JP2000181500A - Speech recognition apparatus and agent apparatus - Google Patents

Abstract

PROBLEM TO BE SOLVED: To lessen the influence of the noises generated in a vehicle and to obtain higher recognition accuracy by detecting the state of noise elements which are the cause for the noises when the recognition accuracy is below a prescribed value and decreasing the noises by controlling the state of the detected noise elements. SOLUTION: An entire part processing section 1 has a navigation processing section 10, an agent processing section 11, etc. The agent processing section 11 judges whether the recognition accuracy of the speeches inputted from a microphone 265 is low or not by a speech control section 14 and if the accuracy is low, the processing section executes the control, such as adjustment, to a noise causative apparatus which is the cause for the noise. The agent processing section 11 executes the evaluation to the recognition result by a speech control section 14 of the speeches inputted from the microphone 26 and executes the judgment as to whether the recognition accuracy is low or not and the detection of the cause (opening or closing of a window, the operating state of an air conditioner or an audio operating state) of the noise to lower the recognition accuracy. The agent processing section 11 executes the noise removal processing for removing the noise cause.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to a speech recognition device and an agent device, and more particularly, to a speech recognition device and an agent device that can be controlled to have higher accuracy when speech recognition accuracy is low.

[0002]

2. Description of the Related Art Techniques for recognizing sounds input from microphones have been put to practical use and have been commercialized in various fields.
For example, in a voice recognition device used in a vehicle such as an automobile, it is possible to control each section of the mounted device in accordance with the recognized voice content. In the case of such a voice recognition device used in a vehicle, the recognition rate of audio generated in the vehicle, various sounds generated by a navigation device, an engine, or the like is reduced as noise. Therefore, Japanese Patent Application Laid-Open No. 6-67689 proposes a technique for reducing the influence of noise in a vehicle and performing speech recognition with high accuracy. In the speech recognition device described in this publication,
The recognition accuracy is improved by adjusting the gain of the microphone amplifier based on the opening and closing amount of the window of the car.

[0003]

However, in the technology described in the above publication, it is difficult to associate the opening and closing amount of the window with the gain of the microphone amplifier. For this reason, a sufficient effect was not obtained as a measure against noise in the vehicle. Therefore, the present invention reduces the effect of noise generated in the vehicle,
A first object is to provide a speech recognition device capable of obtaining higher recognition accuracy.

[0004] By the way, the present applicant makes an anthropomorphic agent appear in the vehicle according to the history of the vehicle and the driver's state, such as the past state of the vehicle, which is currently unknown.
While communicating with drivers and passengers,
We have applied for an agent device that performs various controls as a result of communication. Also in such an agent device, a voice recognition technology is used as an important element for an agent to communicate with a driver or the like in a vehicle, and similarly, improvement in voice recognition accuracy is desired. Accordingly, a second object of the present invention is to provide an agent device capable of performing communication with higher voice recognition accuracy by reducing the influence of noise generated in a vehicle.

[0005]

According to the first aspect of the present invention, a voice recognition means for recognizing a voice, a recognition accuracy obtaining means for obtaining a recognition accuracy by the voice recognition means, and a recognition by the recognition precision obtaining means. A noise element detecting means for detecting a state of a noise element causing noise when the accuracy is equal to or less than a predetermined value; and a noise for controlling the state of the noise element detected by the noise element detecting means to reduce the noise. The first object is achieved by providing a speech recognition device with element control means. According to a second aspect of the present invention, in the speech recognition apparatus according to the first aspect, there is provided a state storage unit that stores a state of the noise element detected by the noise element detection unit, and the noise element control unit includes: When the voice recognition processing is completed, the state of the noise element is returned to the state stored in the state storage unit. According to a third aspect of the present invention, in the speech recognition apparatus according to the first or second aspect, the noise element detecting means detects an open / closed state of a window as a state of the noise element, and the noise element control means When the noise element detecting means detects that the window is open, the open window is closed. According to a fourth aspect of the present invention, in the speech recognition apparatus according to the first or second aspect, the noise element detection unit detects a volume of audio as a state of the noise element, and the noise element control unit includes: When the noise element detecting means detects a sound volume of audio equal to or higher than a predetermined value, the sound volume is reduced. According to a fifth aspect of the present invention, in the voice recognition device according to the first or second aspect, the noise element detecting means detects a flow rate of an air conditioner as a state of the noise element, and the noise element control means comprises: When the noise element detecting means detects an air volume equal to or more than a predetermined amount, the air volume is reduced. According to the invention described in claim 6, a voice recognition unit that recognizes voice, a recognition accuracy obtaining unit that obtains recognition accuracy by the voice recognition unit, an agent appearance unit that causes a personified agent to appear in a vehicle, A noise element detecting means for detecting a state of a noise element causing noise when the recognition accuracy of the recognition accuracy obtaining means is equal to or less than a predetermined value; and controlling a state of the noise element detected by the noise element detecting means. The second object is achieved by providing an agent device with agent control means for causing an agent appearing by the agent appearing means to perform an action of reducing noise.

[0006]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Preferred embodiments of a speech recognition apparatus and an agent apparatus according to the present invention will be described below.
This will be described in detail with reference to FIGS. 1 to 10 using an agent device as an example. (1) Overview of Embodiment In the agent device of the present embodiment, the agent
Recognizes the voices of passengers, such as the driver of the vehicle and passengers,
When it is determined that the recognition accuracy is low, the causes may be (a) when the window is open, (b) when the air conditioner is operating, and (c) when audio is used. . Therefore, when the voice recognition accuracy is low and there are noise causes (a) to (c), the agent closes the window, adjusts the air volume of the air conditioner, and adjusts the audio volume according to the cause. , Is performed. In performing the above noise removal processing,
The agent learns (stores) the state before the process, that is, the position X of the open window and the opening / closing amount R1, the air volume R2 of the air conditioner, and the volume R3 of the audio, and after the voice recognition process ends, A vehicle device state restoration process for returning each vehicle device such as a window to the original state before control is also performed. The agent in the present embodiment is a personified agent, and an image thereof (a planar image, a three-dimensional image such as holography, etc.) appears in the vehicle by the image display device. The processing of the agent includes determining and learning the situation of the vehicle including the vehicle itself, passengers, oncoming vehicles (including not only learning of the situation but also the response and reaction of the driver), and the vehicle situation at each time point. Based on the learning results up to that point, the agent responds to the driver or the vehicle with various variations (action = action and voice). As a result, the driver can freely call and communicate (communicate) with a plurality of agents in the vehicle, and can make the environment in the vehicle comfortable. Here, in the present embodiment, the personified agent has the same identity as a specific person, creature, cartoon character, etc., and the creature with the identity maintains the identity and continuity. The output of the tendency (response by operation or voice) is performed. In addition, identity and continuity are expressed as personalities having unique personalities, and can be regarded as a kind of pseudo-creature in an electronic device. The agent according to the present embodiment that appears in the vehicle is a pseudo-personalized (virtual personalized) subject that is determined in the same manner as a human. Therefore, even in the same vehicle situation, the content of the communication differs depending on the past learning content. Occasionally, there is a misjudgment in a range unrelated to the vehicle, and an unnecessary (burst) response may be made due to the misjudgment. Then, based on the driver's response, it is determined whether or not there is a determination error, and learning is performed.

