by the voice conversion process with the second character information and extracting a mismatched portion;
Unmatched portion for determining the character information of the mismatched portion extracted by comparison in the character information comparison process as the character information of the mismatched portion of the first character information and the second character information by a predetermined method The decision process,
Replacing the inconsistent portion of the first character information or the second character information with the character information determined in the inconsistent portion determination processing, and causing the computer to execute character information generation processing for generating character information. And

  According to the phonetic character conversion device according to the present invention, the conversion accuracy is high because the two voice information of the first voice information and the second voice information is converted into one character information.

Embodiment 1 FIG.
In this embodiment, a phonetic character conversion apparatus 100 having dictionary information for each input item will be described.

FIG. 1 is a conceptual diagram showing an outline of functions of the phonetic character conversion apparatus 100 according to this embodiment.
The user moves the cursor to a predetermined input item (input field) of the application 10 and inputs information by voice from an input device such as a microphone. For example, move the cursor to the amount input field and enter “100,000”. In this case, the speech character conversion device 100 acquires the attribute information of the input item with the cursor from the application 10. Here, “numerical attribute” is acquired as attribute information in the amount input field. In addition, the speech character conversion device 100 acquires speech information input by a user from an input device such as a microphone. Here, the audio information “Juman” indicating “100,000” is acquired. Then, the phonetic character conversion device 100 determines a dictionary to be used based on the attribute information acquired from the application 10, converts the phonetic information acquired by the determined dictionary into character information, and returns the recognition result to the application 10. Here, using the numerical value recognition dictionary corresponding to “numerical attribute”, the speech information “juuman” is converted into character information “100,000” and returned to the application 10. Then, the application 10 sets the character information “100,000” returned from the phonetic character conversion device 100 in the amount input field.
Thus, the phonetic character conversion apparatus 100 has dictionary information for each input item, and converts the phonetic information into character information using dictionary information that matches the item to be input. In general, the smaller the number of words registered in the dictionary information, the higher the hit rate (probability that voice information is converted into intended character information). By having dictionary information for each input item, the number of words registered in the dictionary information to be used can be reduced, and only accurate words can be registered. Therefore, according to the phonetic character conversion device 100, the hit rate can be increased. That is, the voice recognition accuracy can be increased.

FIG. 2 is a functional block diagram showing functions of the phonetic character conversion apparatus 100 according to this embodiment.
The phonetic character conversion device 100 includes a voice information input unit 110, a voice recognition unit 120, an attribute information acquisition unit 130, a character information acquisition unit 140, and a per-attribute dictionary information storage unit 150.
The voice information input unit 110 inputs predetermined voice information output by a user who uses the application 10 by the processing device and stores it in the storage device.
The voice recognition unit 120 converts the voice information acquired by the voice information input unit 110 and generates character information by the processing device. The voice recognition unit 120 converts voice information into character information indicating the sound according to the language model. For example, there are “juu”, “ichi zero” and the like as a way of reading the numerical value “10”. In this case, the voice recognition unit 120 converts the voice information input by the user to pronounce the numerical value “10” as “ju” and converts it into character information “ju”. On the other hand, the voice information that the user pronounces and inputs the numerical value “10” with the reading “1 zero” is converted into character information “1 zero”.
The attribute information acquisition unit 130 acquires attribute information indicating the attribute of the input item at which the cursor is currently located among the plurality of input items displayed on the predetermined terminal by the application 10 from the application 10 and stores the attribute information in the storage device. To do. That is, the attribute information acquisition unit 130 acquires attribute information of the input item that is currently input. The attribute is information indicating the nature of information input to the input item such as a numerical value, an address, and a person name.
The character information acquisition unit 140 is based on the dictionary information corresponding to the attribute indicated by the attribute information acquired by the attribute information acquisition unit 130 among the dictionary information stored in the attribute-specific dictionary information storage unit 150 described later. The character information generated by the conversion is converted into other character information by the processing device and returned to the application 10 as a recognition result. For example, in the above example, the character information acquisition unit 140 converts character information such as “ju” or “first zero” into character information “10” and returns character information “10” to the application 10.
The attribute-specific dictionary information storage unit 150 stores, in the storage device, dictionary information in which the first character information and the second character information are associated with each other for each attribute of the input item. For example, in FIG. 2, the attribute-specific dictionary information storage unit 150 uses a numeric recognition dictionary for input items that input numerical values, an address recognition dictionary for input items that input addresses, and an input item that inputs personal names. In contrast, a personal name recognition dictionary or the like is stored. Here, the 1st character information is the character information after conversion, and is the character information which the character information acquisition part 140 returns to the application 10 as a recognition result. The second character information is character information that is compared with the character information generated by the voice recognition unit 120, and is character information that corresponds to reading the first character information. That is, in the above example, the first character information is “10”, and the second character information is “ju” and “first zero”.

FIG. 3 is a diagram showing a configuration of the phonetic character conversion apparatus 100 different from that in FIG.
3 does not have a function of converting speech information into character information indicating the sound in accordance with the language model, among the functions of the speech character conversion device 100 illustrated in FIG. It has that function. That is, the voice recognition apparatus 101 is provided with a function for converting voice information into character information indicating the sound according to the language model. That is, the voice recognition device 101 includes the voice information input unit 110 and the voice recognition unit 120. Then, the information acquisition unit 160 of the speech character conversion device 100 inputs the character information generated by the speech recognition unit 120 of the speech recognition device 101 by converting the speech information, and stores it in the storage device. Others are the same as the phonetic character conversion apparatus 100 shown in FIG.

