JP6927905B2 - Decision device, decision method, decision program and program - Google Patents

Description

The present invention relates to a determination device, a determination method, a determination program and a program .

Conventionally, a technique related to voice recognition for a user's utterance has been provided. For example, when a voice from a user is input and the voice recognition result is insufficient in certainty, there is known a technique for displaying a request for further input from the user.

Japanese Unexamined Patent Publication No. 2000-148181

However, it is not always possible to improve the accuracy of voice recognition by the above-mentioned conventional technique. For example, in the above-mentioned prior art, when a voice from a user is input and the voice recognition result is insufficient in certainty, only a display requesting the user for further input is performed, and the accuracy of voice recognition is improved. It is not always possible to improve.

The present application has been made in view of the above, and an object of the present application is to provide a determination device, a determination method, a determination program, and a model capable of improving the accuracy of speech recognition.

The determination device according to the present application includes an estimation unit that estimates the utterance content corresponding to the voice uttered by the user, estimation accuracy in the first estimation result estimated from the first voice by the estimation unit, and the first voice. Subsequently, a determination unit that determines the voice recognition result corresponding to the first voice and the second voice based on the estimation accuracy in the second estimation result estimated from the second voice repeatedly uttered by the user. It is characterized by having.

According to one aspect of the embodiment, there is an effect that the accuracy of voice recognition can be improved.

FIG. 1 is a diagram showing an example of a determination process executed by the determination device according to the embodiment. FIG. 2 is a diagram showing an example of a voice recognition result determination process executed by the determination device according to the embodiment. FIG. 3 is a diagram showing a configuration example of the determination device according to the embodiment. FIG. 4 is a diagram showing an example of a voice information storage unit according to the embodiment. FIG. 5 is a diagram showing an example of the estimation result information storage unit according to the embodiment. FIG. 6 is a diagram showing an example of the score information storage unit according to the embodiment. FIG. 7 is a flowchart showing an example of the flow of the determination process executed by the determination device according to the embodiment. FIG. 8 is a diagram showing an example of determining the voice recognition result according to the embodiment. FIG. 9 is a diagram showing an example of a determination process executed by the determination device according to the modified example. FIG. 10 is a hardware configuration diagram showing an example of a computer that realizes the function of the determination device.

Hereinafter, modes for implementing the determination device, determination method, determination program, and model according to the present application (hereinafter, referred to as “embodiments”) will be described in detail with reference to the drawings. It should be noted that this embodiment does not limit the determination device, determination method, determination program and model according to the present application. In addition, each embodiment can be appropriately combined as long as the processing contents do not contradict each other. Further, in each of the following embodiments, the same parts are designated by the same reference numerals, and duplicate description is omitted.

[1. An example of the determination process indicated by the determination device]
An example of the determination process executed by the determination apparatus according to the embodiment will be described with reference to FIG. FIG. 1 is a diagram showing an example of a determination process executed by the determination device according to the embodiment. FIG. 1 shows an example in which a determination process is executed by the determination device 100.

As shown in FIG. 1, the determination system 1 includes a terminal device 10 and a determination device 100. The terminal device 10 and the determination device 100 are communicably connected by wire or wirelessly via a predetermined communication network (not shown). The determination system 1 shown in FIG. 1 may include a plurality of terminal devices 10 and a plurality of determination devices 100.

The terminal device 10 is an information processing device used by a user who accesses web contents such as a web page displayed on a browser and contents displayed on an application. For example, the terminal device 10 is a desktop PC (Personal Computer), a notebook PC, a tablet terminal, a mobile phone, a PDA (Personal Digital Assistant), or the like. Further, the terminal device 10 acquires the voice uttered by the user according to the operation by the user or the function of the terminal device 10, and stores the information related to the voice in the predetermined storage area of the terminal device 10.

For example, it is assumed that the terminal device 10 has an application capable of acquiring voice installed in advance. In this case, the terminal device 10 acquires the voice uttered by the user by the built-in microphone, and stores the information related to the voice in the predetermined storage area of the terminal device 10. The above processing by the terminal device 10 can be realized by a well-known technique such as voice search.

The determination device 100 is a determination device that determines a voice recognition result corresponding to a voice uttered by the user and provides the voice recognition result to the terminal device 10, and is realized by, for example, a server device or the like. This point will be described. In the conventional speech recognition analysis, when the speech recognition is difficult, the speech recognition result corresponding to the speech uttered by the user may be provided to the user as a deviation from the speech. Further, in the conventional voice recognition analysis, when voice recognition is difficult, the user may be made to repeat the utterance, or the user may not respond to the voice uttered by the user. In this case, the user may repeatedly utter the same voice. As a result, the user is forced to repeatedly utter the same voice. Therefore, the determination device 100 according to the embodiment realizes improvement in the accuracy of voice recognition for the voice repeatedly spoken by the user. Specifically, the determination device 100 determines the estimation result of the first voice estimated from the voice first uttered by the user (hereinafter, may be referred to as the first voice) and the voice subsequently uttered by the user. The voice recognition result is determined based on the score for each combination with the estimation result of the second voice estimated from (hereinafter, may be referred to as the second voice). This point will be described in detail below. In the following, when the user performs a voice search on the terminal device 10, the voice uttered by the user a plurality of times will be described as having the same content.

Hereinafter, an example of the determination process of the voice recognition result by the determination device 100 will be described along the flow with reference to FIG.

First, as shown in FIG. 1, the user utters the "associative game" twice to the terminal device 10 (step S1). For example, it is assumed that the user speaks "associative game" to the terminal device 10 in order to search for the type of game related to "associative game". In this case, the user is worried about whether or not the voice spoken by the user himself / herself is recognized by the terminal device 10, and spontaneously speaks the "associative game" twice. Then, the terminal device 10 accepts the "associative game" uttered twice by the user.

In addition, for example, when the voice recognition by the determination device 100 is difficult, the terminal device 10 uses voice navigation such as "Please say again" based on the information regarding the voice recognition failure notified by the decision device 100. The user may be prompted to speak repeatedly.