(2) Details of Embodiment FIG. 1 is a block diagram showing a configuration of an agent device according to this embodiment. In the present embodiment, an overall processing unit 1 that controls the entire communication function of the agent is provided. The overall processing unit includes a navigation processing unit 10, an agent processing unit 11, and an I / F unit for the navigation processing unit 10 and the agent processing unit 11, which search for a route to a set destination and provide route guidance by voice or image display. 12, an image processing unit 13 for processing an image output or an input image such as an agent image or a map image,
A voice control unit 14 that outputs voices such as voices of agents and voices for route guidance, recognizes input voices using a voice recognition dictionary, and detects detection data of various situations regarding vehicles and drivers. It has a situation information processing unit 15 for processing. The agent processing unit 11 causes the agent having a predetermined appearance to appear in the vehicle, and switches to an agent having a newly called name (changes the image of the appearance) to appear in the vehicle. In addition, appropriate conversation and control are performed by learning the situation of the vehicle and past responses by the driver.

The agent processing unit 11 determines whether or not the recognition accuracy of the voice input from the microphone 26 by the voice control unit 14 is low. If the recognition accuracy is low, the noise source It performs control such as adjustment for the equipment. Further, the agent processing unit 11 evaluates the recognition result of the voice input from the microphone 26 by the voice control unit 14 to determine whether the recognition accuracy is low or to determine the cause of the noise that lowers the recognition accuracy (opening / closing of the window). State, operating state of air conditioner, operating state of audio). Further, the agent processing unit 11 performs a noise removal process for removing a noise cause, as described later.

The navigation processing unit 10 and the agent processing unit 11 include a CPU (central processing unit) for controlling data processing and operation of each unit, a ROM connected to the CPU by a bus line such as a data bus and a control bus, and the like. RAM,
A timer and the like are provided. The two processing units 10 and 11 are connected to a network, and can acquire processing data of each other. The ROM is a read-only memory in which various data and programs for controlling by the CPU are stored in advance, and the RAM is a random access memory used by the CPU as a working memory.

In the navigation processing unit 10 and the agent processing unit 11 of the present embodiment, the CPU reads various programs stored in the ROM and executes various processes. The CPU is a storage medium drive 23
The computer program is read from an external storage medium set in the storage device, and stored (installed) in another storage device such as an agent data storage device 29, a navigation data storage device 30, and a hard disk (not shown).
Then, a necessary program or the like may be read (loaded) from the storage device into the RAM and executed. Also, necessary programs and the like are transmitted from the recording medium driving device 23 to the R
The program may be directly read into the AM and executed.

A current position detecting device 21 and a navigation data storage device 30 are connected to the navigation processing unit 10, an agent data storage device 29 is connected to the agent processing unit 11, and an input device 22 is connected to the I / F unit 12. , A storage medium drive device 23, a communication control device 24, a window opening / closing device 201, an air conditioner air volume control device 202, and an audio volume control device 203, and a display device 27 and an imaging device 28 are connected to the image processing unit 13, Control unit 1
4, a voice output device 25 and a microphone 26 are connected, and the status information processing unit 15 is connected to a status sensor unit 40.

The current position detecting device 21 is for detecting the absolute position (depending on latitude and longitude) of the vehicle, and uses a GPS (Global Positioning System) for measuring the position of the vehicle using artificial satellites.
Positioning System) Receiver 211 and bearing sensor 2
12, a steering angle sensor 213, a distance sensor 214, a beacon receiving device 215 for receiving position information from a beacon arranged on the road, and the like. GPS receiver 2
11 and the beacon receiving device 215 can perform position measurement independently, but the GPS receiving device 211 and the beacon receiving device 2
In a place where reception by the receiver 15 is impossible, the direction sensor 212
The current position is detected by dead reckoning navigation using both the distance sensor 214 and the distance sensor 214. In order to detect a more accurate current position, a D-GPS (Differential GPS) that receives a correction signal for a positioning error transmitted from a predetermined base station and corrects the current position may be used. . The azimuth sensor 212 is, for example, a terrestrial magnetism sensor that detects terrestrial magnetism to determine the azimuth of the vehicle, a gyro such as a gas rate gyro or an optical fiber gyro that detects the rotational angular velocity of the vehicle and integrates the angular velocity to determine the azimuth of the vehicle. A wheel sensor or the like is used in which a wheel sensor is disposed, and the amount of azimuth displacement is calculated by detecting the turning of the vehicle based on the output pulse difference (difference in moving distance). The steering angle sensor 213 detects the steering angle α by using an optical rotation sensor, a rotation resistance volume, or the like attached to a rotating part of the steering. As the distance sensor 214, for example, various methods such as a method of detecting and counting the number of rotations of a wheel, or a method of detecting acceleration and integrating twice are used.

The input device 22 is used to input the reading of the name of the agent, and to input user information (age, gender, hobby, personality, etc.) used in performing agent processing.
Is for inputting. In addition, the information about these users is not limited to the case where the user inputs from the input device 22. Alternatively, the agent may make various inquiries regarding the favorite team name and the like to the user, and acquire the information from the user's answer. Input device 2
2 is one means for the driver to respond to all other inquiries of the agent according to the present embodiment. The input device 22 includes a current location (departure point) and a destination (arrival point) at the start of traveling in the navigation processing, a predetermined traveling environment (transmission condition) of a vehicle to which a request for information such as traffic congestion information is transmitted to an information providing station, It is also used to input the type (model) of the mobile phone used in the vehicle. As the input device 22, various devices such as a touch panel (functioning as a switch), a keyboard, a mouse, a light pen, a joystick, a remote controller using infrared rays, and a voice recognition device can be used. Further, a remote control using infrared rays or the like and a receiving unit for receiving various signals transmitted from the remote control may be provided. On the remote controller, various keys such as a menu designation key (button) and a numeric keypad are arranged in addition to a joystick for moving a cursor displayed on the screen.

The storage medium drive unit 23 is a drive unit used by the navigation processing unit 10 and the agent processing unit 11 to read a computer program for performing various processes from an external storage medium. The computer programs recorded on the storage medium include various programs and data. Here, the storage medium refers to a storage medium on which a computer program is recorded, and specifically, a magnetic storage medium such as a floppy disk, a hard disk, a magnetic tape, a semiconductor storage medium such as a memory chip or an IC card, and a CD-ROM. ROM, MO, PD (Phase Change Rewritable Optical Disk), DVD (Digital Video Disk) and other storage media from which information can be read optically, paper cards and tapes, and programs for reading programs using character recognition devices It includes a storage medium using paper such as a printed matter (and a medium having a function equivalent to paper) and a storage medium on which a computer program is recorded by various methods.