FIG. 4 is a diagram illustrating an example of screen information displayed by the application 10 and dictionary information stored in the attribute-specific dictionary information storage unit 150.
The screen information includes three input items: an amount input field, an address input field, and a name input field. The attribute-specific dictionary information storage unit 150 stores three pieces of dictionary information, that is, a numerical value recognition dictionary, an address recognition dictionary, and a personal name recognition dictionary corresponding to three input items of screen information.
Each dictionary stores a plurality of pieces of first character information, and one or more pieces of second character information are stored for each piece of first character information. For example, in the case of a numerical value recognition dictionary, the second character information “1” for the first character information “1”, and the second character information “juu” for the first character information “10”. "," Zero zero ", etc. are stored.

  For example, it is assumed that the user has selected an amount input field from the terminal. In other words, it is assumed that the cursor is placed on the amount input field. In this case, the attribute information acquisition unit 130 acquires a numerical attribute as attribute information indicating the attribute of the amount input field. It is assumed that the user inputs “10” by voice while the user has selected the amount input field from the terminal. In this case, the voice information input unit 110 inputs voice information “ju”, and the voice recognition unit 120 converts the voice information into character information “ju”. The character information acquisition unit 140 first searches and selects dictionary information corresponding to the attribute information acquired by the attribute information acquisition unit 130 from the dictionary information stored in the attribute-specific dictionary information storage unit 150. That is, here, since a numerical value is input, the numerical value recognition dictionary is searched and selected. Next, the character information acquisition unit 140 searches for second character information that matches the character information generated by the voice recognition unit 120 from the selected dictionary information. That is, the character information “ju” is searched from the second character information in the numerical recognition dictionary. And the character information acquisition part 140 acquires the 1st character information corresponding to the searched 2nd character information. That is, the first character information “10” corresponding to the second character information “ju” is acquired. The character information acquisition unit 140 returns the acquired first character information to the application 10 as a recognition result. That is, the character information “10” is returned to the application 10. In the application 10, the returned “10” is entered in the amount input field.

Also, in FIG. 4, in the address recognition dictionary, the first character information is associated with each other by a hierarchical structure. For example, “Tokyo” and “Shinagawa-ku” are associated with each other in a hierarchical structure in which “Tokyo” is a parent and “Shinagawa-ku” is a child. Similarly, “Kanagawa Prefecture”, “Yokohama City”, and “Kamakura City” are related by a hierarchical structure in which “Kanagawa Prefecture” is a parent and “Yokohama City” and “Kamakura City” are children.
For example, if you enter the address "Yokohama City, Kanagawa Prefecture", enter the name of the prefecture "Kanagawa Prefecture", then enter the city name "Yokohama City", and enter the name of the prefecture. It is conceivable that the city name “Yokohama City” can be entered without any problem. Therefore, if you enter the prefecture name “Kanagawa Prefecture” and then enter the city name “Yokohama City”, the character information “Kanagawa Prefecture” is first acquired as recognition information. Next, the character information “Yokohama City” is acquired and returned as recognition information. On the other hand, if you enter the city name “Yokohama City” without entering the prefecture name, when you get the character information “Yokohama City”, the parental character information “Kanagawa Prefecture” Acquire together. Then, character information “Yokohama City, Kanagawa Prefecture” is returned as recognition information. That is, when the first character information of the child is acquired without acquiring the first character information of the parent, the first character information of the parent is acquired together with the first character information of the child. .
In the above description, when the first character information of the child is acquired by associating the plurality of first character information with a hierarchical structure, the parent first character information is also acquired. . However, the associated structure is not limited to the hierarchical structure. For example, two pieces of first character information may be associated with each other in an equal relationship, and when one is acquired, the other may be acquired.
That is, the attribute-specific dictionary information storage unit 150 includes one character information which is first character information having a plurality of first character information and two characters which are first character information different from the one character information. Store information in association with it. When the character information acquisition unit 140 acquires the second character information as the first character information, the character information acquisition unit 140 also acquires the first character information related to the second character information as the first character information. And D character information are returned as recognition information.

