JP7406921B2 - Information processing device, information processing method and program - Google Patents

Description

The present invention relates to an information processing device, an information processing method, and a program.

In a speech recognition device aimed at unspecified people, a dictionary for speech recognition centered on general-purpose and general vocabulary is registered in advance, and the speech recognition device can use the registered dictionary for speech recognition. Recognize speech based on. In such speech recognition devices, if the vocabulary to be recognized can be defined at the time of design, a pre-created speech recognition dictionary is used; however, if the vocabulary cannot be defined or should be changed dynamically. In some cases, vocabulary for speech recognition is typically generated manually or automatically from character string information and registered in a dictionary.

In addition, recent speech recognition devices are able to handle not only formal word pronunciations but also arbitrary abbreviated pronunciations by the user by registering paraphrase expressions such as abbreviations in the speech recognition dictionary. are doing.

For example, Patent Document 1 discloses a speech recognition device that can recognize even abbreviated paraphrase expressions of words with a high recognition rate.

Patent No. 3724649

However, the speech recognition device disclosed in Patent Document 1 uses general and uncommon new words (special When registering (terms) in a dictionary for speech recognition, manual input is required, which places a heavy burden on human labor such as selecting and inputting words to be registered. Therefore, from the viewpoint of appropriately generating a dictionary for speech recognition, this method is still insufficient.

The present invention has been made in view of the above-mentioned circumstances, and an object of the present invention is to provide an information processing device, an information processing method, and a program that can suitably generate a dictionary for speech recognition.

In order to achieve the above object, an information processing device according to a first aspect of the present invention includes:
A speech recognition result that receives a first speech recognition result based on a first dictionary and a second speech recognition result based on a second dictionary that is different from the first dictionary and includes word information generated by the user. receiving means;
Receive a first certainty factor about the first speech recognition result calculated based on a predetermined calculation and a second certainty factor about the second speech recognition result calculated based on the calculation. certainty level receiving means;
As a result of comparing the first certainty factor and the second certainty factor, if the difference in the certainty factor is larger than a predetermined value, the word information included in the second speech recognition result is stored in the first dictionary. word information storage means for storing it as an update list ;
Transmitting means for transmitting an update instruction to add word information included in the update list stored in the word information storage means to the first dictionary;
It is characterized by having the following .

further comprising an extraction means for extracting word information to be stored from the second speech recognition result in accordance with a predetermined standard when the difference in the certainty factor is large ;
The word information storage means stores the word information extracted by the extraction means.
You can do it like this.

Further comprising a classification means for classifying the word information extracted by the extraction means into one of a plurality of predetermined classifications for each appearance frequency,
The word information storage means stores word information classified by the classification means for each classification.
You can do it like this.

The word information includes audio information and text information,
further comprising a first dictionary updating means for updating the first dictionary by adding word information stored as the update list by the word information storage means to the first dictionary;
The second dictionary is newly stored by the user's operation every time the first dictionary is updated.
You can do it like this.

In order to achieve the above object, an information processing method according to a second aspect of the present invention includes:
A speech recognition result that receives a first speech recognition result based on a first dictionary and a second speech recognition result based on a second dictionary that is different from the first dictionary and includes word information generated by the user. a receiving step;
Receive a first certainty factor about the first speech recognition result calculated based on a predetermined calculation and a second certainty factor about the second speech recognition result calculated based on the calculation. a confidence level receiving step;
As a result of comparing the first certainty factor and the second certainty factor, if the difference in the certainty factor is larger than a predetermined value, the word information included in the second speech recognition result is stored in the first dictionary. a step of storing word information as an update list ;
a sending step of sending an update instruction to add the word information included in the update list stored in the word information storage step to the first dictionary;
It is characterized by having the following .