The storage medium drive unit 23 reads a computer program from these various storage media and, when the storage medium is a writable storage medium such as a floppy disk or an IC card, the navigation processing unit 10. And the data of the RAM of the agent processing unit 11 and the storage devices 29 and 30 can be written to the storage medium. For example, learning contents (learning item data, response data) relating to the agent function on the IC card,
The same agent that stores user information and other information and uses this stored IC card even when driving another vehicle is named according to his / her preference and learned according to past response situations. It is possible to communicate with. This makes it possible to cause an agent having a name unique to the driver and learning contents to appear in the vehicle, instead of an agent for each vehicle.

The communication control device 24 is connected to a portable telephone composed of various wireless communication devices. The communication control unit 24 communicates with an information providing station that provides data related to traffic information such as traffic congestion conditions and traffic regulations in addition to telephone calls, and karaoke data used for karaoke communication in the vehicle. Can be communicated with an information providing station that provides the information. Further, it is also possible to transmit and receive learning data and user information related to the agent function via the communication control device 24.

The window opening / closing device 201 closes an open window under the control of the agent processing unit 11,
The closed window is opened to the original opening and closing amount. Under the control of the agent processing unit 11, the air conditioner air volume adjusting device 202 reduces the air volume of the air conditioner to a level that does not affect voice recognition, and increases the reduced air volume to the original air volume. . Under the control of the agent processing unit 11, the audio volume control device 203 reduces the volume of the audio to a level that does not affect the speech recognition, and increases the reduced volume to the original volume. .

The voice output device 25 is composed of a plurality of speakers arranged in the vehicle, and is controlled by the voice control unit 14, for example, guidance voice for performing route guidance by voice, voice or sound of an agent. Is output. The audio output device 25 may be used in whole or in part as a speaker for audio. Note that the voice control unit 14 can control the tone color, accent, and the like of the voice output from the voice output device 25 in response to the driver's input of a tuning instruction. The voice output device 25 is also configured to output a voice synthesized in order to confirm (call back) to the user the introversion of the voice recognized by the voice control unit 14.

The microphone 26 is a voice to be subjected to voice recognition by the voice control unit 14, for example, an input voice of a destination in a navigation process, a conversation of a driver with an agent (a user response to a callback, etc.). Function as a voice input unit for inputting The microphone 26 may be used also as a microphone for performing karaoke such as a communication karaoke, or a dedicated directional microphone may be used to accurately collect a driver's voice. Is also good. The audio output device 25 and the microphone 26 may form a hands-free unit so that a telephone call can be made without using a mobile phone.

The display device 27 displays a road map and various image information for route guidance by the processing of the navigation processing unit 10, and displays various actions (moving images) of the agent by the agent processing unit 11. Has become. Further, images inside and outside the vehicle captured by the image capturing device 28 are also displayed after being processed by the image processing unit 13. The display device 27 is a liquid crystal display device, a CRT
Various display devices are used. This display device 2
Reference numeral 7 may have a function as the input device 22 such as a touch panel.

The image pickup device 28 has a C for picking up an image.
It is composed of a camera equipped with a CD (Charge Coupled Device). In addition to an in-vehicle camera for imaging the driver, an out-of-vehicle camera for imaging the front, rear, right and left sides of the vehicle are arranged. The image captured by each camera of the imaging device 28 is supplied to the image processing unit 13, where processing such as image recognition is performed, and each recognition result is also used by the agent processing unit 11 to determine a program number. ing.

The agent data storage device 29 is a storage device for storing various data and programs necessary for realizing the agent function according to the present embodiment.
The agent data storage device 29 includes, for example, a floppy disk, hard disk, CD-ROM, optical disk, magnetic tape, IC card, optical card, DVD
And various storage media, and its driving device. In this case, for example, the learning item data 292 and the response data 29
3, a plurality of different storage media and drive units, such as an easy-to-carry IC card or a floppy disk, and other data, such as a hard disk. A driving device may be used.

The agent data storage device 29 has an agent program 290, a program selection table 291, learning item data 292, response data 293, image data 294 for displaying the appearance and behavior of the agent as a still image or a moving image, Vehicle equipment status data 29
6. User information 297 for specifying a driver and various data necessary for processing for other agents are stored.

The image data 294 stores the appearance of each agent and various image data for the agent of each appearance to express various expressions and actions. The user
Each of these agents can be selected and freely named (set). The appearance to be stored does not have to be a human (male, female) appearance, for example, a chick, a dog, a cat, a frog, a mouse, etc., and the animal itself or a human figure (illustration). It may be an animal appearance, a robotic appearance,
It may be the appearance of a specific character or the like, and it is possible to give a name corresponding to each appearance. Also, the age of the agent does not need to be constant, and as a learning function of the agent, the appearance of the child is first changed to a child's appearance, and it grows over time and changes in appearance (change to the appearance of an adult, May be changed). The image data 294 stores images of the appearances of these various agents, and can be selected from the input device 22 or the like according to the driver's preference.

The user information 297 includes the user's name, address, date of birth, gender, character, hobby, favorite sport, favorite team, favorite food, religion, user's height, weight, driver's seat (seat). For each user, various information such as a fixed position (front-back position, backrest angle), an angle of a room mirror, a height of a line of sight, digitized data of a face photograph, voice characteristic parameters, and the like are stored. The user information is used when the agent determines the content of the communication with the user, and the latter data group such as the weight of the user is also used to identify the driver.

The state data 296 stores a vehicle equipment control execution flag and vehicle equipment state data. The vehicle device control execution flag is a flag for specifying the vehicle device that has performed the noise removal control in order to increase the voice recognition accuracy. In the present embodiment, there are a window closing execution flag, an air volume adjustment flag, and a volume adjustment flag. , Turn on each flag,
A flag area for turning off is secured. And
The vehicle device status data is data for storing the status of each device of the vehicle, and stores window status data, air conditioner status data, and audio status data. The position X and the opening / closing amount R1 of the open window as the window state data, the air volume R2 of the air conditioner as the air conditioner state data, and the audio volume R3 as the audio state data are:
It is stored in correspondence with each vehicle equipment control execution flag. When there are a plurality of open windows, the window position Xn and the opening / closing amount R1n (n = 1, 2, 3,...) Are stored by the number of open windows. Each vehicle equipment status data is
Stored when the corresponding flag is turned on, and after the voice recognition processing is completed, the vehicle device state data is read in order to return each controlled device to the original state.

In the agent program 290, an agent processing program for realizing the agent function, detailed actions when the agent and the driver communicate with each other are displayed on the display device 27 as images, and conversation corresponding to the actions is performed. Communication programs to be output from the audio output device 25 are stored in program number order. The agent program 290 stores a plurality of types of voice data with respect to the voice of each program number, and the driver can select the voice from the input device 22 or the like together with the selection of the appearance of the agent. It has become. Agent voices include male voices, female voices, child voices, mechanical voices, animal voices, voices of specific voice actors and actors, voices of specific characters, and the like. The driver chooses. The selection of the voice and the appearance can be changed as needed.

The program selection table 291 is a table for selecting a communication program stored in the agent program 290. The selection conditions for selecting a communication program from the program selection table 291 include the state sensor 40
(Time, start-up location, cooling water temperature, shift position, accelerator opening, etc.) determined from various conditions of the vehicle and the driver detected by the above, and learning stored in learning item data 292 and response data 293. There are items determined from the contents (today's IG ON count, elapsed time since last end, total start count, etc.).