FIG. 5 is an explanatory diagram of an implementation example (an example of a phonetic character conversion program) of the phonetic character conversion device 100 according to this embodiment.
FIG. 5 shows an implementation example using program codes obtained by extending HTML and sheets (screen information) displayed by the program codes.
In an implementation example using program codes in which HTML is expanded, tags (for example, a SELECT tag and an INPUT tag) for displaying as screen information and a tag (VOICE tag) for converting character information are mixed. That is, in the implementation example illustrated in FIG. 5, the phonetic character conversion device 100 is incorporated in the application 10. Among tags displayed as screen information, tags for inputting information correspond to tags for converting character information and tag names (input tag 1, input tag 2, input tag 3). For example, a SELECT tag is associated with a VOICE tag having the same name by “input tag 1” as its name.
That is, when the input tag 1 is selected in the sheet, the VOICE tag with the name of the input tag 1 is executed in the program code. Here, in the VOICE tag, dictionary information corresponding to the input item is described, and character information (for example, “ju” ”) generated by converting input voice information is recognized information (for example,“ 10 ”). Convert to In the program code, the character information generated by converting the voice information is represented by alphabets, but may be katakana as described above.
For example, in the VOICE tag of the input tag 1, “10,000”, “100,000”, and “200,000” are registered as the first character information. Second character information is registered for each first character information. For example, for the first character information “100,000”, “Zyuumann”, “ichizero”, and “tou” are registered as the second character information. Furthermore, a general-purpose dictionary “suu” is registered as second character information. The general-purpose dictionary “suu” is dictionary information that collects readings of general numerical attributes, and is an external dictionary stored in another XML file or the like. That is, if a program code is created by mixing a tag to be displayed as screen information and dictionary information, the program code may become complicated. Therefore, the second character information particularly necessary for the program code is written directly in the program code, and the other general second character information is read from the external dictionary. As a result, the program code does not become complicated, and the necessary second character information is directly described in the program code, so that the processing speed is high.
The VOICE tag corresponding to the input tag 2 has a hierarchical relationship as described with reference to FIG. That is, if a city name (for example, “Yokohama City”) is input without inputting a prefecture name, “Yokohama City, Kanagawa Prefecture” is returned as recognition information.

FIG. 6 is a diagram illustrating an implementation example using JAVAScript (registered trademark) or AJAX.
Similar to the implementation example using the program code that is an extension of HTML, the implementation example using JAVAScript (registered trademark) or AJAX is also converted to the code (input type...) For displaying as screen information shown in FIG. Functions to be performed (function inputtag 1 etc.) are mixed. That is, when an input item is selected in the screen information, a function corresponding to the selected input item is called. In the function, the same processing as that of the VOICE tag is executed.
For example, when the input tag 1 is selected, the inputtag1 function is executed. In the inputtag1 function, dictionary information is directly described (the dictionary array is omitted). When the input tag 2 is selected, the inputtag2 function is executed. In the inputtag2 function, an external dictionary described in an external file such as an XML file is called. When the input tag 3 is selected, the inputtag3 function is executed. In the inputtag3 function, the character information conversion process itself by the character information acquisition unit 140 is executed by calling another program.

In addition, for example, by associating HTML and XML tag names with the VOICE tag shown in FIG. 5 in advance, a program code that incorporates a tag that converts character information simply by creating screen information as usual. Can be generated.
For example, when “JUUSHO” is added to the tag name, the tag name is set in advance so as to be associated with a VOICE tag including an “address recognition dictionary”. In this way, when creating an address entry field, simply adding “JUUSHO” to the tag name automatically generates a program code associated with a VOICE tag having an “address recognition dictionary”. .
In addition, in the case of a selection-type tag such as a SELECT tag, a dictionary that uses a word registered as a selection target as a recognition target (first character information) may be generated. And the 2nd character information with respect to 1st character information may be made to acquire by searching 1st character information for a key from a general dictionary, for example. If the recognition target cannot be determined, a general general dictionary may be set.

As described above, according to the phonetic character conversion apparatus 100 according to this embodiment, since the dictionary information is provided for each input item, the voice recognition accuracy can be increased.
In addition, by associating a word (first character information) registered in the dictionary information, necessary input information can be supplemented even when the input information is omitted.
Further, by embedding dictionary information in the document display program, the conversion process can be speeded up. On the other hand, it is possible to prevent the program code from becoming complicated by using an external dictionary for words (second character information) that have a low probability of being used.

Embodiment 2. FIG.
In this embodiment, a method of creating an electronic document by applying the phonetic character conversion apparatus 100 according to Embodiment 1 and acquiring a conversation between a user and an operator as voice information will be described.

FIG. 7 is a conceptual diagram showing an outline of the functions of the phonetic character conversion apparatus 100 according to this embodiment.
The user and the operator have a conversation, for example, by telephone. The voice character conversion device 100 acquires a conversation between a user and an operator as voice information and creates an electronic document. When the operator says “Please tell me”, the phonetic character conversion apparatus 100 moves the cursor to the input item corresponding to “˜”. For example, when the operator says “Please tell me your address,” move the cursor to the “Address” input field. Then, as described in the first embodiment, the phonetic character conversion apparatus 100 acquires the attribute information of the input item on which the cursor is placed, and switches the dictionary information to be used. When the user answers “Please tell me” from the operator, when the user answers “Yes”, the speech character conversion device 100 recognizes “XX” and goes to the input field where the cursor is placed. Fill out. By repeating this process for all items, an electronic document can be created.
As described above, the speech character conversion apparatus 100 recognizes the speech information input from the operator and switches the input item, recognizes the speech information input from the user, and writes the information in the input item. Therefore, an electronic document can be created by simply having a conversation between the user and the operator without performing a terminal operation such as switching the cursor.