In order to achieve the above object, a program according to a third aspect of the present invention,
computer,
A speech recognition result that receives a first speech recognition result based on a first dictionary and a second speech recognition result based on a second dictionary that is different from the first dictionary and includes word information generated by the user. receiving means,
Receive a first certainty factor about the first speech recognition result calculated based on a predetermined calculation and a second certainty factor about the second speech recognition result calculated based on the calculation. confidence receiving means;
As a result of comparing the first certainty factor and the second certainty factor, if the difference in the certainty factor is larger than a predetermined value, the word information included in the second speech recognition result is stored in the first dictionary. word information storage means for storing it as an update list ;
Transmitting means for transmitting an update instruction to add word information included in the update list stored in the word information storage means to the first dictionary;
It is characterized by functioning as

According to the present invention, a dictionary for speech recognition can be suitably generated.

1 is a block diagram showing an example of an information processing system according to an embodiment of the present invention. FIG. 1 is a block diagram showing an example of an information processing device according to an embodiment of the present invention. FIG. 1 is a block diagram showing an example of a speech recognition server according to an embodiment of the present invention. FIG. 2 is an explanatory diagram for explaining the overall processing of the information processing system. 3 is a flowchart illustrating an example of term registration processing. It is a figure showing an example of a voice recognition result. It is a figure showing an example of a morpheme and a part of speech of a speech recognition result.

An information processing apparatus 100 according to the present invention will be explained using an example in which it is applied to an information processing system 1 shown in FIG. In the information processing system 1, as shown in FIG. 1, information processing devices 100A and 100B and a voice recognition server 200 are communicably connected via a network 510. For ease of understanding, this embodiment will be described below using an example in which a user of information processing device 100A and a user of information processing device 100B converse with each other. Note that the information processing device 100A and the information processing device 100B are also simply referred to as the information processing device 100.

The information processing device 100 is an information terminal (so-called computer) such as a mobile phone, a smartphone, a tablet, or a PC (Personal Computer), and has constructed a distributed network 510 such as P2P (Peer to Peer). Note that the information processing system 1 is not limited to a P2P type system, and may be, for example, a cloud computing type system.

The information processing device 100 has a function of outputting voice data and text data (speech recognition results) of conversations of users of other information processing devices 100, which are received from the voice recognition server 200. Furthermore, the information processing device 100 has a function of extracting word information to be registered from the voice recognition result based on the confidence level received from the voice recognition server 200, and registering the word information in the dictionary for voice recognition.

The speech recognition server 200 is any computer device such as a mainframe, a workstation, or a PC (Personal Computer). The voice recognition server 200 recognizes the voice (conversation content) transmitted from the information processing device 100 based on a pre-stored dictionary for voice recognition, and stores the recognized voice data together with text data (as a voice recognition result). ) It has a function of transmitting to other information processing apparatuses 100. The speech recognition server 200 also has a function of calculating a confidence level indicating the probability that the vocabulary obtained as a speech recognition result matches the vocabulary actually uttered, and transmitting it to other information processing devices 100. There is.

Next, with reference to FIG. 2, the configuration of information processing apparatus 100 (information processing apparatus 100A and information processing apparatus 100B shown in FIG. 1) in this embodiment will be described. Although not shown, a functional unit is provided that has the function of converting the user's conversation (voice) into audio data for transmission (from analog to digital) (and vice versa). do.

As shown in FIG. 2, the information processing device 100 includes a storage section 110, a control section 120, an input/output section 130, a communication section 140, and a system bus (not shown) that interconnects these sections. ing.

The storage unit 110 includes a ROM (Read Only Memory), a RAM (Random Access Memory), and the like. The ROM stores programs executed by a CPU (Central Processing Unit) of the control unit 120 and data required in advance to execute the programs (not shown).

