JP5181533B2 - Spoken dialogue device - Google Patents

Description

  The present invention relates to a voice interaction device, and more particularly, to improve the accuracy of voice recognition using a past dialogue history with a user related to a predetermined subject, and to make the dialogue related to the predetermined subject smoother. The present invention relates to a voice interactive apparatus.

  Conventionally, a geographical area that automatically waits for a user's voice input regardless of the button operation such as the voice input start button is set, and when a vehicle enters the geographical area, there is a possibility of being uttered 2. Description of the Related Art A voice input device is known that starts waiting for voice input after changing a voice recognition dictionary used for voice recognition while predicting a high vocabulary based on features of the geographical region (for example, Patent Documents) 1).

  This voice input device sets a geographical area based on information on a genre of a facility, information on user preferences, or setting history information on a destination or waypoint, and names of facilities or places belonging to the geographical area The speech recognition dictionary that includes the proper noun is temporarily adopted to improve the accuracy of speech recognition, and the conversation with the user on the subject relating to the geographical area is facilitated.

As a result, this voice input device can reduce the burden on the user who wants to obtain information on the geographical area through dialogue.
JP 2006-251298 A

  However, since the voice input device described in Patent Document 1 only waits for the user to speak after adopting the voice recognition dictionary corresponding to the geographical area, the degree of accuracy improvement of voice recognition is fixed and limited. There is no way to actively enrich the dialogue.

  In view of the above points, an object of the present invention is to provide a speech dialogue apparatus that actively enhances the content of dialogue while continuously improving the accuracy of voice recognition.

  In order to achieve the above-mentioned object, a voice interaction device according to a first invention is a voice interaction device that controls a dialogue with a user while recognizing a user's utterance using a plurality of voice recognition dictionaries as a reference object. An utterance history recording unit that records a user's utterance history for each subject, and a dialogue scenario determination unit that determines a new scenario for a conversation in each subject based on the utterance history recorded by the utterance history recording unit for each subject. And a reference dictionary determining means for determining a speech recognition dictionary to be referred to based on a new scenario of the dialog determined by the dialog scenario determining means.

  The second invention is a voice interaction device according to the first invention, further comprising an operation history recording means for recording an operation history of the in-vehicle device, wherein the operation scenario recording means includes the operation history recording means. A new scenario of dialogue in each subject is determined based on the operation history of the in-vehicle device related to each recorded subject and the utterance history recorded for each subject by the utterance history recording means.

  A third invention is a voice interaction device according to the first or second invention, wherein the reference dictionary determining means excludes a voice recognition dictionary having a low reference frequency from a reference target.

  With the above-described means, the present invention can provide a voice dialogue apparatus that actively enhances the dialogue contents while continuously improving the accuracy of voice recognition.

  Hereinafter, the best mode for carrying out the present invention will be described with reference to the drawings.

  FIG. 1 is a block diagram showing a configuration example of a voice dialogue apparatus according to the present invention. A voice dialogue apparatus 100 outputs a question and advice in a voice according to a dialogue scenario determined for each subject (theme). An in-vehicle device that continues a conversation with a user exchanged in a vehicle cabin while recognizing utterance content, and includes a control unit 1, a voice input unit 2, a storage unit 3, a voice output unit 4, and a display unit 5, The content of the dialogue with the user is enriched while changing the dialogue scenario on a predetermined subject according to the past dialogue history or the output history of the in-vehicle device 6 or the navigation device 7.

  Here, the “dialog scenario” means the type of question issued by the voice dialog device 100 (a question requesting either affirmation or denial, or a question for which three or more answers are planned). It is a way of proceeding a dialogue composed of the contents of the question, the order of the questions, or the length of the whole dialogue (meaning whether a detailed question is output in small increments, or it enters the main subject directly) It is assumed that a plurality of dialogue scenarios can be prepared and adopted for each subject.

  The control unit 1 is a computer including a CPU (Central Processing Unit), a RAM (Random Access Memory), a ROM (Read Only Memory), and the like. For example, the voice recognition unit 10, the utterance history recording unit 11, the operation history recording While storing programs corresponding to each of the means 12, the dialog scenario determining means 13, the reference dictionary determining means 14 and the voice dialog control means 15 in the ROM, the CPU is caused to execute processing corresponding to each means.