FIG. 8 is a functional block diagram showing functions of the phonetic character conversion apparatus 100 according to this embodiment.
The phonetic character conversion device 100 according to this embodiment includes an item information acquisition unit 170 and an item identification dictionary information storage unit 180 in addition to the functions of the phonetic character conversion device 100 according to the first embodiment. The voice information input unit 110 includes a first voice information input unit 111 and a second voice information input unit 112.
The first voice information input unit 111 inputs the voice information (first voice information) output by the operator from the processing device via the operator terminal 11 and stores it in the storage device.
The second voice information input unit 112 inputs voice information (second voice information) output by the user from the processing device via the user terminal 12 and stores it in the storage device.
The item information acquisition unit 170 is generated by the voice recognition unit 120 converting the voice information input by the first voice information input unit 111 based on the item identification dictionary information stored in the item identification dictionary information storage unit 180 described later. The item character information that matches the character information is searched, and the item identification information corresponding to the searched item character information is acquired by the processing device.
The item identification dictionary information storage unit 180 stores, in the storage device, item identification dictionary information in which item identification information indicating the input item is associated with item character information that is predetermined character information for each input item of a plurality of input items. Remember. For example, if the item identification dictionary information storage unit 180 is an input item for inputting an amount, the item identification information “amount input field” corresponds to character information such as “amount”, “money”, and “price”. Add and remember.
Further, the attribute information acquisition unit 130 acquires attribute information of the input item indicated by the item identification information acquired by the item information acquisition unit 170. The character information acquisition unit 140 searches and selects dictionary information corresponding to the attribute information acquired by the attribute information acquisition unit 130 from the dictionary information stored in the attribute-specific dictionary information storage unit 150. The character information acquisition unit 140 corresponds to the second character information that matches the character information generated by the voice recognition unit 120 converting the voice information input by the second voice information input unit 112 based on the selected dictionary information. First character information to be acquired is acquired. Then, the character information acquisition unit 140 describes (stores) the acquired first character information in the input item.

FIG. 9 is a diagram showing a configuration of a phonetic character conversion apparatus 100 different from that in FIG.
The phonetic character conversion device 100 shown in FIG. 9 is similar to the phonetic character conversion device 100 shown in FIG. 3, and among the functions of the phonetic character conversion device 100 shown in FIG. The voice recognition device 101 is provided with the function of converting to “G”.
Here, the information acquisition unit 160 includes a first information acquisition unit 161 and a second information acquisition unit 162. The first information acquisition unit 161 inputs the character information generated by the voice recognition unit 120 by converting the voice information input by the first voice information input unit 111 by the processing device and stores the character information in the storage device. The second information acquisition unit 162 inputs the character information generated by the voice recognition unit 120 by converting the voice information input by the second voice information input unit 112 by the processing device and stores the character information in the storage device.
Others are the same as the phonetic character conversion apparatus 100 shown in FIG.

In the above description, input items are switched according to voice information input from an operator. However, the input items may be switched based on voice information input not only by the operator but also by the user.
Also, the item identification dictionary information storage unit 180 stores the order of input items, and when voice information such as “next” is input, the item information acquisition unit 170 inputs the current cursor position. The item identification information of the input item next to the item may be acquired.
In addition, the operator may change the input items by operating the buttons on the operator terminal 11. The input information (first character information) may be corrected.

  As described above, according to the phonetic character conversion apparatus 100 according to this embodiment, an electronic document can be created only by a conversation between a user and an operator.

Embodiment 3 FIG.
In this embodiment, a method for improving the recognition accuracy of voice information by using two pieces of voice information, that is, voice information input by a user and voice information input by an operator will be described.

FIG. 10 is a conceptual diagram showing an outline of functions of the phonetic character conversion apparatus 100 according to this embodiment.
FIG. 10 shows an example where the operator asks the user where the address is. Suppose the user answers "Tokyo" when asked for his address. Then, the voice character conversion device 100 inputs the voice information “Tokyo” answered by the user, and converts it into character information indicating the sound. Here, it is assumed that “Toyoto” is converted. On the other hand, the operator repeats “Tokyo” to confirm the contents answered by the user. Then, the phonetic character conversion apparatus 100 inputs the voice information that the operator repeats “Tokyo”, and converts it into character information indicating the sound. Here, it is assumed that “Toyoyoto” is converted. Then, the phonetic character conversion apparatus 100 compares the two character information “Toyoyo” and “Toyoyo” generated by converting the two phonetic information. Then, it can be seen that the character information at the two locations “U” and “O” and “HI” and “KI” do not match. Therefore, it is determined whether the character information converted from the user's voice information or the character information converted from the operator's voice information is likely, and the most likely character information is adopted. For example, here, for "u" and "o", the character information "u" converted from the user's voice information is likely, and for "hi" and "ki", the character information converted from the operator's voice information Suppose that "Ki" seems to be certain. In other words, “U” and “KI” are adopted. Then, the character information “Tokyo” is generated.
Further, the phonetic character conversion device 100 changes a language model in which a rule for converting voice information into character information is determined for each person who inputs the voice information based on the recognition result, and the conversion accuracy (recognition rate). To increase. That is, in the above example, the phonetic character conversion device 100 converts the pronunciation of the user (for example, intonation and sound frequency) that has been converted to “Toyoto” into “Toyoto” for this user. I understand that I have to. Therefore, based on this result, the phonetic character conversion apparatus 100 changes the language model.
As described above, the phonetic character conversion apparatus 100 increases the recognition rate by using two pieces of speech information, and further increases the recognition rate by improving the language model based on the recognition result. Therefore, the recognition rate increases every time the user and the operator have a conversation.