Subsequently, the determination device 100 receives the voice repeatedly spoken by the user (step S2). For example, the determination device 100 receives the voice repeatedly spoken by the user based on transmitting information to the determination device 100 regarding the fact that the terminal device 10 has received the voice spoken by the user. In this case, the determination device 100 determines whether or not the voice spoken by the user is repeatedly spoken. Specifically, the determination device 100 determines whether or not the voice spoken by the user has been repeatedly spoken based on the similarity between the voice waveform of the voice spoken by the user and the voice waveform of another voice. ..

For example, when the similarity between the voice waveform of the voice spoken by the user and the voice waveform of another voice is equal to or higher than a predetermined value, the determination device 100 determines that the voice spoken by the user has been repeatedly spoken. Even if the determination device 100 corrects the voice waveform of the voice by amplitude, phase, frequency, etc. before making a determination based on the similarity between the voice waveform of the voice and the voice waveform of another voice. good. As a result, the determination device 100 can determine the similarity between the voice waveforms of the voices having different timings and volumes.

The processing executed by the determination device 100 is, for example, by correcting the amplitude, phase, frequency, etc. of the voice waveform with respect to the voice waveform and calculating the similarity between the voice waveforms of the voice by using a well-known technique related to voice signal processing and the like. It is feasible. Further, for example, in the process of determining the similarity between voice waveforms executed by the determination device 100, it may be determined whether or not the voice uttered by the user is repeated by a well-known technique such as machine learning.

Subsequently, the determination device 100 estimates the utterance content corresponding to the voice repeatedly spoken by the user, and ranks the estimation result estimated from the voice (step S3). Specifically, the determination device 100 estimates the utterance content corresponding to the first voice among the voices determined to have been repeatedly uttered by the user. Then, the determination device 100 estimates the utterance content of the first voice and calculates the estimation accuracy, which is information indicating the accuracy of the estimation result of the first voice. In addition, the determination device 100 estimates the utterance content corresponding to the second voice among the voices determined to have been repeatedly uttered by the user. Then, the determination device 100 estimates the utterance content of the second voice and calculates the estimation accuracy of the estimation result of the second voice.

For example, the determination device 100 estimates a "performance game", a "chlorine game", and an "associative game" as estimation results estimated from the first voice corresponding to the voice "associative game" uttered by the user. Then, the determination device 100 has "1.0", "0.9", and "0" as the estimation accuracy corresponding to the "performance game", "chlorine game", and "associative game" which are the estimation results of the first sound. It is calculated as "0.8". Further, the determination device 100 has a "cleaning game", an "associative game", and an "inflammation game" as estimation results estimated from the second voice corresponding to the voice "associative game" uttered by the user following the first voice. Is estimated. Then, the determination device 100 has "1.0", "0.9", and "0" as the estimation accuracy corresponding to the "cleaning game", "associative game", and "inflammation game" which are the estimation results of the second voice. It is calculated as "0.8".

Then, the determination device 100 ranks the estimation results estimated from the voice uttered by the user based on the estimation accuracy of the estimation results. For example, the determination device 100 ranks the "performance game", the "chlorine game", and the "associative game" in this order based on the estimation accuracy corresponding to the estimation result of the first voice. Further, the determination device 100 ranks the "cleaning game", the "associative game", and the "inflammation game" in this order based on the estimation accuracy corresponding to the estimation result of the second voice.

In the above processing executed by the determination device 100, for example, the utterance content corresponding to the voice uttered by the user is estimated by a well-known technique related to voice recognition analysis and the like, and the estimation accuracy of the estimation result estimated from the voice is estimated. It can be realized by calculating.

Subsequently, the determination device 100 calculates a score for each combination of the estimation result of the first voice and the estimation result of the second voice (step S4). For example, the determination device 100 calculates the score SC1 by the following formula (1).

Score SC1 = Acc1 + Acc2 + Rep ... (1)

In the above equation (1), "Acc1" indicates the estimation accuracy of the estimation result of the first voice, "Acc2" indicates the estimation accuracy of the estimation result of the second voice, and "Rep" indicates the estimation accuracy of the first voice. Information on the degree of duplication of words contained in the estimation result and the estimation result of the second voice (hereinafter referred to as the degree of duplication) is shown. Then, the determination device 100 determines the combination of the estimation result of the first voice and the estimation result of the second voice having the highest score SC1.

Here, an example of the process of determining the voice recognition result will be described with reference to FIG. FIG. 2 is a diagram showing a combination of the estimation result of the first voice and the estimation accuracy of the estimation result, and the estimation result of the second voice and the estimation accuracy of the estimation result. For example, in the example shown in FIG. 2, as shown in the estimation result WT1 of the first voice, the “performance game” has an estimation accuracy of “1.0”.

In the example shown in FIG. 2, the determination device 100 calculates a score in which the estimation accuracy of the estimation result of the first voice, the estimation accuracy of the estimation result of the second voice, and the multiplicity are added by the above equation (1). For example, in a combination in which the estimation result of the first voice is a "playing game" and the estimation result of the second voice is a "cleaning game" (hereinafter, may be referred to as "playing game x cleaning game"), the morphological element Since the "games" match by analysis or the like, it is assumed that the degree of duplication is calculated as "1.0". In this case, the determination device 100 uses the formula shown in the calculation formula score CT1 in FIG. 2 to determine the estimation accuracy “1.0” of the first voice estimation result WT1 “playing game” and the second voice estimation result WT2 “. Since the estimation accuracy of the "cleaning game" is "1.0" and the degree of overlap is "1.0", the score CT1 is calculated as "3.0".

Further, for example, in the "associative game x associative game", since the "associative" and the "game" match by morphological analysis or the like, the degree of duplication is calculated to be "2.0". In this case, the determination device 100 uses the formula shown in the calculation formula score CT2 in FIG. 2 to determine the estimation accuracy “0.8” of the first voice estimation result WT3 “associative game” and the second voice estimation result WT4 “. Since the estimation accuracy of the "associative game" is "0.9" and the degree of overlap is "2.0", the score is calculated as CT2 "3.7".