Specifically, the storage unit 110 in this embodiment stores, as a list of registered terms 111, audio data of words to be registered as a dictionary for speech recognition and their text data. Note that the audio data and the text data corresponding to the audio data are also collectively referred to as word information. Note that the registered term list 111 shows a list of word information to be registered, and includes a plurality of word information. The word information in the registered term list 111 is stored in the storage unit 110 for each category through a term registration process that will be described later. The storage unit 110 also stores, as a registered classification 112, a list of registered classifications that are classified based on user specifications and their classification criteria. As registered classifications, for example, classifications such as "universally used internal terminology" and "internal organizational terminology used within a specific organization" are registered based on user specifications. As a classification standard, for example, the frequency of occurrence of the word information to be registered in a conversation is memorized, and word information that appears five or more times is classified as "universally used company terminology". If it is less than 5 times, it may be set as desired by the user, such as classifying it into "internal terms used within a specific organization".

The control unit 120 includes a CPU, an ASIC (Application Specific Integrated Circuit), and the like. The control unit 120 operates according to a program stored in the storage unit 110 and executes processing according to the program. The control unit 120 includes a certainty comparison unit 121, a morpheme extraction unit 122, a part of speech estimation unit 123, a term classification unit 124, and a term registration unit as main functional units provided by the program stored in the storage unit 110. 125.

The confidence level comparison unit 121 is a functional unit that compares the confidence levels transmitted from the speech recognition server 200. As will be described in detail later, the speech recognition server 200 provides speech recognition results (based on the first registered term list 211, which will be described later) when the first registered term list 211 is used as a speech recognition dictionary (first dictionary). text data and its audio data), its confidence level A (first confidence level), and the speech recognition result when the second registered term list 212 is used as a dictionary for speech recognition (second dictionary). 2 text data and its audio data based on the list of registered terms 212) and its confidence level B (second confidence level) are transmitted. The confidence level comparison unit 121 compares the confidence level A and the confidence level B. Specifically, the confidence level comparison unit 121 compares the confidence levels by determining whether a value obtained by subtracting the confidence level A from the confidence level B is greater than or equal to a predetermined threshold. The threshold value may be set by the user to a different value depending on, for example, the content of the meeting or the language used.

For example, the morpheme extraction unit 122 uses the speech recognition result (first speech recognition result) when the first registered term list 211 is used as a dictionary for speech recognition, and the second registered term list 212 as a dictionary for speech recognition. This is a functional unit that divides each of the speech recognition results (second speech recognition results) into morphemes by morpheme analysis and extracts different morphemes. Specifically, the morpheme extracting unit 122 extracts different morphemes by subtracting the morphemes in common with the first speech recognition result divided for each morpheme from the second speech recognition result divided for each morpheme.

The part-of-speech estimation unit 123 is a functional unit that extracts morphemes with different parts of speech by comparing the parts of speech of each morpheme in the first speech recognition result and the second speech recognition result. Specifically, the part-of-speech estimation unit 123 compares the morpheme of the first speech recognition result with the morpheme of the second speech recognition result, and determines that although the part of speech of the morpheme of the second speech recognition result is a noun, Extract morphemes other than nouns. That is, the morpheme extraction unit 122 extracts a word morpheme (a different character string) from the second speech recognition result that is different from the first speech recognition result, whereas the part-of-speech estimation unit 123 extracts a word morpheme (a different character string) from the second speech recognition result. A morpheme of a part of speech that is different from the first speech recognition result is extracted. In other words, it can be said that the morpheme extraction unit 122 is a functional unit that extracts morphemes from the perspective of character strings, and the part-of-speech estimation unit 123 is a functional unit that extracts morphemes from the perspective of parts of speech. Note that special terms such as "universally used internal terminology" and "internal organizational terminology used within a specific organization" are often nouns. Therefore, the part of speech estimating unit 123 in this embodiment extracts a morpheme whose part of speech of the morpheme in the second speech recognition result is a noun, but whose morpheme in the first speech recognition result is other than a noun. Alternatively, morphemes of different parts of speech may simply be output to the input/output unit 130, and the user may select whether or not to extract them.