FIG. 11 is a functional block diagram showing functions of the phonetic character conversion apparatus 100 according to this embodiment.
The speech recognition unit 120 of the speech to character conversion device 100 according to this embodiment includes a speech conversion unit 121, a character information comparison unit 122, a mismatched part determination unit 123, a character information generation unit 124, a language model storage unit 125, and a language model update. Part 126 is provided. Others are the same as the phonetic character conversion apparatus 100 according to the second embodiment.
The voice conversion unit 121 converts the first voice information input by the operator (first voice information input unit 111) into character information by the processing device based on the language model for the operator. Similarly, the voice conversion unit 121 converts the second voice information input by the user (second voice information input unit 112) into character information by the processing device based on the language model for the user. In addition, when converting the voice information to the character information, the voice conversion unit 121 calculates, with the processing device, the accuracy with which the conversion is correct for each character of the converted character information.
The character information comparison unit 122 compares the two pieces of character information recognized by the voice conversion unit 121 and extracts a mismatched portion by the processing device.
The non-matching portion determination unit 123 determines the character information of the non-matching portion extracted by comparison by the character information comparison unit 122 as the character information of the non-matching portion between the two pieces of character information. The mismatching part determination unit 123 determines which character information is used for each character of the character information of the mismatching part based on the accuracy calculated by the voice conversion unit 121.
The character information generation unit 124 replaces any mismatched portion of the two character information with the character information determined by the mismatched portion determination unit 123, and generates the character information by the processing device.
The language model storage unit 125 stores, in a storage device, a language model for converting speech information into character information for each person who inputs speech information (that is, for each user and operator).
The language model update unit 126 updates the language model of the operator based on the character information generated by the character information generation unit 124 and the first voice information, and the character information generated by the character information generation unit 124 and the second information The language model of the user is updated by the processing device based on the voice information.

FIG. 12 is a diagram for supplementarily explaining the process of converting voice information to character information described with reference to FIG.
As described above, the language model storage unit 125 stores a language model for each person who inputs speech information, and the speech conversion unit 121 converts speech information into character information based on the language model of the person who inputs speech information. To do. That is, a so-called specific speaker type speech recognition process is performed. In this case, the accuracy of recognition changes depending on how much information the language model has about the pronunciation of the person (the person who inputted the speech information), that is, how much information is learned. In other words, the more you learn about the person's pronunciation, the higher the accuracy of recognition. The accuracy of recognition also changes depending on the clarity of pronunciation. The voice conversion unit 121 calculates the recognition accuracy for each character while converting the voice information into the character information.
For example, in the example shown in FIG. 12, it is assumed that the user is a new customer, the language model is in an unlearned state, the operator is an operator with a high level of proficiency, and the language model is being learned. Therefore, the character information “Toyo Hyoto” obtained by converting the voice information input by the user has a low recognition accuracy as a whole. On the other hand, the character information “Toyoyoto” obtained by converting the voice information input by the operator has a high recognition accuracy as a whole. However, “O” of “Toyoyo” obtained by converting the voice information input by the operator has an unclear pronunciation, so the recognition accuracy is low. Here, the recognition accuracy of the character information of the mismatched portion extracted by comparing the two character information by the character information comparison unit 122 is 60% of “U” of “Toyoyo” converted from the voice information input by the user. , “Hi” is 30%, “O” of “Toyokyo” converted voice information input by the operator is 50%, and “Ki” is 90%.
The non-matching part determination unit 123 compares “U” (60%) and “O” (50%), adopts “U” with high recognition accuracy, and “H” (30%) and “K” (“ 90%) and adopt “ki” with high recognition accuracy.
For example, the character information generation unit 124 replaces “U” and “HI”, which are inconsistent parts of “TOYO HYOTO”, with “U” and “KI”, respectively, and changes the character information “TOYOTO” Generate.
Further, the language model update unit 126 updates the user's language model based on the character information “Tokyo” generated by the character information generation unit 124 and the pronunciation of the user. The language model update unit 126 updates the language model of the operator based on the character information “Tokyo” generated by the character information generation unit 124 and the pronunciation of the operator.

  Any method may be used for extracting specific audio information from the audio information input by the user or operator (for example, extracting “Tokyo” from the audio information “Tokyo is”). . For example, “is”, “sound” that is generally added to the end of a word, “it” that is added to the beginning of a word (in the case of “that is ...”), etc. may be omitted. Further, this omitted information may be learned in the same manner as the language model.

As described above, according to the phonetic character conversion device 100 according to this embodiment, the recognition accuracy can be increased by converting the text information into one character information using the two voice information of the user and the operator. .
Moreover, since the language model is updated using the character information generated from the two pieces of voice information and the input voice information, the recognition accuracy increases as the user and the operator have a conversation.
It should be noted that the accuracy of deriving the recognition result from the speech information can be further increased by carrying out the processing together with the character information conversion processing shown in the first embodiment.

Embodiment 4 FIG.
In this embodiment, an example in which the phonetic character conversion device 100 described in the above embodiment is applied to a call center system will be described.