Further, for example, in the example shown in FIG. 2, the scores CT3 to CT9 are calculated based on the above calculation method in the following combinations.
Score CT3 ("playing game x associative game") = 2.9 ... (2)
Score CT4 ("playing game x inflammation game") = 2.8 ... (3)
Score CT5 ("Chlorine game x cleaning game") = 2.9 ... (4)
Score CT6 ("chlorine game x associative game") = 2.8 ... (5)
Score CT7 ("Chlorine game x inflammation game") = 2.7 ... (6)
Score CT8 ("associative game x cleaning game") = 2.8 ... (7)
Score CT9 ("associative game x inflammation game") = 2.6 ... (8)

Then, the determination device 100 compares the scores CT1 to CT9. For example, the determination device 100 has the largest score CT2 between the first voice estimation result WT3 "associative game" and the second voice estimation result WT4 "associative game", so that the first voice estimation result WT3 and the second The voice recognition result is selected from the voice estimation result WT4, and the "associative game" is determined as the voice recognition result.

Returning to FIG. 1, an example of the determination process according to the embodiment will be described. The determination device 100 determines the voice recognition result C1 as the "associative game" based on the score SC1 (step S5). For example, the determination device 100 has a score SC1 of "3.7" in the "associative game x associative game" and a score SC1 of "3.0" in the "playing game x cleaning game", and thus the "associative game". Since the score SC1 of the "associative game" is the largest, the "associative game" selected from the estimation result of the first voice and the estimation result of the second voice is determined as the voice recognition result. Then, the determination device 100 provides the voice recognition result C1 “associative game” to the terminal device 10 (step S6). For example, the determination device 100 converts the voice recognition result C1 “associative game” into text and provides it to the terminal device 10. Then, the terminal device 10 displays the text "associative game" to the user. The terminal device 10 may read aloud "associative game" to the user by using the voice reading function.

As described above, the determination device 100 according to the embodiment has an estimation result estimated from the first voice corresponding to the voice uttered by the user and a second voice corresponding to the voice uttered following the first voice by the user. The speech recognition result is determined based on the score corresponding to the combination with the estimation result estimated from. As a result, the determination device 100 according to the embodiment can improve the accuracy of voice recognition. This point will be described. Explaining with reference to the example of FIG. 1, the determination device 100 selects the estimation result of the first voice having high estimation accuracy from the estimation results of the plurality of first voices, and among the estimation results of the plurality of second voices. , Select the estimation result with high estimation accuracy. Then, the determination device 100 calculates the score based on the degree of overlap for each combination of the estimation result of the first voice and the estimation result of the second voice with high estimation accuracy. As a result, the determination device 100 determines the estimation result of the first voice and the estimation result of the second voice in the combination having a high score as the voice recognition result, so that the voice recognition result can be determined with higher accuracy. Therefore, the determination device 100 can determine the voice recognition result based on the score calculated for each combination of voices repeatedly uttered by the user, so that the accuracy of voice recognition can be improved.

[2. Configuration of determination device]
Next, the configuration of the determination device 100 according to the embodiment will be described with reference to FIG. FIG. 3 is a diagram showing a configuration example of the determination device 100 according to the embodiment. As shown in FIG. 3, the determination device 100 includes a communication unit 110, a storage unit 120, and a control unit 130.

(About communication unit 110)
The communication unit 110 is realized by, for example, a NIC (Network Interface Card) or the like. Then, the communication unit 110 is connected to the network by wire or wirelessly, and transmits / receives information to / from the terminal device 10.

(About storage unit 120)
The storage unit 120 is realized by, for example, a semiconductor memory element such as a RAM (Random Access Memory) or a flash memory (Flash Memory), or a storage device such as a hard disk or an optical disk. The storage unit 120 includes a voice information storage unit 121, an estimation result information storage unit 122, and a score information storage unit 123.

(About voice information storage unit 121)
The voice information storage unit 121 according to the embodiment stores information related to voice spoken by the user. Here, FIG. 4 shows an example of the voice information storage unit 121 according to the embodiment. In the example shown in FIG. 4, the voice information storage unit 121 has items such as "voice ID (Identifier)", "voice", and "similarity with the first voice".

The "voice ID" is an identifier that identifies the voice spoken by the user. The “voice” is voice waveform data associated with the “voice ID”. The "similarity with the first voice" is information regarding the similarity with the voice received immediately before being associated with the "voice ID". For example, in FIG. 4, “V2” identified by the voice ID is similar to the first voice, which is the voice “WV2” and has a degree of similarity to the voice “WV1” identified by the voice ID “V1”. The degree is "0.9".

(About voice recognition result information storage unit 122)
The estimation result information storage unit 122 according to the embodiment stores various information regarding the estimation result corresponding to the voice uttered by the user. Here, FIG. 5 shows an example of the estimation result information storage unit 122 according to the embodiment. In the example shown in FIG. 5, the estimation result information storage unit 122 has items such as "voice ID", "estimation result ID", "ranking order of estimation result", "estimation result", and "estimation accuracy".

The "voice ID" is an identifier that identifies the voice spoken by the user. The "estimation result ID" indicates identification information for identifying the estimation result estimated from the voice. The "ranking rank of the estimation result" indicates the ranking ranked by the magnitude of the estimation accuracy calculated together with the estimation result corresponding to the voice. The "estimation result" indicates an estimation result corresponding to the voice spoken by the user. "Estimation accuracy" indicates the estimation accuracy calculated together with the "estimation result". For example, in FIG. 5, the ranking rank of the estimation result corresponding to "V1" identified by the voice ID and corresponding to "VC1" identified by the estimation result ID is "1st place", and the estimation result " The estimation accuracy of the "playing game" is "1.0".

(About score information storage unit 123)
The score information storage unit 123 according to the embodiment stores information regarding the score for each combination of the estimation result of the first voice and the estimation result of the second voice. Here, FIG. 6 shows an example of the score information storage unit 123 according to the embodiment. In the example shown in FIG. 6, the score information storage unit 123 has items such as “score ID”, “combination of estimation results”, and “score”.

The “score ID” indicates identification information for identifying the score corresponding to the combination of the estimation result of the first voice and the estimation result of the second voice. The "combination of estimation results" indicates a score corresponding to the combination of the estimation result of the first voice and the estimation result of the second voice. For example, in FIG. 6, the combination of estimation results corresponding to "SCC1" identified by the score ID is "associative game x associative game", and the score in the combination of such estimation results is "3.7". show.