Learning item data 292 and response data 29
Reference numeral 3 denotes data for storing a result learned by the agent based on a driver's driving operation or response. Therefore, the learning item data 292 and the response data 293 are stored and updated (learned) for each driver. The response data 293 stores the history of the user's response to the action of the agent for each communication program number.

The learning item data 292 includes the total number of activations for determining the selection conditions of the program selection table 291,
The last end date and time, today's ignition ON frequency, the last five refueling remaining times, etc. are stored, and are used to determine whether to start the program selected according to the selection conditions (whether to take a rest). The number of rests / date and time, default value, and other data are stored.

In the learning item data 292, the air volume H2 of the air conditioner and the audio volume H3 which do not affect the speech recognition are stored as threshold values. If the accuracy of voice recognition is low, the air conditioner air volume control device 202 and the audio volume control device 203 are controlled to reduce the air volume and volume until they become equal to the threshold values H2 and H3. Learned about.
That is, if the speech recognition rate is still low even if the air volume and volume are reduced to the same as the stored thresholds H2 and H3,
The values of the thresholds H2 and H3 stored in the learning item data 292 are changed so as to be lowered by a predetermined amount so that the air volume and volume are further lowered by a predetermined amount. When lowering the threshold value, for example, in the case of an air conditioner, the air volume is lowered by one step, and in the case of audio, the volume is lowered by 10% of the adjustable range. This air volume and volume threshold H
2, H3 is differently stored for each user (driver) stored in the respective user information 297 because the loudness and the sound quality of the user vary from user to user.

The agent processing section 11 selects from the program selection table 291 these learning item data 292, response data 293, and program numbers corresponding to various situations of the vehicle detected by the situation sensor section 40, and assigns them to the program numbers. Corresponding agent program 29
By selecting and executing 0, communication between the agent and the driver or the like is performed. For example, when the temperature of the engine coolant is low, the agent "sleeps ..."
Act. As a sleepy expression, make an image of a facial expression with eyelids lowered, perform a predetermined action (bowing etc.) after yawning or stretching, rub your eyes first, make movement and vocalization slower than usual It can be expressed by doing. These sleepy expressions are not always the same, but the expressions are appropriately changed by learning the number of actions and the like.
For example, the eyes are rubbed once every three times (A action), yawn once every ten times (B action), and the facial expression with the eyelid lowered (C action) otherwise. These changes are
This is realized by combining the additional program of the action B and the action C with the basic program of the action A. As for which action is to be combined, the number of program executions of the basic action A is counted as a learning item, and an additional program is combined according to the number of times. In addition, a program is also provided in which an agent takes an action such as "swiping", "stepping on", and making a surprise voice on condition that a sudden brake is applied. The selection of each action by the agent is changed by learning for the sudden braking. For example, the first time from the first sudden braking, the third time, "Shimo-mochi", and the fourth to the tenth time, "Tatara", step on, 1
From the 0th time onward, take the action of "stepping on one foot just one step forward" and let the agent gradually get used to sudden braking. Then, if there is an interval of one week from the last sudden braking, the vehicle is moved backward by one step.

The navigation data storage device 30 stores various data files used for route guidance and the like, such as a communication area data file, a rendered map data file, a road network data file, and data on a travel route searched for a destination. A search data file is stored. In the road network data file, intersection data, node data, and road data are stored as various data used for the route search. The communication area data file includes a mobile phone connected to the communication control device 24,
The communication area data for the mobile phone used in the car without connection to display the area where the mobile phone can communicate at the vehicle position on the display device 27 and to use the area where the communication can be performed for the route search is stored in the mobile phone. Stored by type.

The situation sensor unit 40 includes a window opening amount detection sensor 401, an air conditioner air volume detection sensor 402, an audio volume detection sensor 403, and other sensors 404. The window opening amount detection sensor 401 is a sensor that detects the position of an open window among the windows arranged in the vehicle and the opening / closing amount thereof. Air conditioner air volume sensor 401
Is a sensor for detecting whether the air conditioner is on, off, and when the air conditioner is on. Audio volume detection sensor 4
Reference numeral 03 denotes a speaker (also used as a speaker when the audio devices such as a CD (compact disk) player, a cassette tape recorder, an MD (mini disk) player, a radio, a television, and a video tape recorder are on, off, and on. In this case, it is a sensor for detecting the volume output from the audio output device 25). The values detected by these sensors are stored in the state data 296.

The other sensors 404 include various sensors for detecting a vehicle condition, a driver condition, a vehicle interior condition, and the like. These various sensors are arranged at predetermined positions according to the respective sensing purposes. Note that the case where these sensors do not exist as independent sensors includes the case where sensing is performed indirectly from other sensor detection signals. For example, a tire pressure drop detection sensor
A decrease in air pressure is indirectly detected by a change in the signal of the vehicle speed sensor. Other sensors 404 include an ignition sensor for detecting ignition ON and OFF,
For example, a vehicle speed sensor that calculates the vehicle speed by detecting the rotational angular velocity or the number of rotations of the speedometer cable is an accelerator sensor that detects the amount of depression of the accelerator pedal, detects the amount of depression of the brake, and detects the depression force and the speed of depression. A brake sensor that detects whether a sudden brake is applied, a side brake detection sensor that detects whether a side brake is applied, a shift position detection sensor that detects a shift lever position, blinking and blinking of a blinker. A turn signal detection sensor for detecting a direction, a wiper detection sensor for detecting a driving state (speed or the like) of a wiper, a light detection sensor for detecting a lighting state of each lamp such as a head lamp, a tail lamp, a fog lamp, a room lamp, a driver, and Passengers (auxiliary seats, rear seats) wear seat belts Seat belt detection sensor to detect whether the agent is wearing or not, and to encourage the agent to wear it to the extent that it is not disliked when not wearing it, driver's door, passenger's seat door, rear driver's seat side door, rear passenger's seat side A door open / close detection sensor for detecting an open / close state of each door according to a vehicle type such as a door, and in the case of a so-called half-door, an agent notifying the fact, from an image of the inside of the vehicle taken by the image pickup device 28. Detects or detects whether or not a passenger is in the front passenger seat or rear seat by using a pressure sensor or a weight scale placed in the auxiliary seat, etc.Passenger detection sensor, indoor temperature detection that detects indoor air temperature A sensor, an outdoor temperature detecting sensor for detecting the temperature outside the vehicle, detecting the remaining amount of fuel such as gasoline and light oil, and detecting values for the past 5 times immediately before refueling are stored in the learning item data 292, and an average value thereof. The fuel detection sensor for notifying the agent that it is time to refuel when it becomes, detects the temperature of the cooling water, and the agent takes a sleepy action immediately after the ignition is turned on (when the detected water temperature is low) A water temperature detection sensor that alerts the agent to a "sloppy" action before overheating when the water temperature is too high, and ABS to prevent tire lock due to sudden braking and ensure maneuverability and vehicle stability ABS detection sensor that detects whether or not it has been activated, detects the weight of the driver, and from the detected weight,
Alternatively, the driver is specified from the weight and the image of the imaging device 28, and a weight sensor for causing an agent learned in relation to the driver to appear, and the distance to other vehicles or obstacles ahead of the vehicle are detected. Front inter-vehicle distance sensor, rear inter-vehicle distance sensor that detects the distance to another vehicle or obstacle behind, for example, placed on the steering wheel surface to detect the driver's body temperature, heart rate, and sweating state from the driver's hand state A body temperature sensor, a heart rate sensor, a perspiration sensor, a sensor for detecting a driver's brain wave, for example, an α-wave or a β-wave for detecting a driver's arousal state, etc. During normal driving, searching for objects outside the vehicle, searching for objects inside the vehicle, eye tracer to judge arousal state etc., infrared sensor to detect user's hand movement and face movement, tire Air pressure drop Various sensors such as a detection sensor, a belt looseness detection sensor, a window open / closed state sensor, a horn sensor, an indoor humidity sensor, an outdoor humidity sensor, an oil temperature detection sensor, and a hydraulic pressure detection sensor are provided.