The term classification unit 124 determines whether or not the morpheme extracted by the function of the morpheme extraction unit 122 matches the morpheme extracted by the function of the part of speech estimation unit 123, and if they match, the term classification unit 124 certifies the morpheme as a registration target. , is a functional unit that classifies the word information of the certified morpheme to be registered based on the classification based on the registration classification 112. Specifically, when the extracted morphemes match, the term classification unit 124 classifies any of the registered morphemes according to the classification criteria set as the registration classification 112 based on the frequency of appearance of the word information to be registered. Classify into one of the following categories.

The term registration unit 125 is a functional unit that registers the word information classified by the term classification unit 124 into the registered term list 111 for each classification. The term registration unit 125 also has a function of transmitting an update instruction to update the content of the first registered term list 211 to the speech recognition server 200 based on the content of the word information registered in the registered term list 111. There is. Note that the term registration section 125 functions as a word information registration means.

These functional units cooperate to realize a function of registering word information to be registered in a speech recognition dictionary in the information processing apparatus 100.

The input/output unit 130 is a device for inputting/outputting data, and includes a keyboard, a mouse, a camera, a microphone, a liquid crystal display, an organic EL (Electro-Luminescence) display, and the like.

The communication unit 140 is a device for communicating with other information processing devices 100 and the voice recognition server 200 via the network 510.

The above is the configuration of the information processing device 100. Next, with reference to FIG. 3, the configuration of the speech recognition server 200 in this embodiment will be described. As shown in FIG. 3, the speech recognition server 200 includes a storage section 210, a control section 220, an input/output section 230, a communication section 240, and a system bus (not shown) that interconnects these sections. ing.

The storage unit 210 includes ROM, RAM, and the like. The ROM stores programs to be executed by the CPU of the control unit 220 and data required in advance to execute the programs (not shown).

Specifically, the storage unit 210 in this embodiment stores a first registered term list 211 and a second registered term list 212 as a dictionary for speech recognition. The first registered term list 211 is a list of word information, and the registered word information is updated every time the term registration process described below is executed. Note that the initial first registered term list 211 may be a list of word information centered on general-purpose and general vocabulary, and may be, for example, generated by a user or published on a network. You can also obtain it by downloading something.

On the other hand, the second registered term list 212 contains more special terms than the first registered term list 211, such as "universally used in-house terms" and "internal terms used within a specific organization." This is a list of word information generated by the user to include as much information as possible. Note that the second registered term list 212 may be generated by the user for each conference or lecture, for example, based on the materials of the scheduled conference or lecture. In the information processing apparatus 100 in this embodiment, the second registered term list 212 is newly stored for each meeting (in other words, each time the first registered term list 211 is updated), and the term registration process described below is performed. It will be done. In the term registration process, target word information is registered by comparing the second registered term list 212 and the first registered term list 211. Therefore, special terms such as "universally used in-house terms" can be conveniently registered in the speech recognition dictionary, and the speech recognition dictionary can be updated by repeatedly executing the process. becomes.

The control unit 220 is composed of a CPU, an ASIC, and the like. The control unit 120 operates according to a program stored in the storage unit 110 and executes processing according to the program. The control unit 220 includes a speech recognition processing unit 221 and a certainty calculation unit 222 as main functional units provided by the program stored in the storage unit 210.

The voice recognition processing unit 221 has a function of converting, for example, voice data received from the information processing device 100 into text data based on the first registered term list 211 and text data based on the second registered term list 212, respectively. Department. Note that the conversion from voice data to text data may be performed using conventionally used voice recognition technology based on the first registered term list 211 and the second registered term list 212. Note that the voice recognition processing unit 221 also has a function of transmitting each converted text data to another information processing device 100 together with the voice data.