(About control unit 130)
The control unit 130 is a controller, and for example, various programs (as an example of a determination program) stored in a storage device inside the determination device 100 by a CPU (Central Processing Unit), an MPU (Micro Processing Unit), or the like. (Equivalent) is realized by executing the RAM as a work area. Further, the control unit 130 is a controller, and is realized by, for example, an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array).

As shown in FIG. 3, the control unit 130 includes a reception unit 131, a determination unit 132, an estimation unit 133, a calculation unit 134, a determination unit 135, and a provision unit 136, and the information described below. Realize or execute the function or action of processing. The internal configuration of the control unit 130 is not limited to the configuration shown in FIG. 3, and may be another configuration as long as it is a configuration for performing information processing described later. Further, the connection relationship of each processing unit included in the control unit 130 is not limited to the connection relationship shown in FIG. 3, and may be another connection relationship.

(About reception desk 131)
The reception unit 131 receives the voice uttered by the user. For example, the reception unit 131 receives the voice uttered by the user based on transmitting the information regarding the reception of the voice uttered by the user by the terminal device 10 to the reception unit 131. Then, the reception unit 131 stores the waveform data of the voice in the voice information storage unit 121.

(About judgment unit 132)
In the determination unit 132, the second voice follows the first voice based on the similarity between the first voice received by the reception unit 131 and the second voice received after the first voice by the reception unit 131. Determine if the voice is spoken repeatedly. Specifically, the determination unit 132 refers to the voice information storage unit 121, stores the similarity between the first voice and the second voice in the voice information storage unit 121, and the second voice follows the first voice. It is determined that the voice is repeatedly spoken.

For example, it is assumed that the first voice "WV1" and the second voice "WV2" are identified by the voice ID. Further, it is assumed that the predetermined value of the similarity is 0.8. In this case, the determination unit 132 stores the similarity “0.9” between the voice waveform of the first voice “WV1” and the voice waveform of the second voice “WV2” in the voice information storage unit 121. Then, the determination unit 132 determines that the first voice "WV1" and the second voice "WV2" have been repeatedly uttered because the similarity is equal to or higher than a predetermined value.

(About estimation unit 133)
The estimation unit 133 estimates the utterance content corresponding to the voice uttered by the user. Specifically, the estimation unit 133 refers to the voice information storage unit 121 and estimates the utterance content corresponding to the first voice, which is the voice first spoken by the user. Then, the estimation unit 133 estimates the utterance content of the first voice and calculates the estimation accuracy of the estimation result of the first voice. Then, the estimation unit 133 stores the estimation result and the estimation accuracy of the first voice in the estimation result information storage unit 122. The estimation unit 133 estimates the utterance content based on, for example, information on the correspondence between the wording and parameters such as the waveform and amplitude of the voice. Then, the estimation unit 133 calculates the estimation accuracy based on, for example, the degree of coincidence between the waveform of the voice and the waveform of the estimation result.

In addition, the estimation unit 133 refers to the voice information storage unit 121 to estimate the utterance content corresponding to the second voice, which is the voice uttered by the user following the first voice. Then, the estimation unit 133 estimates the utterance content of the second voice and calculates the estimation accuracy of the estimation result of the second voice. Then, the estimation unit 133 stores the estimation result and the estimation accuracy of the second voice in the estimation result information storage unit 122.

For example, it is assumed that the first voice "WV1" and the second voice "WV2" are identified by the voice ID. In this case, the estimation unit 133 refers to the voice information storage unit 121 and estimates the "performance game", the "chlorine game", and the "associative game" as the estimation result of the first voice corresponding to the first voice "WV1". do. Then, the estimation unit 133 sets the estimation accuracy corresponding to the first sound estimation result of the "performance game", "chlorine game", and "associative game" to be "1.0", "0.9", and "0". It is calculated as "0.8". Then, the estimation unit 133 stores the estimation result and the estimation accuracy of the first voice in the estimation result information storage unit 122. Further, the estimation unit 133 refers to the voice information storage unit 121 and estimates the "cleaning game", the "associative game", and the "inflammation game" as the estimation result of the second voice corresponding to the second voice "WV2". .. Then, the estimation unit 133 sets the estimation accuracy corresponding to the estimation result of the second voice, "cleaning game", "associative game", and "inflammation game", to be "1.0", "0.9", and "0". It is calculated as "0.8". Then, the estimation unit 133 stores the estimation result and the estimation accuracy of the second voice in the estimation result information storage unit 122.

In addition, the estimation unit 133 ranks the estimation results estimated from the voice uttered by the user based on the estimation accuracy of the estimation results. Then, the estimation unit 133 stores the ranking information of the estimation result in the estimation result information storage unit 122. For example, the estimation unit 133 ranks the "performance game", the "chlorine game", and the "associative game" in this order based on the estimation accuracy corresponding to the estimation result of the first voice. Then, the estimation unit 133 stores the ranking information of the estimation result of the first voice in the estimation result information storage unit 122. Further, the estimation unit 133 ranks the "cleaning game", the "associative game", and the "inflammation game" in this order based on the estimation accuracy corresponding to the estimation result of the second voice. Then, the estimation unit 133 stores the ranking information of the estimation result of the second voice in the estimation result information storage unit 122.

(About calculation unit 134)
The calculation unit 134 uses the calculation unit 134 for each combination of the first estimation result (corresponding to the estimation result of the first voice) and the second estimation result (corresponding to the estimation result of the second voice) estimated by the estimation unit 133. The score is calculated based on the estimation accuracy of the estimation result of the first voice and the estimation accuracy of the estimation result of the second voice. For example, the calculation unit 134 reads out the “associative game × associative game” and the “playing game × cleaning game” from the estimation result information storage unit 122. For example, in the "associative game x associative game", since the "associative" and the "game" match by morphological analysis or the like, the degree of duplication is calculated to be "2.0". In this case, the calculation unit 134 has an estimation accuracy of "0.8" for the estimation result "associative game" of the first voice in the "associative game x associative game", and the estimation result "associative game" for the second voice. Since the estimation accuracy is "0.9" and the degree of overlap is "2.0", the score SC1 is calculated as "3.7". Further, for example, in the "playing game x cleaning game", since the "games" match by morphological analysis or the like, the degree of duplication is calculated to be "1.0". In this case, the calculation unit 134 has an estimation accuracy of "1.0" for the estimation result "playing game" of the first sound in the "playing game x cleaning game", and the estimation result "cleaning game" for the second sound. Since the estimation accuracy is "1.0" and the degree of overlap is "1.0", the score SC1 is calculated as "3.0". Then, the calculation unit 134 stores the score SC1 corresponding to each combination in the score information storage unit 123.