  The voice input unit 2 is a device for inputting user utterances. For example, a directional microphone that recognizes only utterances from a predetermined direction, or a plurality of directions using a phase difference of a plurality of sounds. There are microphone sets equipped with a plurality of sound receiving sections so that utterances can be recognized.

  The storage unit 3 is a device for storing various types of information. For example, the storage unit 3 is a storage medium such as a hard disk or a DVD (Digital Versatile Disk), and stores a voice recognition dictionary 30, an utterance history database 31, and an operation history database 32. To do.

  The voice recognition dictionary 30 is a group of dictionaries that are referred to by the voice recognition unit 10 described later in order to convert voice data acquired via the voice input unit 2 into text data. This is a function that automatically turns on and off the headlight according to the brightness of the vehicle.) Expressions that are expected to be uttered on each subject such as a method for recovering regenerative energy or a function explanation of follow-up driving control (for example, Is a set of dictionaries in which a predetermined number of vocabularies (for example, 10 vocabularies) are collected.

  Expressions expected to be uttered are set in advance based on questions, advices, etc. issued by the voice interaction apparatus 100 in each subject. In other words, questions, advices, etc. issued by the voice interaction apparatus 100 are predetermined utterances by the user. Configured to induce.

  Note that the vocabulary stored in the speech recognition dictionary 30 may be a word unit, a phrase unit, or a sentence unit.

  The utterance history database 31 is a database for storing a user's utterance history, and stores, for example, utterance time, utterance content, vehicle position information (latitude, longitude, altitude) at the time of utterance and the like.

  The operation history database 32 is a database for storing an operation history of the in-vehicle device 6 and stores, for example, operation time, operation content, vehicle position information (latitude, longitude, altitude) at the time of operation in association with each other.

  The voice output unit 4 is a device for outputting various types of information as voices. For example, the voice output unit 4 is an in-vehicle speaker and is based on a dialogue scenario determined by a dialogue scenario determination unit 13 described below based on a control signal from the control unit 1. Voice output of questions and advice.

  The display unit 5 is a device for displaying various types of information. For example, the display unit 5 is a liquid crystal display or the like, and is text information corresponding to audio information output by the audio output unit 4 or text for supplementing the audio information. Display information and graphic information.

  The in-vehicle device 6 is a device that outputs information for allowing the voice interactive device 100 to grasp the running state of the vehicle. For example, the in-vehicle device 6 includes an auto light control, a fog light switch, a two-wheel drive / four-wheel drive changeover switch, a shift selector, There are devices that output their own setting status, such as a tracking control device, etc., and measurements such as steering angle sensor, accelerator opening sensor, brake pedal force sensor, illuminance sensor, gradient sensor, raindrop sensor, vehicle speed sensor, etc. Including devices that output values. Furthermore, the in-vehicle device 6 may include a communication device that receives traffic jam information and weather information.

  The navigation device 7 is a device for guiding the vehicle to the destination based on the electronic map information and the current position of the vehicle. For example, the navigation device 7 receives the GPS signal by a GPS (Global Positioning System) receiver while receiving the vehicle position ( (Latitude, Longitude, Altitude) is measured, the optimal route to the destination is derived using the Dijkstra method as the shortest route search algorithm, and the derived optimal route is superimposed on the electronic map and voice guidance is output. Guide the vehicle to the destination.

  Next, various units included in the control unit 1 will be described.

  The voice recognition unit 10 is a unit for recognizing a voice uttered by a user as text data. For example, the voice recognition unit 10 recognizes a user utterance by a grammar-based voice recognition engine using an acoustic model, or an acoustic model or a language model. While using it, the user's utterance is recognized by a dictation-based speech recognition engine.

  The utterance history recording unit 11 is a unit for recording the utterance history. For example, the text data corresponding to the user utterance recognized by the voice recognition unit 10 is converted into the utterance time, the subject of the utterance (for example, auto light control The function is recorded in the utterance history database 31 of the storage unit 3 while being associated with the position information (latitude, longitude, altitude) of the vehicle at the time of utterance.

  Further, when a plurality of users use one vehicle, the utterance history recording unit 11 may store the utterance history for each user using a user identification number or the like.