The confidence level calculation unit 222 is a functional unit that calculates the confidence level corresponding to the text data converted by the speech recognition processing unit 221. Specifically, the certainty factor calculation unit 222 calculates the certainty factor A of the text data based on the first registered term list 211 and the certainty factor B of the text data based on the second registered term list 212, respectively. For example, the confidence level is based on the degree of similarity between the audio feature amount (waveform, cycle, etc.) of the word information registered in the first registered term list 211 and the second registered term list 212 and the audio feature amount based on the received audio data. (calculated based on a predetermined calculation). Note that the confidence level calculation unit 222 also has a function of transmitting each calculated confidence level to other information processing apparatuses 100.

These functional units work together to convert the voice data received from the information processing device 100 into text data (speech recognition) in the voice recognition server 200, and perform other information processing along with the voice data as a voice recognition result. A function of transmitting data to the device 100 is realized. Furthermore, a function of transmitting the certainty factor to other information processing apparatuses 100 is realized.

The input/output unit 230 is a device for inputting and outputting data, and is composed of a keyboard, a mouse, a camera, a microphone, a liquid crystal display, an organic EL display, and the like.

The communication unit 240 is a device for communicating with the information processing apparatus 100 via the network 510.

The above is the configuration of the voice recognition server 200. Next, the operation of the information processing device 100 will be explained with reference to FIGS. 4 to 7. First, as an operation of the information processing system 1, the overall processing flow will be described with reference to FIG. In the illustrated example, a case will be described below in which the user of the information processing device 100B makes a statement of the content of example sentence 1 to the user of the information processing device 100A.

As shown in FIG. 4, when the user of the information processing device 100B inputs the voice of example sentence 1 into the input/output unit 130, it is converted into voice data by the function of the control unit 120, and the voice data is sent to the voice recognition server 200. ((1) in Figure 4). Note that in the illustrated example, in order to facilitate understanding, the voice data is transmitted from the information processing device 100B to the voice recognition server 200, but for example, the voice data is transmitted from the information processing device 100B to the information processing device 100A. and voice data may be transmitted, and specific voice data extracted by the information processing device 100A may be transmitted to the voice recognition server 200.

When the voice recognition server 200 receives voice data from the information processing device 100B, the voice recognition processing unit 221 performs voice recognition (converts to text data) based on the first registered term list 211, and converts the voice data into text data. The text data is transmitted to the information processing device 100A as the first voice recognition result ((2) in FIG. 4). Furthermore, the speech recognition server 200 uses the function of the confidence calculation unit 222 to calculate the confidence A of speech recognition based on the first registered term list 211, and transmits it to the information processing device 100A ((3) in FIG. 4). .

Further, the speech recognition server 200 performs speech recognition (converts to text data) based on the second registered term list 212 using the function of the speech recognition processing unit 221, and converts the speech data and text data into the second speech recognition result. ((4) in FIG. 4). Furthermore, the speech recognition server 200 uses the function of the confidence calculation unit 222 to calculate the confidence B of speech recognition based on the second registered term list 212, and transmits it to the information processing device 100A ((5) in FIG. 4). . Note that (2) to (5) in FIG. 4 may be performed all at once.

On the side of the information processing device 100A, voice data and text data based on the second registered term list 212 received from the voice recognition server 200 are outputted from the input/output unit 130 (the contents shown in FIG. 6(B) are outputted). ). Further, upon receiving the first voice recognition result and the second voice recognition result (including confidence level A and confidence level B) from the voice recognition server 200, the information processing device 100A sends the first voice recognition result and the second voice recognition result (including confidence level A and confidence level B) to the voice recognition result receiving means and the confidence level receiving means. equivalent), performs term registration processing to register special terms to be registered in the speech recognition dictionary. That is, the information processing device 100A performs processing for registering special terms included in the utterances of the user of the information processing device 100B in a speech recognition dictionary. Note that in the following, audio data and text data with the content shown in FIG. 6(A) are received as the first speech recognition result, and audio data and text data with the content shown in FIG. 6(B) are received as the second speech recognition result. A case will be described in which the second speech recognition result "NTT" is registered as a special term (assuming that the confidence level is also the value shown in the figure).