Next, the operation of the present embodiment configured as described above will be described. FIG. 2 is a flowchart showing the main operation of the processing by the agent of the present embodiment. The agent processing unit 11 sets the ignition to O
When the ignition sensor 401 detects that N has been performed, first, initialization is performed (step 11). As the initial setting, processes such as clearing the RAM, setting a work area for each process in the RAM, loading the program selection table 291 into the RAM, setting a flag to 0, and the like are performed. In the agent processing according to the present embodiment, the start of the processing is set to the ignition ON. However, the processing may be started when, for example, the opening / closing of any door is detected by the door opening / closing detection sensor 411.

Next, the agent processing section 11 specifies a driver mainly based on various data stored in the user information 297 (step 12). That is,
The agent processing unit 11 analyzes a voice when a greeting is first given by the driver to identify the driver,
The driver is identified by analyzing the captured image, the driver is identified from the weight detected by the weight sensor of the situation sensor unit 40, and the driver is identified from the set seat position and the angle of the rearview mirror. Or In addition, about the identified driver, separately from the processing of the agent described below,
A special communication program for inquiring, such as "Are you?" Is activated, and the driver is confirmed.

When the driver is specified, the agent processing unit 11 grasps the current situation (step 1).
3). That is, the agent processing unit 11 detects the detection value supplied from each sensor of the situation sensor unit 40 to the situation information processing unit 15, the processing result of the image captured by the imaging device 28, and the vehicle detected by the current position detection device 21. The data such as the current position is acquired and stored in a predetermined area of the RAM, and the current situation is grasped from the stored data. For example, the position of the open window and the opening / closing amount are grasped from the output of the window opening amount detection sensor 401, the air volume is grasped from the output of the air conditioner air volume detection sensor 402, and the volume is grasped from the audio volume detection sensor 403. When the temperature of the cooling water detected by the water temperature detection sensor is t1, the agent processing unit 1
1 stores the temperature t1 in the RAM and, when t1 is equal to or less than the predetermined threshold value t2, recognizes that the cooling water temperature is low as the current state of the vehicle. As the current situation, there are other requests from the driver who have performed voice recognition based on the input from the microphone 26, for example, "Please call the number XXX." And "Please play CD" are also recognized as the current situation. In this case, the words “CD”, “Kake” and the like included in the recognized voice are stored in the program selection table 291.
Selection condition. Further, the agent processing unit 11
By checking the learning item data 292 and the response data 293 in the agent data storage device 29 as the current situation, the state (learning data) that the agent has learned so far is grasped.

Upon grasping the current situation, the agent processing section 11 performs an agent process according to the grasped situation (step 14). The processing of the agent here includes various kinds of processing such as judgment, action (action + utterance), control, learning, and inspection by the agent, for example, noise removal processing to be described later. This includes cases where nothing works.

Next, the agent processing section 11 determines whether or not to end the processing of the main operation (step 1).
5) If not terminated (step 15; N), return to step 13 and repeat the process. When one of the processes is terminated, that is, the ignition sensor detects that the ignition has been turned off (step 13), and after the termination process such as turning off the interior light (step 14) is completed (step 15; Y), the main process is performed. The operation of is ended.

FIGS. 3 and 4 are flowcharts showing the processing operation of the noise removal processing according to the present embodiment. This noise elimination processing is based on the state of the voice recognition switch, which is grasped as the current situation in the main operation in FIG.
Based on the position of the open window and the opening / closing amount, the air volume of the air conditioner, and the volume of the audio (step 13), the process is executed as an agent process (step 14) according to the grasped state. The agent processing unit 11 determines whether or not to start speech recognition based on the current situation grasped in step 13 (step 21). Regarding whether to start speech recognition,
For example, when the destination setting in the navigation processing is selected, and when a predetermined voice recognition switch is turned on from the input device 22, the agent processing unit 11 determines that the voice recognition has started.

If speech recognition is started (step 21;
Y) Further, it is determined whether or not a voice has been input from the microphone 26 (step 22), and if no voice has been input (; N).
Return to the main routine. On the other hand, when a voice is input (Step 22; Y), the input voice is recognized by the voice control unit 14 (Step 23), and a callback voice is output by the voice control unit 14 to confirm the recognition result. The voice is synthesized and output from the voice output device 25 (step 24).

When the user responds to this callback with "O
"K" (approval) is confirmed (step 25;
Recognition accuracy acquisition means). That is, the agent processing unit 1
Reference numeral 1 denotes whether an approval key such as “Yes”, “Yes”, or “OK” indicating the approval has been selected (including touch by the touch panel or selection by the joystick) from the input result by the input device 22, or “No” indicating the denial. "N
Confirm whether a denial key such as "o" has been selected. In addition, the agent processing unit 11 outputs a voice “Yes”, “Yes”, “OK”, “OK” indicating the user's approval by voice.
It is input from the microphone 26 as to whether an approval sound such as “matched” is issued or a sound indicating denial such as “No”, “No”, “No” or “No” is issued. The voice is recognized by the voice control unit 14, and the callback is confirmed from the recognition result. In this embodiment, in order to confirm the callback by voice, a voice dictionary dedicated to the callback is prepared, and step 2 is performed.
A predetermined time (for example, within 3 seconds, 5
Any time, such as within seconds, can be set as the callback answer time) for the voice input from the microphone 26,
By using a voice dictionary dedicated to callback, it is possible to accurately recognize whether to approve or reject.

The user's answer to the callback is OK
If not, that is, if the recognition result is rejected (Step 25; N), the agent processing unit 11 determines that the state in the vehicle is a state where the voice recognition accuracy is low, and removes the noise that is the cause. Process for The agent processing unit 11 first determines a window closing execution flag from the flag state of the vehicle device control execution flag in the state data 296,
It is checked whether all of the air volume adjustment flag and the volume adjustment flag are on (step 26). If all the flags are not ON (Step 26; N), the agent control unit 11 checks whether there is any open window from the current situation grasped in Step 13 of the main routine (Step 27). If there is an open window, the position X and the opening amount (opening / closing amount) R1 of the opening window are stored in the state data 296 of the agent data storage device 29 (step 2).
8).