  Further, the utterance history recording means 11 may record the fact that the speech could not be recognized in the utterance history database 31 of the storage unit 3 while associating it with the utterance time, the subject of the utterance, the position information of the vehicle at the time of utterance, and the like. Good. This is to make it possible to use the fact that the speech could not be recognized to estimate the user's understanding of the subject.

  The operation history recording unit 12 is a unit for recording an operation history. For example, information output from the in-vehicle device 6 is changed to an operation time, an operation content (for example, a fact that a switch of a certain device is changed from on to off, etc. The vehicle position information (latitude, longitude, altitude) at the time of operation is stored in association with each other.

  Similarly to the utterance history recording unit 11, the operation history recording unit 12 stores an operation history for each user using a user identification number or the like when a plurality of users use one vehicle. It may be.

  The dialogue scenario determining unit 13 is a unit for determining a new dialogue scenario based on the utterance history recorded by the utterance history recording unit 11 or the operation history recorded by the operation history recording unit 12.

  For example, the dialogue scenario determination means 13 determines whether or not the operation method of the auto light control is correctly transmitted to the user based on the user's utterance history in the conversation on the subject of the auto light control performed in the past, and the correct transmission is performed. If it is determined that it is not, the condition for starting the next dialogue with the subject of auto light control (hereinafter referred to as “dialogue start condition”) or the dialogue scenario is changed.

  The dialogue start condition includes, for example, a predetermined time (for example, dusk at this time) when a distance to a predetermined point (for example, a point where the surroundings become dark like a tunnel in this case) is less than a threshold. Etc.), or when a predetermined vehicle-mounted device 6 (for example, a headlight switch in this case) is operated. The distance between the host vehicle and the tunnel is acquired based on the output of the navigation device 7.

  In addition, the change of the dialogue scenario includes, for example, changes of the contents of questions to be output by voice, the order of questions, terms used for explanation, and the like.

  The dialogue scenario determination means 13 determines that the explanation is not correctly transmitted when there is no user utterance (answer) for the question or when an answer different from the expected answer is obtained.

  In addition, the dialogue scenario determining means 13 refers to the operation history database 32, acquires the operation contents performed by the user for the autolight control after the dialogue on the subject of the autolight control is completed, and during the dialogue By checking whether or not the operation as described above is performed, it may be determined whether or not the operation method of the auto light control is correctly transmitted to the user, and between the utterance history and the operation history Based on the relationship, it may be determined whether or not the operation method of the auto light control has been correctly transmitted to the user.

  For example, when there is no expected answer, if an operation as described is performed, it may be assumed that the operation method has been known for a long time, or that the operation method has been correctly transmitted, On the other hand, if there is an expected response but the operation is not performed as described, it is only an operation that is different from the procedure described intentionally, and that the operation method is correctly communicated. This is because there are cases where it can be estimated.

  The dialogue scenario determination means 13 estimates a driving purpose (for example, commuting, attending school, shopping, etc.) from information on the current time, destination, schedule, etc. output from the navigation device 7, and sets the estimated driving purpose. The dialogue scenario may be changed based on this. For example, the dialogue scenario determination means 13 adopts a dialogue scenario that goes directly to the main subject during commuting or school, and the speech recognition dictionary 30 including a short command word is used as a reference dictionary by the reference dictionary determination means 14 described later. To be adopted.

  As a result, the voice interaction device 100 can continue the conversation without causing discomfort to the user, and can effectively support driving by the information provided through the conversation.

  The reference dictionary determining unit 14 is a unit for determining the speech recognition dictionary 30 to be referred to for each subject. For example, the reference dictionary determining unit 14 corresponds to the dialog start condition or the dialog scenario changed based on the determination by the dialog scenario determining unit 13. It is determined which speech recognition dictionary 30 is to be referred to.

  Further, the reference dictionary determining unit 14 may exclude the speech recognition dictionary 30 having a low reference frequency from the reference target based on the utterance history recorded by the utterance history recording unit 11. This is to prevent erroneous recognition of the user utterance and further improve the accuracy of speech recognition.

  The reference dictionary determining unit 14 uses user information (for example, a certain user uses “yes” as an utterance indicating affirmation and “no” as an utterance indicating a denial based on the utterance history recorded by the utterance history recording unit 11. , Another user refers to information on the user's characteristics such as “Yes” as an utterance indicating affirmation and “No” as an utterance indicating denial), and the voice to be referred to based on the extracted user information The recognition dictionary 30 may be determined.