FIG. 5 is a flowchart illustrating an example of term registration processing. In the term registration process, the information processing device 100A uses the function of the certainty comparison unit 121 to determine whether the value obtained by subtracting the certainty A from the certainty B is equal to or greater than a predetermined threshold (predetermined conditions (step S101). If it is less than the threshold, the information processing device 100A concludes that there is no target to be registered and ends the term registration process. Specifically, in the process of step S101, the certainty factor 0.16 shown in FIG. 6(A) is subtracted from the certainty factor 0.89 shown in FIG. 6(B), and it is determined whether the certainty factor is equal to or greater than a threshold value. Note that the threshold value in this example is assumed to be set in advance by the user as 0.5.

If it is equal to or greater than the threshold value (step S101; Yes), the information processing device 100A uses the function of the morpheme extraction unit 122 to extract each of the first speech recognition result and the second speech recognition result received from the speech recognition server 200 for each morpheme. Then, different morphemes are extracted from the second speech recognition result (step S102). Note that in step S102, the text data of the first speech recognition result may be divided into morphemes, different morphemes may be extracted, and then the audio data of the portion corresponding to the morpheme may be extracted. Alternatively, both the text data and the audio data in the first speech recognition result may be divided into morphemes, and different morphemes may be extracted for each morpheme. Specifically, in step S102, since the morphemes of "Venditti" shown in FIGS. 6(A) and 7(A) and "NTT" shown in FIGS. 6(B) and 7(B) are different, B) and the morpheme of "NTT" shown in FIG. 7(B) are extracted. Note that "Venditti" shown in FIGS. 6A and 7A is a coined word used in this embodiment to facilitate understanding, and the part of speech is an adjective. In addition, in the following, it is assumed that the frequency of appearance of the "NTT" is 5 times, and the frequency of appearance in example sentence 1 is 6 times.

After executing the process of step S102, the information processing device 100A uses the function of the part-of-speech estimation unit 123 to compare the morphemes of the first speech recognition result and the morphemes of the second speech recognition result, and compares the morphemes of the second speech recognition result. A morpheme whose part of speech is a noun but whose morpheme in the first speech recognition result is other than a noun is extracted (step S103). Note that, as described above, in step S103, morphemes of different parts of speech may simply be output to the input/output unit 130, and the user may select whether or not to extract them. Specifically, in the process of step S103, since the part of speech of "Venditti" shown in FIG. 7(A) is "adjective" and the part of speech of "NTT" shown in FIG. 7(B) is "noun", The morpheme of "NTT" shown in FIG. 7(B) is extracted. Furthermore, in this embodiment, as shown in FIG. 7, a preposition such as "of" is not included in the comparison because it is an unnecessary part of speech from the perspective of registration in a speech recognition dictionary.

After executing the process in step S103, the information processing device 100A uses the function of the term classification unit 124 to determine whether or not the morpheme extracted in step S102 and the morpheme extracted in step S103 match (step S104). . If they do not match (step S104; No), the term registration process ends. If they do not match, the user is asked to select whether or not to register the word information corresponding to the morpheme extracted in step S102 and the morpheme extracted in step S103 in the registered term list 111, and neither is registered. If the term registration process is ended and at least one of the terms is registered, the process may proceed to step S105. In this embodiment, it is determined whether the morphemes extracted in the process of step S102 and the process of step S103 match, but only the process of step S102 or only the process of step S103 is performed, and the process of step S105 is performed. You may proceed to processing. Furthermore, if the process in steps S102 to S104 is not executed and the determination is Yes in step S101, the process may proceed to step S105. In this case, for example, the confidence level for each morpheme may be transmitted from the speech recognition server 200.