Then, the agent processing unit 11 performs voice control for specifying a vacant window and requesting permission to close the window, such as "Can the window at the X position be closed?" The sound is synthesized by the unit 14 and output from the audio output device 25 (step 29). This audio output is
Actually, the process is performed as communication with the user by the agent, and as shown in FIG. 5, the user is asked whether or not the agent displayed on the display device 27 is allowed to close the window. Note that the balloon shown in FIG. 5 and the characters displayed therein are not displayed because the agent speaks in this embodiment, but are displayed so that the user can visually confirm it. (The same applies hereinafter).

Thereafter, the agent processing section 11 confirms whether or not the user has "OK" for the voice output by the agent, that is, whether or not the user has approved closing the window (step 30). Confirmation of this approval and rejection is performed in the same manner as in step 25. When the user enters "O
If "K" (step 30; Y), the agent processing unit 11 controls the window opening / closing device 201 via the interface unit 12, closes the window at the open position X, and executes the vehicle device control of the state data 296. The window closing execution flag is turned on among the flags (step 31).

After the execution of the window closing (step 31), if it is denied that the window is closed (step 30; N), or if there is no open window (step 27; N), the agent processing unit 11 Next, the air conditioner air volume R2 is acquired from the current situation grasped in step 13 of the main routine, and the threshold value H2 corresponding to the driver (user) specified in step 12 of the main routine is obtained from the learning item data 292. The two are read and compared (step 32). The agent processing unit 11 has an air conditioner air volume R2
Is greater than the threshold value H2 (step 32;
Y), the air volume of the air conditioner is stored in the state data 296 (step 33).

Then, the agent processing unit 11 synthesizes a voice requesting permission to reduce the air volume of the air conditioner by the voice control unit 14, for example, "Can the air volume of the air conditioner be reduced?" Output from the device 25 (step 34). This voice output is actually processed as communication with the user by the agent, as in the case of requesting permission to close the window (step 29), and is displayed on the display device 27 as shown in FIG. The user is asked whether the air volume can be reduced by the agent.

Thereafter, the agent processing section 11 confirms whether or not the user has "OK" with respect to the voice output by the agent, that is, whether or not the user has approved that the air volume should be reduced, in the same manner as in step 25 (step 25). Step 3
5). When the user has "OK" (step 35;
Y), the agent processing unit 11 controls the air conditioner air volume adjusting device 202 via the interface unit 12 to adjust the air volume of the air conditioner to the air volume of the threshold value H2, and to execute the vehicle device control of the state data 296. Turn on the airflow adjustment execution flag among the flags (step 3
6). Note that the threshold value H2 for the air volume is the minimum value “0”.
In this case, the agent processing unit 11 turns off the power of the air conditioner by the air conditioner air volume adjusting unit 202.

After the execution of the air volume adjustment (step 36), if it is determined that the air volume is to be reduced (step 35;
N) or when the air conditioner air volume is equal to or less than the threshold value H2 (including when the air conditioner is stopped) (step 3).
2; N), the agent processing unit 11 acquires the air volume R3 of the audio from the current situation grasped in step 13 of the main routine, and responds to the driver (user) specified in step 12 of the main routine. The threshold value H3 to be read is read from the learning item data 292, and the two are compared (step 37). Agent processing unit 11
Stores the audio volume in the status data 296 when the audio volume R3 is larger than the threshold value H3 (step 37; Y) (step 38).

Then, the agent processing unit 11 synthesizes a voice requesting permission to reduce the audio volume, for example, "Can the audio volume be reduced?" Output device 2
5 is output (step 39). This voice output is actually processed as communication with the user by the agent, as in the case of requesting permission to close the window (step 29), and as shown in FIG.
An inquiry is made to the user as to whether or not the agent whose image is displayed in FIG.

Thereafter, the agent processing section 11 confirms whether or not the user has "OK" with respect to the voice output by the agent, that is, whether or not the user has approved the reduction of the volume in the same manner as in step 25 (step 25). Step 4
0). When the user has “OK” (Step 40;
Y), the agent processing unit 11 controls the audio volume control device 203 via the interface unit 12,
The audio volume is adjusted so as to be equal to the threshold value H3, and the volume adjustment execution flag among the vehicle device control execution flags in the state data 296 is turned on (step 41). When the threshold value of the volume is the minimum value “0”, the agent processing unit 11 turns off the power of the audio by the audio volume adjusting unit 203.

After executing the volume control (step 41), if it is denied that the volume should be lowered (step 40;
N) or when the audio volume is equal to or lower than the threshold value H3 (including the case where no audio is used) (Step 37; N), the agent processing unit 11 proceeds to Step 21.

After the cause of the noise has been eliminated by the vehicle equipment control processing from step 27 to step 41, voice input is performed (step 22; Y), and voice recognition is performed (step 23). When the callback of the recognition result is approved, that is, when the recognition result of the voice is correct (Step 25; Y), the agent processing unit 11 executes an instruction corresponding to the recognition result (Step 42). ), And return to the main routine. As the execution of the instruction corresponding to the recognition result, for example, in the destination setting of the navigation processing, when the user wants to search for the destination from a genre such as a station or an amusement park, the user utters “genre” and is correctly recognized. (Step 25;
Y). As the command execution in this case, an image in which the agent appears on the display device 27 is displayed with a panel listing various selectable genre names. When the image of the information list to be displayed is large, the display of the list is prioritized, and the agent is hidden behind the list.

On the other hand, even if the cause of noise is removed by the vehicle equipment control processing from step 27 to step 41, the recognition accuracy for the subsequently input voice is still low,
That is, voice recognition is performed (step 23), and when the callback (step 24) is rejected (step 25; N), if all the vehicle device control execution flags are on (step 26; Y) 8, the agent processing unit 11 synthesizes, in the voice control unit 14, a voice that prompts the user to perform an effective action to increase the recognition rate, such as "Please be quiet", as shown in FIG. Output from the audio output device 25 (step 43). This voice output is actually processed as communication with the user by the agent, as in the case of requesting permission to close the window (step 29), and is displayed on the display device 27 as shown in FIG. The agent makes a comment urging the user to be quiet. As an effective user action to increase the recognition rate, when there is a passenger other than the driver and the conversation is a cause of noise, a “quiet” action is applicable, as shown in FIG. "Please be quiet," the agent comments. If there is only one driver and there is no passenger, the action of the user is “speak loudly”, and the agent comments “speak a little more loudly”. In addition, about existence of passenger,
The determination is made based on the current situation grasped in step 13 from the passenger detection sensor, the wearing state of the seat belt other than the driver's seat, and the like.

Next, the agent processing unit 11 changes the threshold values H2 and H3 of the air volume and the volume for the driver specified in step 12 of the main routine so as to decrease by a predetermined amount (step 44). That is, the threshold value H2 is reduced so that the air volume of the air conditioner is reduced by one step, and the threshold value H3 is reduced so that the audio volume is reduced by 10% of the adjustable range. In addition, if the air conditioner is not an air conditioner that can adjust the air flow in several steps, but an air conditioner that can adjust the air flow in several tens of steps, or an air conditioner that can be adjusted to an arbitrary position in an analog manner, Similarly, the threshold value H2 is reduced so that the air volume is reduced by 10% of the adjustable range.