  The reference dictionary determining unit 14 stores only the dictionary identification number corresponding to the speech recognition dictionary 30 to be referred to by the speech recognition unit 10 among the dictionary identification numbers assigned to the speech recognition dictionaries for each subject. The voice recognition dictionary 10 is referred to by the voice recognition means 10.

  Further, the reference dictionary determining means 14 can refer to each voice recognition dictionary by switching the value of the reference flag of each voice recognition dictionary from “0 (non-reference)” to “1 (reference)”. Thus, the speech recognition means 10 may be referred to by storing the values of the reference flags of all the speech recognition dictionaries 30 for each subject.

  Moreover, although the reference dictionary determination means 14 determines the some speech recognition dictionary 30 as a reference object for every theme, it is good also considering the single speech recognition dictionary 30 as a reference object.

  Further, the reference dictionary determining means 14 determines which speech recognition dictionary 30 is to be referred to for each subject, but which speech recognition dictionary 30 is to be referred to for each part of the subject or for each question. You may make it determine whether to do.

  The voice dialogue control means 15 is a means for controlling a dialogue with the user according to a dialogue scenario. For example, when a dialogue start condition for a predetermined subject is satisfied, a predetermined question is sent from the voice output unit 4. Output audio.

  The voice interaction control means 15 operates the voice input unit 2 for a predetermined time (for example, 5 seconds) after outputting a predetermined question, and the voice recognition means 10 cannot recognize the user's utterance (hereinafter, “ Transition to a state in which the speech recognition means 10 can recognize the user's utterance (hereinafter referred to as a “standby state”).

  This is to reduce the user's load on voice input (for example, there is a load caused by forcing the pressing of the voice input button for voice input), and voice recognition is performed while not accepting meaningless user utterances. This is to further improve the accuracy.

  Further, the voice dialogue control means 15 selects a voice recognition engine (for example, grammar base or dictation base) used by the voice recognition means 10 in accordance with the dialogue scenario decided by the dialogue scenario decision means 13. Also good.

  Next, a process (hereinafter referred to as “speech recognition dictionary determination process”) in which the voice interaction apparatus 100 determines a speech recognition dictionary 30 to be referred to will be described with reference to FIG. FIG. 2 is a flowchart showing the flow of the speech recognition dictionary determination process. For example, when the engine is started, the speech dialogue apparatus 100 prepares for the next conversation with respect to a predetermined subject. Shall be executed.

  First, the control unit 1 of the voice interaction apparatus 100 refers to the utterance history database 31 and acquires the utterance history for the subject “method of using the two-wheel drive / four-wheel drive switch” (step S1). This is to estimate the user's understanding of the two-wheel drive / four-wheel drive switch, and the control unit 1 obtains an answer different from the expected answer, for example, when there is no user utterance (answer) to the question. In this case, it is estimated that the user's understanding of the two-wheel drive / four-wheel drive switch is low.

  At this time, the control unit 1 may extract the user's characteristics regarding the response to the dialogue based on the utterance history. This is because the user's characteristics regarding the response to the dialogue are important in determining the dialogue scenario and the speech recognition dictionary 30.

  Further, the control unit 1 refers to the operation history database 32 and acquires the operation history of the two-wheel drive / four-wheel drive changeover switch (step S2). Similarly to the above, this is for estimating the user's understanding of the two-wheel drive / four-wheel drive changeover switch. For example, when the operation in accordance with the advice provided in the previous dialogue is not performed, the control unit 1 It is estimated that the user's understanding of the drive / four-wheel drive switch is low.

  Furthermore, the control unit 1 may estimate the user's level of understanding of the two-wheel drive / four-wheel drive changeover switch based on the travel history (there is a transition of travel speed, etc.).

  Thereafter, the control unit 1 uses the dialogue scenario determination unit 13 to change the dialogue scenario in accordance with the user's estimated comprehension level with respect to the two-wheel drive / four-wheel drive changeover switch (step S3).

  In the previous dialogue scenario, when “how to use the two-wheel drive / four-wheel drive switching switch” is not sufficiently communicated to the user, and it is estimated that the user's understanding level is low, the control unit 1 explains the usage in more detail. A dialogue scenario, a dialogue scenario in which the order of explanation is changed, a dialogue scenario explaining from another viewpoint (for example, improvement in fuel consumption, etc.), etc. are adopted to enhance the user's understanding.