If it is determined that they match (step S104; Yes), the information processing device 100A uses the function of the term classification unit 124 to certify the word information corresponding to the extracted morpheme as a registration target, and uses the function of the term classification unit 124 to certify the word information corresponding to the extracted morpheme as a registration target. The word information is classified based on the classification based on the registered classification 112 (step S105). Specifically, in the process of step S105, since the word information of "NTT" appears six times, "NTT" is classified into the "universally used in-house term" category. Note that "universally used in-house terms" includes, for example, terms that are commonly used in multiple projects.

After executing the process of step S105, the information processing device 100A uses the function of the term registration unit 125 to input the audio data and text data as word information classified in the process of step S104 to the registered term list 111 according to the classification. (Step S106). Specifically, in the process of step S106, the audio data and text data of "NTT" classified into the "universally used in-house terminology" category are associated with each other, and the "universal Register as a classification of "internal terminology used in the company".

After executing the process of step S106, the information processing device 100A uses the function of the term registration unit 125 to update the contents of the first registered term list 211 based on the contents of the word information registered in the registered term list 111. An update instruction is sent to the speech recognition server 200 (step S107), and the term registration process is ended. Specifically, in the process of step S107, the voice data and text data of "NTT" registered as a category of "universally used in-house terms" in the registered term list 111 are sent to the voice recognition server 200 along with an update instruction. Then, the voice data and text data of "NTT" are additionally registered in the first registered term list 211 stored in the voice recognition server 200. As a result, the contents of the first registered term list 211 will be updated.

Returning to FIG. 4, the voice recognition server 200 updates the contents of the first registered term list 211 based on receiving the update instruction from the information processing device 100A. Although not shown, if the user of the information processing device 100A speaks to the user of the information processing device 100B after this, the function of the control unit 120 of the information processing device 100A converts it into voice data. The voice data is sent to the voice recognition server 200. Then, term registration processing is performed on the information processing device 100B side, and the contents of the first registered term list 211 in the speech recognition server 200 are updated. Such processing will be repeatedly executed until the conversation at the conference, lecture, etc. ends. In this way, term registration processing is performed for each conversation and the contents of the first registered term list 211 are updated, so the dictionary for speech recognition is updated in real time, and the dictionary for speech recognition can be suitably used. can be generated. In this embodiment, a conversation between two parties is taken as an example, but the same applies to a conversation between three or more parties. Furthermore, the dictionary generated in this manner can also be used as a dictionary in known Japanese input software.

(Modified example)
Note that the present invention is not limited to the embodiments described above, and various modifications and applications are possible. For example, the information processing device 100 does not need to have all the technical features described in the above embodiments, but may include some of the technical features described in the above embodiments so that at least one problem in the conventional technology can be solved. It may also have the structure of a section. Furthermore, at least a portion of each of the following modified examples may be combined.

In the above embodiment, an example was shown in which the process of step S107 in FIG. It may also be done. For example, the process of step S107 shown in FIG. 5 may be executed by the user operating the input/output unit 130 at the timing when the conversation ends. Further, for example, when a predetermined specific voice (multiple voices may be set) such as "end" is received, it is determined that the conversation is over and the process of step S107 in FIG. 5 is executed. Good too. Alternatively, the process of step S107 in FIG. 5 may be executed every time the number of word information set by the user is registered in the registered term list 111. According to these, the load on the updating process of the first registered term list 211 can be reduced.

Furthermore, for example, the word "PoC" may be pronounced differently depending on the user, such as one user who pronounces it as "P-OC" and another who pronounces it as "Pock." Regarding such words, a plurality of audio data are registered in advance as the second registered term list 212 in correspondence with one text data, and in step S106 of FIG. The audio data may be registered in the registered term list 111. Then, in the process of step S107, an instruction to update the first registered term list 211 with the content may be issued. According to this, it is possible to suitably generate a dictionary for speech recognition even when a word is pronounced differently depending on the user.

Furthermore, the information processing device 100 may include the configuration of the voice recognition server 200 in the above embodiment. In this case, in step S107 of FIG. 5, the first registered term list 211 stored in the own storage unit 110 is updated to synchronize with the first registered term list 211 stored in the other information processing device 100. Just do it.