FIG. 13 is a conceptual diagram showing an outline of functions of the phonetic character conversion apparatus 100 according to this embodiment.
For example, when a product is sold in a financial business or the like, it may be obliged to give a user a predetermined explanation about the risk of the product in advance. In the call center, when a user offers to purchase a product, the operator gives a mandatory explanation. However, since the required explanation is determined for each product, the operator simply reads out the explanation.
Therefore, the voice character conversion device 100 provides the explanation by voice instead of the operator. When the voice description ends, the voice character conversion device 100 checks whether or not the user understands the contents of the explanation. At this time, the user inputs whether or not he / she understands by voice as in the case of talking with the operator. Then, the voice character conversion apparatus 100 recognizes the voice information input using the confirmation recognition dictionary and transmits the recognition result to the operator. The operator can know from the recognition result whether or not the user understands the explanation. If the operator agrees, the operator talks with the user and proceeds to a process for purchasing a product.
In addition, the voice character conversion device 100 accepts voice input from the user even during voice explanation. For example, when the user wants to pause the explanation, he / she inputs that fact by voice. Then, the phonetic character conversion apparatus 100 recognizes the input voice using the interrupt recognition dictionary and pauses the explanation. Further, when the user has a question about the contents of explanation, the fact is input by voice as well. Then, the voice character conversion apparatus 100 recognizes the input voice using the interrupt recognition dictionary and transmits the recognition result to the operator. The operator can know from the recognition result that the user has a question in the explanation. Thus, the operator can talk to the user and answer the questions.
In this way, the voice character conversion device 100 performs a voice explanation on behalf of the operator, and accepts a response from the user by voice. Therefore, the operator can perform other work while the phonetic character conversion apparatus 100 is explaining, and work efficiency can be increased. In addition, the user can receive the same explanation as that given by the operator. Further, when there is a question, the user can tell that there is a question not by operating the machine but by the voice as when the operator is responding, and does not need to perform an unfamiliar machine operation.

FIG. 14 is a functional block diagram showing functions of the phonetic character conversion apparatus 100 according to this embodiment.
Spoken character conversion apparatus 100 according to this embodiment includes voice output section 190, confirmation information requesting section 200, and transmission section 210 in addition to the functions of voice / character conversion apparatus 100 according to the first embodiment.
The audio output unit 190 outputs predetermined explanation information stored in the storage device as audio information. The voice output unit 190 may output the description information stored as the voice information in advance as the voice information, or may convert the description information stored as the character information or the like into the voice information and output it.
When the audio output unit 190 finishes outputting the description information, the confirmation information request unit 200 requests the input of predetermined confirmation information from the processing device. That is, the confirmation information requesting unit 200 requests an input as to whether or not the contents explained are accepted. On the other hand, the voice information input unit 110 inputs the confirmation information input by the user as voice information, and the voice recognition unit 120 converts it into character information. Then, the character information acquisition unit 140 matches the character information converted by the voice recognition unit 120 based on the confirmation recognition dictionary that is the dictionary information for confirmation information among the dictionary information stored in the attribute-specific dictionary information storage unit 150. First character information corresponding to the second character information is acquired. For example, “OK”, “Cancel”, etc. are registered as the first character information. In addition, “Ryokai”, “Kakunin”, etc. are registered as the second character information for the first character information “OK”, and “Torikeshi”, as the second character information for the first character information “Cancel”, “Cancel” or the like is registered.
The transmission unit 210 transmits the first character information acquired by the character information acquisition unit 140 as confirmation information to the operator's terminal via the communication device.

  Also, the voice information input unit 110 inputs the interrupt information input by the user as voice information while the voice output unit 190 is outputting description information. The voice recognition unit 120 converts the input voice information into character information. The character information acquisition unit 140 matches the character information converted by the speech recognition unit 120 based on the interrupt recognition dictionary that is dictionary information for interrupt information among the dictionary information stored in the attribute-specific dictionary information storage unit 150. First character information corresponding to the second character information is acquired. The transmission unit 210 transmits the first character information acquired by the character information acquisition unit 140 as interrupt information to the operator's terminal via the communication device.

The attribute information acquisition unit 130 acquires an interrupt attribute as attribute information from the audio output unit 190 when the audio output unit 190 outputs audio information, and when the audio output unit 190 finishes outputting the audio information, A confirmation attribute is acquired from the output unit 190 as attribute information. The character information acquisition unit 140 switches the dictionary information to be used according to the attribute information acquired by the attribute information acquisition unit 130 as in the first embodiment.
Further, the input from the user may be received by the button operation or the like together with the voice information.
In addition, the description may display not only audio but also video or the like on the user terminal.

FIG. 15 is an explanatory diagram of an implementation example (an example of a bidirectional explanation confirmation program) of the phonetic character conversion apparatus 100 according to this embodiment.
In the interactive explanation confirmation program shown in FIG. 15, the explanatory text is displayed on the user's terminal and output as a voice. Similarly, in a scene where confirmation is requested, a confirmation / cancel button is displayed on the user's terminal and is output as a voice to accept confirmation / cancellation input by voice as well as confirmation / cancellation input by the button.
A tag to be displayed (for example, an INPUT tag) and a tag for voice output (VOICE tag) are associated with each other by a tag name.
In FIG. 15, an example of implementation is shown by using program codes obtained by extending HTML. However, implementation using JAVAScript (registered trademark) or AJAX as described with reference to FIG. 6 is also possible.

As described above, according to the phonetic character conversion device 100 according to this embodiment, the phonetic character conversion device 100 performs a voice explanation on behalf of the operator and accepts a response from the user by voice. While the voice character conversion apparatus 100 is explaining, other work can be performed and work efficiency can be improved.
Further, since the user only needs to operate by voice, there is no inconvenience for the user.
Further, since the voice character conversion device 100 outputs the set explanatory text without adapting it, the contents can be conveyed more accurately than the explanation by the operator.