  On the other hand, when it is estimated that the degree of understanding of the user is high because “how to use the two-wheel drive / four-wheel drive switch” can be sufficiently communicated to the user in the previous dialogue scenario, the control unit 1 simplifies the explanation. The dialogue scenario or the like is adopted, and the dialogue start condition is changed to reduce the frequency of the dialogue, so that unnecessary advice is not repeatedly output by voice.

  Accordingly, the control unit 1 uses the reference dictionary determination unit 14 to increase the speech recognition accuracy of the user utterance in the new dialog scenario adopted by the dialog scenario determination unit 13, and includes the vocabulary that is expected to be uttered. The recognition dictionary 30 is newly adopted as a reference object, and the speech recognition dictionary 30 with a low reference frequency is excluded from the reference object (step S4).

  With the above configuration, the voice interaction device 100 can actively enrich the conversation contents by changing the conversation scenario based on the past conversation history.

  In addition, the voice interaction apparatus 100 can continuously improve the accuracy of voice recognition by changing the voice recognition dictionary 30 according to the dialogue scenario.

In addition, since the voice interaction apparatus 100 continuously verifies the effectiveness of the past conversation scenario while referring to the operation history of each in-vehicle device in addition to the past conversation history, the voice interaction apparatus 100 actively uses a better conversation scenario. The content of dialogue can be enhanced.
In addition, the voice interaction apparatus 100 prevents the recognition error of the user utterance from temporarily increasing or the accuracy of the voice recognition from temporarily decreasing by changing the voice recognition dictionary 30 according to the conversation scenario. can do.

  In addition, since the voice interaction apparatus 100 excludes the voice recognition dictionary 30 with a low reference frequency from the reference target, it is possible to further improve the accuracy of voice recognition while reducing the probability of erroneous recognition.

  Although the preferred embodiments of the present invention have been described in detail above, the present invention is not limited to the above-described embodiments, and various modifications and substitutions can be made to the above-described embodiments without departing from the scope of the present invention. Can be added.

  For example, in the embodiment described above, the voice interaction device 100 changes the dialogue scenario and the voice guidance dictionary by determining whether or not the user has correctly understood the operation method when voice guidance is given for the operation method of the automatic light control. However, when the guidance for improving the fuel efficiency is voice-guided because the fuel efficiency is reduced, it is possible to change the dialogue scenario and the voice guidance dictionary by judging whether or not the operation is performed as advised. Good.

It is a block diagram which shows the structural example of the voice interactive apparatus which concerns on this invention. It is a flowchart which shows the flow of a speech recognition dictionary determination process.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Control part 2 Audio | voice input part 3 Memory | storage part 4 Audio | voice output part 5 Display part 6 In-vehicle apparatus 7 Navigation apparatus 10 Voice recognition means 11 Speech history recording means 12 Operation history recording means 13 Dialog scenario determination means 14 Reference dictionary determination means 15 Voice dialog Control means 30 Speech recognition dictionary 31 Utterance history database 32 Operation history database 100 Spoken dialogue apparatus

Claims (3)

A speech dialogue apparatus for controlling a dialogue with a user while recognizing a user's utterance using a plurality of voice recognition dictionaries as a reference object,
Utterance history recording means for recording text data corresponding to the user's utterance as utterance history for each subject,
Dialog scenario determining means for determining a dialog scenario based on the utterance history recorded for each subject by the utterance history recording means ;
Reference dictionary determining means for determining a speech recognition dictionary to be referred to based on the dialog scenario determined by the dialog scenario determining means;
A spoken dialogue apparatus comprising: For example Bei the operation history recording means to record the operation history of the car mounting equipment,
The dialogue scenario determining means determines an interaction scenario in each subject based on the operation history of the in-vehicle device related to each subject recorded by the operation history recording means and the utterance history recorded for each subject by the utterance history recording means,
The reference dictionary determining means determines a speech recognition dictionary to be referred to based on the dialog scenario determined by the dialog scenario determining means;
The spoken dialogue apparatus according to claim 1. The reference dictionary determining means excludes a speech recognition dictionary having a low reference frequency from a reference target;
The spoken dialogue apparatus according to claim 1 or 2, characterized by the above.

Images