Next, the agent processing section 11 turns off the air volume adjustment execution flag and the volume adjustment execution flag (step 45), and proceeds to step 21. By turning off both flags, the air conditioner air volume and audio volume are further reduced to the changed threshold values H2 and H3 by the vehicle device control processing from step 27 to step 41.

As described above, the vehicle device control process provides
Speech recognition is performed with high accuracy, and thereafter, it is confirmed whether the speech recognition is completed (step 21). The speech recognition is terminated, for example, when the speech recognition switch is turned off, when a sound of a certain level or higher is not input to the microphone 26 for a certain period of time, or when the user selects another mode that does not require the speech recognition. It is determined that the speech recognition has ended.

When the voice recognition is completed (step 2
1; N), the agent processing unit 11 executes a vehicle device state restoring process of returning the window closed by the vehicle device control process, the lowered air volume, and the volume to the original state. That is, the agent processing unit 11 confirms from the status data 296 of the agent data storage device 29 whether or not the window closing execution flag is turned on (step 46). The window position X and the opening amount R1 stored in the data 296 are read, and the window opening / closing device 201 is controlled to open the window X at the corresponding position by R1 to return to the original state and turn off the window closing execution flag ( Step 23). Similarly, the agent processing unit 11
Checks whether the air volume adjustment execution flag is on, and if it is on (Step 48; Y), the air conditioner air volume adjustment device 202 is controlled to change the air volume of the air conditioner to the original air volume, R
2 (step 49). Similarly, the agent processing unit 11 checks whether the volume adjustment execution flag is turned on, and if it is on (step 50; Y), the agent processing unit 11 controls the audio volume adjustment device 203 to reduce the audio volume to the original volume R3. (Step 51), and returns to the main routine.

As described above, according to the voice recognition device of this embodiment and the agent device to which the voice recognition device is applied, when the voice recognition rate in the vehicle is low, the open window is closed and the air conditioner is closed. Since the cause of the noise is eliminated by lowering the air volume and lowering the volume of the audio, it is possible to prevent the voice recognition from being performed in a low noise state and prevent the recognition rate from being lowered.

Although the preferred embodiment of the present invention has been described above, the present invention is not limited to the configuration of this embodiment, and other embodiments may be adopted within the scope of the invention described in each claim. , And can be deformed. For example, in the embodiment described above, all the windows, air conditioners, and audio are controlled to remove noise, but in the present invention, only windows, only air conditioners, only audio, windows and air conditioners, windows and audios, One of the air conditioner and the audio, that is, at least one of the window, the air conditioner, and the audio may be controlled as a noise element.

In the embodiment described above, it is determined whether or not control is to be performed for all of window closing, air conditioning air volume adjustment, and audio volume adjustment. However, priority may be assigned to the control. FIG. 9 is a flowchart corresponding to FIG. 4 in the case where priorities are assigned to the control of each device in the vehicle device control process in the noise removal process. Note that the processes with the same step numbers as those in FIG. 4 are the same or similar, and thus the description will be appropriately simplified. That is, when it is determined that the accuracy of the speech recognition is low, the agent processing unit 11 pays attention to the window as a cause of the noise, confirms the open window (Step 27; Y), and determines the window position X and the open position. The amount R2 is acquired and stored (step 28), and if the confirmation of closing the window is approved by the user by the agent (step 29, step 30; Y), the open window is closed (step 31), and step 21 ( The process proceeds to FIG. 3) and speech recognition is continued. The process proceeds to step 21 without determining whether the air conditioner air volume and the audio volume are higher than the threshold values H2 and H3 and without performing adjustment control. If the low voice recognition accuracy is eliminated by closing the window, Continue speech recognition.

On the other hand, if the accuracy of voice recognition is still low even if the window is closed (step 27; N because the window was closed in the previous processing), if the user is refused to close the window (step 30; N), and when there is no originally open window (Step 27; N), the air flow rate R of the air conditioner
It is determined whether or not 2 is higher than a threshold value H2 (step 32). When it is high (step 32; Y), the air volume R2 is stored (step 33), and when the air volume adjustment is approved by the agent (step 34, step 35;
Y), reduce the air volume of the air conditioner to H2 (step 3).
6) Go to step 21. Set the air volume of the air conditioner to H2
If the voice recognition accuracy is increased by lowering the threshold value to a lower level, the voice recognition is continued without determining and adjusting the audio volume.

If the voice recognition accuracy is low even if the air conditioner air volume is reduced to the threshold value H2 (Step 32; N since the air volume was adjusted in the previous processing), if the air volume adjustment is denied (Step 35; N), and when the air volume R2 is equal to or less than the threshold value H2 (Step 32; N), the agent processing unit 11 determines whether or not the audio volume R3 is higher than the threshold value H3 (Step 37). ). If the volume is high (Step 37; Y), the volume R3 is stored (Step 38), and if the volume adjustment is approved by the agent (Step 39, Step 40; Y), the audio volume is reduced to H3 (Step 41). ), And proceed to step 21. If the voice recognition accuracy is increased by lowering the volume of the audio to H3, the voice recognition is continued as it is.

Even if the audio volume is lowered to the threshold value H2, if the voice recognition accuracy is still low, it is effective to increase the recognition rate of "Please speak loudly" or "Please be quiet". The voice that prompts the user to perform an appropriate action is synthesized by the voice control unit 14 and output from the voice output device 25 as a conversation by the agent. After that, the agent processing unit 11 lowers the thresholds H2 and H3 of the air volume and the volume and turns off both flags, as described in steps 44 and 45. In this way, by prioritizing and sequentially controlling the devices that cause noise, it is possible to improve the voice recognition accuracy by controlling only the minimum necessary devices.

In the above-described modified example, control is performed by assigning priorities to the order of closing the window, adjusting the air-conditioning air volume, and adjusting the audio volume. However, the priorities are changed between when the vehicle is running and when the vehicle is stopped. It may be. For example, the priority order while the vehicle is running is controlled in the order of closing the window, air conditioner air volume adjustment, and audio volume adjustment. The priority order while the vehicle is stopped is controlled in the order of air conditioner air volume adjustment, audio volume adjustment, and window closing.

In the described embodiment, as described in the order from step 43 to step 45, after prompting the user to take an effective action to increase the recognition rate (step 43), the threshold value H2 of the air volume and the volume is set. , H3 (step 44) and both flags are turned off (step 45). On the other hand, in the present invention, in the case of Yes in step 26, the threshold values of the air volume and the volume are changed and both flags are turned off, and the process proceeds to step 21, where the threshold value H2,
If H3 becomes the lowest value and the voice recognition accuracy is still low even after the power of the air conditioner and the audio is turned off, an effective user action to increase the recognition rate is encouraged (step 43). Good.