Next, the hardware configuration of the phonetic character conversion apparatus 100 in the above embodiment will be described.
FIG. 16 is a diagram illustrating an example of a hardware configuration of the phonetic character conversion apparatus 100.
As shown in FIG. 16, the phonetic character conversion apparatus 100 includes a CPU 911 (also referred to as a central processing unit, a central processing unit, a processing unit, an arithmetic unit, a microprocessor, a microcomputer, and a processor) that executes a program. . The CPU 911 is connected to the ROM 913, the RAM 914, the LCD 901 (Liquid Crystal Display), the keyboard 902, the communication board 915, and the magnetic disk device 920 via the bus 912, and controls these hardware devices. Instead of the magnetic disk device 920, a storage device such as an optical disk device or a memory card read / write device may be used.

  The ROM 913 and the magnetic disk device 920 are examples of a nonvolatile memory. The RAM 914 is an example of a volatile memory. The ROM 913, the RAM 914, and the magnetic disk device 920 are examples of storage devices. The communication board 915 and the keyboard 902 are examples of input devices. The communication board 915 is an example of an output device. Furthermore, the communication board 915 is an example of a communication device. Furthermore, the LCD 901 is an example of a display device.

  An operating system 921 (OS), a window system 922, a program group 923, and a file group 924 are stored in the magnetic disk device 920 or the ROM 913. The programs in the program group 923 are executed by the CPU 911, the operating system 921, and the window system 922.

The program group 923 stores a program for executing each process of the phonetic character conversion apparatus 100 in the above description and other programs. The program is read and executed by the CPU 911. That is, it has been described as “voice information input unit 110”, “voice recognition unit 120”, “attribute information acquisition unit 130”, “character information acquisition unit 140”, “information acquisition unit 160”, and “item information acquisition unit 170”. A program for executing the function and other programs are stored.
The file group 924 stores information, data, signal values, variable values, and parameters handled by the phonetic character conversion apparatus 100 in the above description as items of “file” and “database”. That is, information stored in the “attribute-specific dictionary information storage unit 150” and the “item identification dictionary information storage unit 180” is stored as items of “file” and “database”. The “file” and “database” are stored in a recording medium such as a disk or a memory. Information, data, signal values, variable values, and parameters stored in a storage medium such as a disk or memory are read out to the main memory or cache memory by the CPU 911 via a read / write circuit, and extracted, searched, referenced, compared, and calculated. Used for the operation of the CPU 911 such as calculation / processing / output / printing / display. Information, data, signal values, variable values, and parameters are temporarily stored in the main memory, cache memory, and buffer memory during the operation of the CPU 911 for extraction, search, reference, comparison, calculation, calculation, processing, output, printing, and display. Is remembered.
In addition, the arrows in the flowcharts in the above description mainly indicate input / output of data and signals, and the data and signal values are recorded in a memory of the RAM 914 and other recording media such as an optical disk. Data and signals are transmitted online via a bus 912, signal lines, cables, or other transmission media.

  In addition, what is described as “to part” in the above description may be “to circuit”, “to device”, “to device”, “to means”, and “to function”. It may be “step”, “˜procedure”, “˜processing”. In addition, what is described as “˜device” may be “˜circuit”, “˜device”, “˜device”, “˜means”, “˜function”, and “˜step”, “ ~ Procedure "," ~ process ". Furthermore, what is described as “to process” may be “to step”. That is, what is described as “˜unit” may be realized by firmware stored in the ROM 913. Alternatively, it may be implemented only by software, or only by hardware such as elements, devices, substrates, and wirings, by a combination of software and hardware, or by a combination of firmware. Firmware and software are stored in a recording medium such as ROM 913 as a program. The program is read by the CPU 911 and executed by the CPU 911. That is, the program causes a computer or the like to function as the “˜unit” described above. Alternatively, the computer or the like is caused to execute the procedures and methods of “to part” described above.

FIG. 3 is a conceptual diagram showing an outline of functions of the phonetic character conversion device 100 according to the first embodiment. FIG. 3 is a functional block diagram showing functions of the phonetic character conversion device 100 according to the first embodiment. FIG. 3 is a diagram showing a configuration different from FIG. 2, which is a phonetic character conversion apparatus 100 according to the first embodiment. The figure which shows an example of the screen information displayed by the application 10, and the dictionary information which the dictionary information storage part 150 for every attribute memorize | stores. Explanatory drawing of the example of implementation (an example of a phonetic character conversion program) of the phonetic character conversion apparatus 100 which concerns on Embodiment 1. FIG. The figure which shows the example of mounting by JAVAScript (trademark) or AJAX. The conceptual diagram which shows the outline | summary of the function of the audio | voice character conversion apparatus 100 which concerns on Embodiment 2. FIG. FIG. 4 is a functional block diagram showing functions of the phonetic character conversion device 100 according to the second embodiment. FIG. 9 is a phonetic character conversion device 100 according to Embodiment 2 and shows a configuration different from that in FIG. 8. FIG. 9 is a conceptual diagram showing an outline of functions of the phonetic character conversion device 100 according to the third embodiment. FIG. 9 is a functional block diagram showing functions of the phonetic character conversion device 100 according to the third embodiment. FIG. 11 is a supplementary explanatory diagram of processing for converting voice information to character information described based on FIG. 10. The conceptual diagram which shows the outline | summary of the function of the audio | voice character conversion apparatus 100 which concerns on Embodiment 4. FIG. FIG. 6 is a functional block diagram showing functions of the phonetic character conversion device 100 according to the fourth embodiment. Explanatory drawing of the implementation example (an example of a bidirectional | two-way description confirmation program) of the phonetic character conversion apparatus 100 which concerns on Embodiment 4. FIG. The figure which shows an example of the hardware constitutions of the voice character conversion apparatus 100.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 10 Application, 11 Operator terminal, 12 User terminal, 100 Voice character conversion apparatus, 101 Voice recognition apparatus, 110 Voice information input part, 111 1st voice information input part, 112 2nd voice information input part, 120 Voice recognition part 121, speech conversion unit, 122 character information comparison unit, 123 mismatched part determination unit, 124 character information generation unit, 125 language model storage unit, 126 language model update unit, 130 attribute information acquisition unit, 140 character information acquisition unit, 150 attribute Each dictionary information storage unit, 160 information acquisition unit, 161 first information acquisition unit, 162 second information acquisition unit, 170 item information acquisition unit, 180 item identification dictionary information storage unit, 190 voice output unit, 200 confirmation information request Part, 210 transmitting part.