(About decision unit 135)
The determination unit 135 has the estimation accuracy of the first estimation result (corresponding to the estimation result of the first voice) estimated from the first voice by the estimation unit 133, and the second voice repeatedly uttered by the user following the first voice. The voice recognition result corresponding to the first voice and the second voice is determined based on the estimation accuracy in the second estimation result (corresponding to the estimation result of the second voice) estimated from. Specifically, the determination unit 135 converts the combination of the estimation result of the first voice and the estimation result of the second voice having the largest score SC1 stored in the score information storage unit 123 into the first voice and the second voice. Determine the corresponding speech recognition result. For example, the determination unit 135 refers to the score information storage unit 123, and the score SC1 in the “associative game × associative game” is “3.7”, and the score SC1 in the “playing game × cleaning game” is “3. Since it is "0", the score SC1 of "associative game x associative game" is the largest, so the "associative game" selected from the estimation result of the first voice and the estimation result of the second voice is determined as the voice recognition result. ..

(About the provider 136)
The providing unit 136 provides the terminal device 10 with the voice recognition result determined by the determining unit 135. For example, assume that the voice recognition result is an "associative game". In this case, the providing unit 136 converts the "associative game" into text and provides it to the terminal device 10. Then, the terminal device 10 displays the text "associative game" to the user.

[3. Flowchart of decision processing]
Next, the procedure of the determination process executed by the determination device 100 according to the embodiment will be described with reference to FIG. 7. FIG. 7 is a flowchart showing an example of the flow of the determination process executed by the determination device according to the embodiment.

As shown in FIG. 7, the reception unit 131 receives the voice uttered by the user (step S101). Then, when the similarity between the first voice and the second voice is equal to or higher than a predetermined value, the determination unit 132 determines that the first voice and the second voice have been repeatedly uttered (step S102). When the determination unit 132 does not determine that the voice is repeatedly spoken by the user (step S102; No), the determination unit 132 waits until the voice spoken by the user is received.

On the other hand, when the determination unit 132 determines that the voice is repeatedly spoken by the user (step S102; Yes), the estimation unit 133 determines the utterance content corresponding to the first voice and the second voice determined by the determination unit 132. Estimate and rank the estimation results (step S103).

The determination unit 135 determines the estimation result of the first voice and the estimation result of the second voice having the highest score calculated by the calculation unit 134 in the combination of the estimation results estimated by the estimation unit 133 as the voice recognition result (step S104). ). Then, the providing unit 136 provides the terminal device 10 with the voice recognition result determined by the determining unit 135 (step S105).

[4. Speech recognition result determination process]
Next, the determination of the voice recognition result in the determination system 1 according to the embodiment will be described with reference to FIG. FIG. 8 is a diagram showing an example of determining the voice recognition result according to the embodiment.

A case where the estimation result of the first voice and the estimation result of the second voice are different from each other will be described with reference to FIG. FIG. 8 is a diagram showing a combination of the estimation result of the first voice and the estimation accuracy of the estimation result, and the estimation result of the second voice and the estimation accuracy of the estimation result. In the example shown in FIG. 8, “Matsushima” has an estimation accuracy of “1.0” as shown in the estimation result WT5 of the first voice.

In the example shown in FIG. 8, the determination device 100 calculates the score in consideration of the estimation accuracy of the estimation result of the first voice, the estimation accuracy of the estimation result of the second voice, and the degree of overlap. For example, in "Matsushima x Yatsushima", the multiplicity is calculated as "3.0" because "tsu", "shi", and "ma" match by analyzing the multiplicity of characters. It shall be done. In this case, the determination device 100 uses the formula shown in the calculation formula score CT21 in FIG. 8 to determine the estimation accuracy “1.0” of the first voice estimation result WT5 “Matsushima” and the second voice estimation result WT6 “. Since the estimation accuracy of "Yatsushima" is "0.9" and the degree of overlap is "3.0", the score is calculated as CT21 "4.9".

Further, for example, in the example shown in FIG. 8, the scores CT22 to CT24 are calculated based on the above calculation method in the following combinations.
Score CT22 ("Matsushima x Hatsushima") = 4.8 ... (9)
Score CT23 ("Tatsushima x Yatsushima") = 4.8 ... (10)
Score CT24 ("Tatsushima x Hatsushima") = 4.6 ... (11)

Then, the determination device 100 compares each score CT21 to 24. For example, the determination device 100 has the largest score CT21 between the first voice estimation result WT5 "Matsushima" and the second voice estimation result WT6 "Yatsushima", so that the first voice estimation result WT5 and the second voice The voice estimation result WT6 is determined as the voice recognition result.

[5. Modification example]
The determination device 100 described above may be implemented in various different forms other than the above embodiment. Therefore, another embodiment of the determination device 100 will be described below.

[5-1. Determining device (1)]
Each component of each of the illustrated devices is a functional concept and does not necessarily have to be physically configured as shown in the figure. That is, the specific form of distribution / integration of each device is not limited to the one shown in the figure, and all or part of the device is functionally or physically dispersed / physically distributed in any unit according to various loads and usage conditions. Can be integrated and configured. For example, the determination device 100 is dispersed into a reception device including a reception unit 131 and a determination unit 132, and a determination device including an estimation unit 133, a calculation unit 134, a determination unit 135, and a provision unit 136. May be good.