In addition, when the above-mentioned functions are realized by sharing between an OS (Operating System) and an application, or by cooperation between the OS and an application, only the parts other than the OS may be stored on the medium.

It is also possible to superimpose a program on a carrier wave and distribute it via a communication network. For example, the program may be posted on a bulletin board (BBS, Bulletin Board System) on a communication network and distributed via the network. The above-described process may be executed by starting these programs and executing them under the control of the operating system in the same way as other application programs.

1 Information processing system, 100, 100A, 100B Information processing device, 110, 210 Storage unit, 111 List of registered terms, 112 Registered classification, 120, 220 Control unit, 121 Confidence comparison unit, 122 Morphological extraction unit, 123 Part of speech estimation unit , 124 term classification unit, 125 term registration unit, 130, 230 input/output unit, 140, 240 communication unit, 200 speech recognition server, 211 first registered term list, 212 second registered term list, 221 speech recognition processing unit, 222 Confidence calculation unit, 510 network

Claims (6)

A speech recognition result that receives a first speech recognition result based on a first dictionary and a second speech recognition result based on a second dictionary that is different from the first dictionary and includes word information generated by the user. receiving means;
Receive a first certainty factor about the first speech recognition result calculated based on a predetermined calculation and a second certainty factor about the second speech recognition result calculated based on the calculation. certainty level receiving means;
As a result of comparing the first certainty factor and the second certainty factor, if the difference in the certainty factor is larger than a predetermined value, the word information included in the second speech recognition result is stored in the first dictionary. word information storage means for storing it as an update list ;
Transmitting means for transmitting an update instruction to add word information included in the update list stored in the word information storage means to the first dictionary;
An information processing device comprising : further comprising an extraction means for extracting word information to be stored from the second speech recognition result in accordance with a predetermined standard when the difference in the certainty factor is large;
The word information storage means stores the word information extracted by the extraction means.
The information processing device according to claim 1, characterized in that: Further comprising a classification means for classifying the word information extracted by the extraction means into one of a plurality of predetermined classifications for each appearance frequency,
The word information storage means stores word information classified by the classification means for each classification.
The information processing device according to claim 2, characterized in that: The word information includes audio information and text information,
further comprising a first dictionary updating means for updating the first dictionary by adding word information stored as the update list by the word information storage means to the first dictionary;
The second dictionary is newly stored by the user's operation every time the first dictionary is updated.
The information processing device according to any one of claims 1 to 3, characterized in that: A speech recognition result that receives a first speech recognition result based on a first dictionary and a second speech recognition result based on a second dictionary that is different from the first dictionary and includes word information generated by the user. a receiving step;
Receive a first certainty factor about the first speech recognition result calculated based on a predetermined calculation and a second certainty factor about the second speech recognition result calculated based on the calculation. a confidence level receiving step;
As a result of comparing the first certainty factor and the second certainty factor, if the difference in the certainty factor is larger than a predetermined value, the word information included in the second speech recognition result is stored in the first dictionary. a step of storing word information as an update list ;
a sending step of sending an update instruction to add the word information included in the update list stored in the word information storage step to the first dictionary;
An information processing method comprising : computer,
A speech recognition result that receives a first speech recognition result based on a first dictionary and a second speech recognition result based on a second dictionary that is different from the first dictionary and includes word information generated by the user. receiving means,
Receive a first certainty factor about the first speech recognition result calculated based on a predetermined calculation and a second certainty factor about the second speech recognition result calculated based on the calculation. confidence receiving means;
As a result of comparing the first certainty factor and the second certainty factor, if the difference in the certainty factor is larger than a predetermined value, the word information included in the second speech recognition result is stored in the first dictionary. word information storage means for storing it as an update list ;
Transmitting means for transmitting an update instruction to add word information included in the update list stored in the word information storage means to the first dictionary;
to function as
A program characterized by:

Images