Claims (6)

A first voice information input unit for inputting the first voice information input by the first terminal and storing the first voice information in a storage device;
A second voice information input unit for inputting the second voice information input by the second terminal and storing the second voice information in a storage device;
The processing apparatus, the conversion of the first audio information converted into the first character information, the converted one character first audio information for the first character information first voice information input unit inputs The first accuracy indicating the correct accuracy is calculated, and the processing device converts the second speech information input by the second speech information input unit into the second character information, and the second character information. A speech conversion unit that calculates a second accuracy indicating the accuracy with which the conversion from the second speech information character by character is correct ;
A character information comparison unit that compares the first character information converted by the voice conversion unit with the second character information and extracts a mismatched portion by a processing device;
The character information of the mismatched portion extracted by comparison by the character information comparison unit is compared character by character, and the first accuracy and the second accuracy are compared one character at a time . inconsistent area determining unit determining the processing device to one of the characters of the second character information,
A character information generation unit that replaces the mismatched portion of the first character information or the second character information with the character information determined by the mismatched portion determination unit one character at a time , and generates character information by a processing device; A phonetic character conversion device characterized by that. The mismatch area determining section, character information of the upper Symbol unmatched portion, when the first probable determines in the first character information, when the second accuracy is higher the second The phonetic character conversion apparatus according to claim 1, wherein the character information is determined. The phonetic character conversion device further includes:
For each user who uses the first terminal and the second terminal, a language model storage unit that stores a language model for converting speech information into character information in a storage device,
The speech conversion unit converts the first speech information into first character information based on a first language model stored by the language model storage unit for a first user who uses the first terminal. In addition to the conversion, the language model storage unit converts the second speech information into second character information based on the second language model stored for the second user who uses the second terminal. The phonetic character conversion device according to claim 1 or 2 , The phonetic character conversion device further includes:
Based on the character information generated by the character information generation unit and the first voice information, the language model of the first user who uses the first terminal is updated by the processing device, and the character information generation unit based on the generated character information and said second audio information, according to claim 3, characterized in that it comprises a language model updating unit that updates the processing unit a second user language model using the second terminal The phonetic character conversion device described in 1. A first audio information input step in which the processing device inputs the first audio information input by the first terminal;
A second audio information input step in which the processing device inputs the second audio information input by the second terminal;
The processing device converts the first voice information input in the first voice information input step into first character information, and the converted first character information is converted from the first voice information character by character. The first accuracy indicating the correct accuracy is calculated , the second speech information input in the second speech information input step is converted into second character information, and the second character information is converted character by character. A speech conversion step of calculating a second accuracy indicating the accuracy of conversion from the speech information of 2 ;
A character information comparison step in which the processing device compares the first character information converted in the voice conversion step with the second character information and extracts a mismatched portion;
The processing device compares the first accuracy and the second accuracy one character at a time for the character information of the inconsistent portion extracted by comparison in the character information comparison step, and compares the first character one character at a time. A non- matching part determination step for determining any character of the information and the second character information;
A character information generation step in which the processing device replaces the mismatched portion of the first character information or the second character information with the character information determined in the mismatched portion determination step one by one , and generates character information. A phonetic character conversion method comprising: A first audio information input process for inputting the first audio information input by the first terminal;
A second audio information input process for inputting the second audio information input by the second terminal;
The first voice information input in the first voice information input process is converted into first character information, and the conversion from the first voice information character by character is correct for the converted first character information. While calculating 1st accuracy, it converts the 2nd audio | voice information input by the said 2nd audio | voice information input process into 2nd character information, The 2nd audio | voice information per character about the said 2nd character information A voice conversion process for calculating a second accuracy indicating that the conversion from is correct ;
A character information comparison process for comparing the first character information converted by the voice conversion process with the second character information and extracting a mismatched portion;
The character information of the mismatched portion extracted by comparison in the character information comparison process is compared character by character, the first accuracy and the second accuracy, and the first character information and the first character by character. inconsistent area determining process for determining the one of the characters of the second character information,
Causing the computer to execute character information generation processing for generating character information by replacing the inconsistent portion of the first character information or the second character information one by one with the character information determined in the inconsistent portion determination processing. A phonetic character conversion program characterized by that.

Images