[5-2. Determining device (2)]
In the above embodiment, an example of the determination process in which the determination device 100 determines the voice recognition result based on the score in the combination of the estimation results corresponding to the voice uttered twice by the user has been described. It is not limited to this. For example, the determination process executed by the determination device 100 may be executed standalone by the terminal device 10.

[5-3. Number of utterances]
In the above embodiment, an example of the determination process in which the determination unit 135 of the determination device 100 determines the voice recognition result based on the score for each combination of the estimation results corresponding to the voice uttered twice by the user has been described. The number of utterances is not limited to this. For example, the determination unit 135 may determine the voice recognition result based on the score in the combination of the estimation results corresponding to the voice uttered by the user three or more times.

[5-4. Determination process〕
In the above embodiment, the determination unit 132 of the determination device 100 repeatedly utters the second voice following the first voice based on the similarity between the first voice and the second voice received after the first voice. Although an example of the determination process for determining whether or not the voice is a voice has been described, the determination unit 132 removes a voice waveform presumed to be an exclamation word or the like from the voice uttered by the user before performing the determination process. Based on the voice, it may be determined whether the second voice is a voice repeatedly spoken following the first voice. For example, it is assumed that the first voice "WV3" includes a voice waveform presumed to be "A". Further, it is assumed that the first voice "WV4" includes a voice waveform presumed to be "er". In this case, the determination unit 132 deletes the voice waveform corresponding to "A" included in the first voice and deletes the voice waveform corresponding to "Eh" included in the second voice. Then, the determination unit 132 determines whether the second voice is a voice repeatedly uttered following the first voice based on the similarity between the first voice "WV3" and the second voice "WV4". May be good.

[5-5. Calculation process]
In the above embodiment, an example of the calculation process in which the calculation unit 134 of the determination device 100 calculates the score for each combination of the estimation results has been described, but the calculation unit 134 is not limited to the degree of duplication, but is based on the content of the word. The score may be calculated. For example, the calculation unit 134 calculates the score SC2 by the following formula (12).

Score SC2 = Acc1 + Acc2 + Con ... (12)

In the above equation (12), "Acc1" indicates the estimation accuracy of the estimation result of the first voice, "Acc2" indicates the estimation accuracy of the estimation result of the second voice, and "Con" indicates the estimation accuracy of the first voice. Information on the meaning of the word included in the estimation result and the estimation result of the second voice (hereinafter referred to as the degree of meaning duplication) is shown. For example, suppose that the score SC2 in "photograph application x image application" and "performance application x cleaning application" is calculated. In addition, in the "photo application x image application", the degree of semantic overlap is "" because the meanings of "photo" and "image" match and the "application" matches by semantic analysis or the like. It shall be calculated as "2.0". In this case, in the "photograph application x image application", the estimation accuracy of the first sound estimation result "photograph application" is "0.8", and the estimation accuracy of the second sound estimation result "image application" is "0". Since it is "0.9" and the degree of semantic overlap is "2.0", the score SC2 is calculated as "3.7". Further, for example, in the "performance application x cleaning application", since the "applications" match by semantic analysis or the like, the semantic overlap degree is calculated to be "1.0". In this case, in the "playing application x cleaning application", the calculation unit 134 has an estimation accuracy of "1.0" for the estimation result "playing application" of the first sound, and the estimation result "cleaning application" of the second sound. Since the estimation accuracy is "1.0" and the semantic overlap degree is "1.0", the score SC2 is calculated as "3.0".

[5-6. Decision processing]
In the above embodiment, an example of the determination process in which the determination unit 135 of the determination device 100 determines the voice recognition result based on the score for each combination of the estimation results estimated from the voice uttered twice by the user has been described. , The determination unit 135 may determine the voice recognition result based on the learning model generated based on the correct estimation result estimated from the voice repeatedly spoken by the user and the other estimation results. For example, the determination unit 135 learns the correct estimation result estimated from the voice repeatedly spoken by the user as a positive example, and learns the other estimation results as a negative example. Then, the determination unit 135 determines the voice recognition result based on the score corresponding to the estimation result when the voice repeatedly uttered by the user is input as the voice to be predicted.

An example of the determination process executed by the determination device 100 according to the modified example will be described with reference to FIG. FIG. 9 is a diagram showing an example of a determination process executed by the determination device 100 according to the modified example. Hereinafter, an example of the determination process executed by the determination device 100 will be described along the flow with reference to FIG.

First, as shown in FIG. 9, the determination device 100 generates the learning model M1 based on the correct estimation result estimated from the voice repeatedly spoken by the user and the other estimation results (step S21). .. For example, the determination device 100 learns the correct estimation result estimated from the voice repeatedly spoken by the user as a positive example, and learns the other estimation results as a negative example. Then, the determination device 100 generates a learning model M1 that calculates a score for the voice to be predicted when the voice to be predicted is input. Then, the determination device 100 inputs the voice to be predicted into the learning model M1 (step S22). For example, the determination device 100 inputs the voice repeatedly spoken by the user into the learning model M1 as the voice to be predicted. Subsequently, the determination device 100 calculates a score corresponding to the estimation result based on the learning model M1 (step S23). For example, the determination device 100 inputs the voice repeatedly spoken by the user into the learning model M1 as the voice to be predicted, sets the score of the positive example to 1, and sets the score of the negative example to 0, and estimates the result for the voice to be predicted. The score corresponding to is calculated as a score of 0 to 1. Then, the determination device 100 determines the voice recognition result based on the score (step S24).

The determination device 100 may generate the learning model M1 using any learning algorithm. For example, the determination device 100 generates a learning model M1 using learning algorithms such as a neural network, a support vector machine, clustering, and reinforcement learning. As an example, when the determination device 100 uses a neural network to generate a learning model M1, the learning model M1 includes an input layer containing one or more neurons, an intermediate layer containing one or more neurons, and one or more neurons. It has an output layer including.

As a result, the determination unit 135 of the determination device 100 according to the embodiment determines the speech recognition result corresponding to the repeated speech input as the prediction target with high accuracy by the learning model that learns the tendency of the speech spoken by the user. can do.