In addition to the three elements of the window, the air conditioner, and the audio, a device that may cause noise may be controlled as a noise element. For example, when there is a karaoke device arranged in the vehicle or a communication karaoke device which acquires performance data and images via the communication control device 24 and performs in the vehicle, the volume of the performance and the input from the microphone are provided. The volume of the singing voice may be adjusted. However, when the audio output device 25 is also used as a speaker for outputting a performance or a singing voice, karaoke or communication karaoke may be included in the audio and controlled. Also, when a wireless telephone such as a car phone, a mobile phone, or a PHS (Personal Handy Phone System) is used in the vehicle, or when a wireless communication device such as an amateur radio or a transceiver is used. In particular, even when a mobile phone or a wireless device or the like is used by a passenger other than the person whose voice recognition is to be performed, such as a driver, the mobile phone is connected so as to be controllable by the agent processing unit 11 via the interface unit 12. If so, noise reduction processing such as lowering the volume may be performed. If these devices are not controllably connected by the agent processing unit 11, the agent requests that these voices be used at a low level, and also makes the voice recognition target utter a loud voice. Request.

In the embodiment described above, when the user does not approve the callback in step 25 of FIG. 3, it is determined that the speech recognition accuracy is low (the recognition accuracy is equal to or less than a predetermined value). However, it is also possible to adopt another criterion for determining whether the recognition accuracy is equal to or less than a predetermined value. For example, if the user rejects the callback of the recognition result (step 25; N) for m times, it is determined that the voice recognition accuracy is equal to or less than a predetermined value, and the vehicle device control processing of steps 27 to 41 is performed. It may be. Also, the number of approvals for the callback of the last 10 times, for example, the last 10 times of the speech recognition result is q times the predetermined number of times,
For example, in the case of seven times or less, it may be determined that the voice recognition accuracy is low, and the vehicle device control processing may be performed. Further, when the S / N ratio (signal-to-noise ratio) of the microphone 26 for voice input is equal to or less than the threshold value H4, it may be determined that the accuracy of voice recognition is low, and vehicle device control processing may be performed. .

In the described embodiment, in the state restoring process from step 46 to step 51, if each flag is on, the state is unconditionally returned to the original state. On the other hand, in the present invention, when each flag is ON, before returning the corresponding vehicle device to the original state, as shown in FIG. 10, whether or not the agent may return to the original state May be confirmed, and if approved, the state may be returned to the original state. As a confirmation word, for example, the agent asks, "Do you want to open the window at the X position?""Would you like to open the window at the X position to the original position?" Do you want to restore the volume? "

In the embodiment described above, voice recognition and vehicle equipment control processing and the like are performed as an agent's action. However, in the present invention, voice recognition and processing are performed not as an agent's action but as processing by a voice recognition device.
Confirmation of recognition accuracy, vehicle equipment control processing, and vehicle equipment state restoration processing may be performed.

[0072]

According to each of the speech recognition devices according to the first to fifth aspects, the effect of noise generated in the vehicle can be reduced, and higher recognition accuracy can be obtained. According to the agent device of the sixth aspect, it is possible to reduce the influence of noise generated in the vehicle and perform communication with higher voice recognition accuracy.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of an agent device according to an embodiment of the present invention.

FIG. 2 is a flowchart showing a main operation of processing by an agent according to the first embodiment;

FIG. 3 is a flowchart showing a part of the processing operation of noise removal processing by the agent.

FIG. 4 is a flowchart showing the rest of the processing operation of the noise removal processing by the agent.

FIG. 5 is an explanatory diagram showing a confirmation state of closing a window by an agent in the noise removal processing.

FIG. 6 is an explanatory diagram showing a confirmation state of air conditioner air volume adjustment by an agent in the noise removal processing.

FIG. 7 is an explanatory diagram showing a confirmation state of audio volume adjustment by an agent in the noise removal processing.

FIG. 8 is an explanatory diagram showing a state in which the agent prompts an effective user action to increase the recognition rate in the noise removal processing.

FIG. 9 is a flowchart illustrating a modified example in which control of each device is prioritized in the vehicle device control process in the noise removal process.

FIG. 10 is an explanatory diagram showing a confirmation state of window opening by an agent in the noise removal processing.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Whole processing part 10 Navigation processing part 11 Agent processing part 12 I / F part 13 Image processing part 14 Voice control part 15 Situation information processing part 21 Current position detection device 22 Input device 23 Storage medium drive device 24 Communication control device 201 Window opening and closing Device 202 Air conditioner air volume control device 203 Audio volume control device 25 Audio output device 26 Microphone 27 Display device 28 Imaging device 29 Agent data storage device 290 Agent program 291 Program selection table 292 Learning item data 293 Response data 294 Image data 296 State data 297 User Information 30 Navigation data storage device 40 Situation sensor unit 401 Window opening amount detection sensor 402 Air conditioner air volume detection sensor 403 Audio volume detection sensor 404 Other status sensors Support

 ──────────────────────────────────────────────────続 き Continued on the front page (72) Manabu Matsuda 2-19-12 Sotokanda, Chiyoda-ku, Tokyo Inside Equos Research Co., Ltd. (72) Kazuhide Adachi 2--19 Sotokanda, Chiyoda-ku, Tokyo 12 Equos Research Co., Ltd. (72) Inventor Koji Mukai 2--19-12 Sotokanda, Chiyoda-ku, Tokyo F-term within Equos Research Co., Ltd. 5D015 EE05 KK01 9A001 DD11 HH17 HH26 HZ19 JZ77 KK32 KK56

Claims (6)

[Claims] 1. A voice recognition means for recognizing a voice, a recognition accuracy obtaining means for obtaining recognition accuracy by the voice recognition means, and a source of noise when the recognition accuracy by the recognition accuracy obtaining means is equal to or less than a predetermined value. Noise element detecting means for detecting the state of the noise element, and noise element control means for reducing the noise by controlling the state of the noise element detected by the noise element detecting means. Voice recognition device. 2. A state storage means for storing a state of the noise element detected by the noise element detection means, wherein the noise element control means is stored in the state storage means when the speech recognition processing is completed. 2. The speech recognition apparatus according to claim 1, wherein the state of the noise element is returned to a state where the noise element has been set. 3. The noise element detecting means detects an open / close state of a window as a state of the noise element. The noise element control means opens the window when the noise element detecting means detects that the window is open. The speech recognition device according to claim 1, wherein a closed window is closed. 4. The noise element detection means detects an audio volume as a state of the noise element, and the noise element control means detects when the noise element detection means detects an audio volume of a predetermined value or more. The voice recognition device according to claim 1, wherein the volume is reduced. 5. The noise element detecting means detects an air volume of the air conditioner as a state of the noise element, and the noise element control means detects the air volume when the noise element detecting means detects an air volume of a predetermined amount or more. The voice recognition device according to claim 1, wherein the voice recognition device lowers the voice recognition value. 6. A speech recognition means for recognizing speech, a recognition accuracy acquisition means for acquiring recognition accuracy by the speech recognition means, an agent appearance means for causing an anthropomorphized agent to appear in a vehicle, and the recognition accuracy acquisition A noise element detecting means for detecting a state of a noise element causing noise when the recognition accuracy by the means is equal to or less than a predetermined value; and controlling the state of the noise element detected by the noise element detecting means to reduce noise. Agent control means for causing an agent appearing by the agent appearing means to perform an act of decreasing the agent.