[6. Hardware configuration]
Further, the terminal device 10 and the determination device 100 according to the above-described embodiment are realized by, for example, a computer 1000 having a configuration as shown in FIG. Hereinafter, the determination device 100 will be described as an example. FIG. 10 is a hardware configuration diagram showing an example of a computer 1000 that realizes the function of the determination device 100. The computer 1000 includes a CPU 1100, a RAM 1200, a ROM (Read Only Memory) 1300, an HDD (Hard Disk Drive) 1400, a communication interface (I / F) 1500, an input / output interface (I / F) 1600, and a media interface (I / F). ) Has 1700.

The CPU 1100 operates based on a program stored in the ROM 1300 or the HDD 1400, and controls each part. The ROM 1300 stores a boot program executed by the CPU 1100 when the computer 1000 is started, a program depending on the hardware of the computer 1000, and the like.

The HDD 1400 stores a program executed by the CPU 1100, data used by such a program, and the like. The communication interface 1500 receives data from another device via the network N and sends it to the CPU 1100, and the CPU 1100 transmits the data generated by the CPU 1100 via the network N to the other device.

The CPU 1100 controls an output device such as a display or a printer, and an input device such as a keyboard or a mouse via the input / output interface 1600. The CPU 1100 acquires data from the input device via the input / output interface 1600. Further, the CPU 1100 outputs the data generated via the input / output interface 1600 to the output device.

The media interface 1700 reads a program or data stored in the recording medium 1800 and provides the program or data to the CPU 1100 via the RAM 1200. The CPU 1100 loads the program from the recording medium 1800 onto the RAM 1200 via the media interface 1700, and executes the loaded program. The recording medium 1800 is, for example, an optical recording medium such as a DVD (Digital Versatile DiSC1) or PD (Phase change rewritable Disk), a magneto-optical recording medium such as an MO (Magneto-Optical disk), a tape medium, a magnetic recording medium, or a semiconductor memory. And so on.

For example, when the computer 1000 functions as the determination device 100 according to the embodiment, the CPU 1100 of the computer 1000 realizes the function of the control unit 130 by executing the program loaded on the RAM 1200. Further, the data in the storage unit 120 is stored in the HDD 1400. The CPU 1100 of the computer 1000 reads and executes these programs from the recording medium 1800, but as another example, these programs may be acquired from another device via the network N.

[7. others〕
Further, among the processes described in the above-described embodiments and modifications, all or part of the processes described as being automatically performed can be manually performed, or are described as being manually performed. It is also possible to automatically perform all or part of the processed processing by a known method. In addition, the processing procedure, specific name, and information including various data and parameters shown in the above document and drawings can be arbitrarily changed unless otherwise specified. For example, the various information shown in each figure is not limited to the illustrated information.

Further, each component of each of the illustrated devices is a functional concept, and does not necessarily have to be physically configured as shown in the figure. That is, the specific form of distribution / integration of each device is not limited to the one shown in the figure, and all or part of the device is functionally or physically dispersed / physically distributed in any unit according to various loads and usage conditions. Can be integrated and configured.

Further, the above-described embodiments and modifications can be appropriately combined as long as the processing contents do not contradict each other.

Further, the above-mentioned "section, module, unit" can be read as "means" or "circuit". For example, the determination unit can be read as a determination means or a determination circuit.

[8. effect〕
As described above, the determination device 100 according to the embodiment includes an estimation unit 133 and a determination unit 135. The estimation unit 133 estimates the utterance content corresponding to the voice uttered by the user. The determination unit 135 has the estimation accuracy of the first estimation result (corresponding to the estimation result of the first voice) estimated from the first voice by the estimation unit 133, and the second voice repeatedly uttered by the user following the first voice. The voice recognition result corresponding to the first voice and the second voice is determined based on the estimation accuracy in the second estimation result (corresponding to the estimation result of the second voice) estimated from.

As a result, the determination device 100 according to the embodiment can determine the voice recognition result based on the score calculated for each combination of voices repeatedly uttered by the user, so that the accuracy of voice recognition can be improved. can.

Further, in the determination device 100 according to the embodiment, the determination unit 135 determines either the first estimation result or the second estimation result based on the estimation accuracy in the first estimation result and the estimation accuracy in the second estimation result. It is determined as the voice recognition result corresponding to the first voice and the second voice.

As a result, when the estimation result of the first voice and the estimation result of the second voice are different in the combination of the voices repeatedly uttered by the user, the determination device 100 according to the embodiment is based on the estimation accuracy of each estimation result. Since the voice recognition result can be determined, the accuracy of voice recognition can be improved.

Further, in the determination device 100 according to the embodiment, the estimation unit 133 estimates a plurality of utterance contents corresponding to one voice uttered by the user, and the determination unit 135 estimates a plurality of utterances estimated from the first voice. One of a plurality of first estimation results and a plurality of second estimation results based on each estimation accuracy in one estimation result and each estimation accuracy in a plurality of second estimation results estimated from the second voice. This is determined as the voice recognition result corresponding to the first voice and the second voice.

As a result, the determination device 100 according to the embodiment has the estimation result of the first voice and the estimation result of the second voice in the combination determined based on the score calculated for each combination of the voices repeatedly spoken by the user. When is different, the voice recognition result can be determined based on the estimation accuracy of each estimation result, so that the accuracy of voice recognition can be improved.

Further, in the determination device 100 according to the embodiment, the score is calculated based on the estimation accuracy of the first estimation result and the estimation accuracy of the second estimation result for each combination of the first estimation result and the second estimation result. The calculation unit 134 is further provided, and the determination unit 135 includes the first estimation result and the second estimation result included in the combination of the first estimation result and the second estimation result selected based on the score calculated by the calculation unit 134. One of the results is determined as the voice recognition result corresponding to the first voice and the second voice.

As a result, the determination device 100 according to the embodiment can determine the voice recognition result based on the score calculated for each combination of voices repeatedly uttered by the user, so that the accuracy of voice recognition can be improved. can.

Further, in the determination device 100 according to the embodiment, the calculation unit 134 calculates a score based on the degree of overlap between the first estimation result and the second estimation result for each combination of the first estimation result and the second estimation result. calculate.

As a result, the determination device 100 according to the embodiment can determine the voice recognition result based on the degree of duplication of each voice of the same content repeatedly uttered by the user, thus improving the accuracy of voice recognition. Can be done.

Further, in the determination device 100 according to the embodiment, the calculation unit 134 uses a word having a similar meaning included in the first estimation result and the second estimation result for each combination of the first estimation result and the second estimation result. Calculate the score based on the degree of overlap with respect to.

As a result, the determination device 100 according to the embodiment can determine the voice recognition result based on the similarity regarding the meaning of the words used in the voice for the voice of the same content uttered a plurality of times by the user. The accuracy of voice recognition can be improved.

Further, in the determination device 100 according to the embodiment, the reception unit 131 that receives the voice spoken by the user, the first voice received by the reception unit 131, and the second voice received by the reception unit 131 after the first voice. Further, a determination unit 132 for determining whether the second voice is a voice repeatedly spoken following the first voice based on the similarity with the voice is provided, and the determination unit 135 is provided with the second voice by the determination unit 132. When it is determined that is a voice repeatedly spoken following the first voice, the voice recognition result corresponding to the first voice and the second voice is determined.

As a result, the determination device 100 according to the embodiment can accurately determine whether or not a plurality of voices have been uttered by the user a plurality of times, so that the voice recognition can be performed in a state where the burden on the user is minimized. It can be carried out.

Although some of the embodiments of the present application have been described in detail with reference to the drawings, these are examples, and various modifications are made based on the knowledge of those skilled in the art, including the embodiments described in the disclosure column of the invention. It is possible to practice the present invention in other improved forms.

1 Decision system 10 Terminal device 100 Decision device 110 Communication unit 120 Storage unit 121 Voice information storage unit 122 Estimated result information storage unit 123 Score information storage unit 130 Control unit 131 Reception unit 132 Judgment unit 133 Estimate unit 134 Calculation unit 135 Decision unit 136 Providing department

Claims (10)

An estimation unit that estimates the utterance content corresponding to the voice spoken by the user,
The estimation accuracy in the first estimation result estimated from the first voice by the estimation unit, and the estimation accuracy in the second estimation result estimated from the second voice repeatedly spoken by the user following the first voice . A calculation unit that calculates a score for each combination of the first estimation result and the second estimation result based on the degree of overlap indicating the degree of overlap between the first estimation result and the second estimation result.
Based on the score calculated by the calculation unit, the determination unit that determines the voice recognition result corresponding to the first voice and the second voice, and
A determination device characterized by being equipped with. The decision unit
Based on the estimation accuracy in the first estimation result and the estimation accuracy in the second estimation result, either the first estimation result or the second estimation result corresponds to the first voice and the second voice. Determined as a voice recognition result,
The determination device according to claim 1. The estimation unit
Estimate multiple utterance contents corresponding to one voice uttered by the user,
The decision unit
Based on the respective estimation accuracy in the plurality of first estimation results estimated from the first voice and the respective estimation accuracy in the plurality of second estimation results estimated from the second voice, the plurality of first Any one of the estimation result and the plurality of second estimation results is determined as the voice recognition result corresponding to the first voice and the second voice.
The determination device according to claim 1 or 2. Before Symbol determining unit,
Before the selected on the basis of the kiss core first estimated result and said second estimation result the included in the combination of the first estimated result and said second estimated 1 wherein said one of the results voice and the second Determined as the voice recognition result corresponding to the voice,
The determination device according to any one of claims 1 to 3. The calculation unit
For each combination of the first estimation result and the second estimation result, the score is calculated based on the degree of overlap between the first estimation result and the second estimation result.
The determination device according to claim 4, wherein the determination device is characterized by the above. The calculation unit
For each combination of the first estimation result and the second estimation result, the score is calculated based on the degree of duplication of words having similar meanings contained in the first estimation result and the second estimation result. ,
The determination device according to claim 4, wherein the determination device is characterized by the above. A reception unit that receives voice uttered by the user,
Based on the similarity between the first voice received by the reception unit and the second voice received by the reception unit after the first voice, the second voice follows the first voice. Further equipped with a determination unit for determining whether the voice is repeatedly spoken,
The decision unit
When the determination unit determines that the second voice is a voice repeatedly uttered following the first voice, the voice recognition result corresponding to the first voice and the second voice is determined.
The determination device according to any one of claims 1 to 6, characterized in that. It ’s a decision method that a computer makes.
An estimation process that estimates the utterance content corresponding to the voice spoken by the user,
The estimation accuracy in the first estimation result estimated from the first voice by the estimation step, and the estimation accuracy in the second estimation result estimated from the second voice repeatedly spoken by the user following the first voice . A calculation step of calculating a score for each combination of the first estimation result and the second estimation result based on the degree of overlap indicating the degree of overlap between the first estimation result and the second estimation result.
Based on the score calculated by the calculation step, a determination step of determining the voice recognition result corresponding to the first voice and the second voice, and
A determination method characterized by having. An estimation procedure for estimating the utterance content corresponding to the voice spoken by the user, and
The estimation accuracy in the first estimation result estimated from the first voice by the estimation procedure, and the estimation accuracy in the second estimation result estimated from the second voice repeatedly spoken by the user following the first voice . A calculation procedure for calculating a score for each combination of the first estimation result and the second estimation result based on the degree of overlap indicating the degree of overlap between the first estimation result and the second estimation result.
A determination procedure for determining the voice recognition results corresponding to the first voice and the second voice based on the score calculated by the calculation procedure, and
A decision program characterized by having a computer execute. The first estimation result estimated from the first voice spoken by a predetermined user, the second estimation result estimated from the second voice repeatedly spoken by the user following the first voice, and the first estimation result . An input layer in which an input layer indicating the degree of overlap between the estimation result and the second estimation result is input, and
Output layer and
A first element that is any layer from the input layer to the output layer and belongs to a layer other than the output layer.
A second element, a a computer with a including model value is calculated based on the weight of said first element and said first element,
With respect to the first estimation result, the second estimation result, and the multiplicity input to the input layer, each element belonging to each layer other than the output layer is set as the first element, and the first element and the first element are used. By performing the calculation based on the weight of the element, the voice recognition result corresponding to the first voice and the second voice is output from the output layer.
A program that makes a computer work.

Images