JP2017090612A - Voice recognition control system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve voice recognition accuracy in relation to a voice recognition control system.SOLUTION: A voice recognition control system includes: an identification unit 2 for identifying a person uttering a voice on the basis of at least the voice: a selection unit 3 for selecting an acoustic model corresponding to the person identified by the identification unit 2; and a recognition unit 4 for recognizing the voice by using the acoustic model selected by the selection unit 3.SELECTED DRAWING: Figure 2

Description

  The present invention relates to a voice recognition control system for controlling an in-vehicle device with a voice of a vehicle occupant.

  2. Description of the Related Art Conventionally, a voice recognition control system that can control an in-vehicle device using a voice command has been developed. That is, the operation state of various in-vehicle devices (car navigation device, air conditioner device, car audio device, etc.) is controlled using the voice of the vehicle occupant as an input signal. In general, spoken speech is analyzed based on an acoustic model (a model in which the correspondence between sound waveform samples and phonemes is specified), and based on a language model (a model in which phonemes are connected) The contents are recognized (see Patent Document 1). By using such a voice recognition technology, it is possible to operate the vehicle-mounted device in a hands-free manner.

JP 2002-189492 A

  By the way, the above acoustic model and language model are created based on the voice of a standard speaker. On the other hand, the actual speech includes acoustic fluctuations (frequency fluctuations, waveform fluctuations) due to the structure of the vocal organs of the speaker, the vocalization method, the social environment (dialect, intonation), etc. It does not have the same characteristics as the voice. Therefore, sufficient speech recognition accuracy may not be obtained using a standard model. In particular, in a vehicle voice recognition control system, the same person does not always get on the vehicle, so that the voice recognition accuracy and the controllability of various in-vehicle devices are likely to deteriorate.

  One of the objects of the present case was invented in view of the above-described problems, and is to provide a voice recognition control system that improves the voice recognition accuracy of a vehicle occupant. It should be noted that the present invention is not limited to this purpose, and is an operational effect that is derived from each configuration shown in “Mode for Carrying Out the Invention” to be described later. Can be positioned as a purpose.

  (1) The voice recognition control system disclosed here is a voice recognition control system that controls an in-vehicle device using a voice of a vehicle occupant as an input signal. The system includes a specifying unit that specifies a person who has emitted the sound based on at least the sound. A selection unit configured to select an acoustic model corresponding to the person specified by the specifying unit; and a recognition unit configured to recognize the voice using the acoustic model selected by the selection unit.

  It is preferable that the recognition unit recognizes the voice of the person using an acoustic model corresponding to the person specified by the specifying unit. Moreover, it is preferable that the said specific | specification part pinpoints the position (speaking position) where the said voice was uttered, and the said person who exists in the said position. The acoustic model preferably includes a language model.

(2) It is preferable to provide an indoor camera that captures an image of the interior of the vehicle, and the specifying unit specifies the person who has emitted the sound using the sound and the image.
(3) It is preferable that the specifying unit specifies the person by comparing the movement of the lips of the person and the timing of the voice.
(4) It is preferable to provide a data updating unit that creates an acoustic model corresponding to the person specified by the specifying unit using the voice.
(5) It is preferable that a control unit that controls the control target is provided when the person specified by the specifying unit corresponds to the control target recognized by the recognition unit.

  By selecting an acoustic model corresponding to the person specified by the specifying unit, it is possible to improve the voice recognition accuracy in the passenger compartment where a plurality of passengers can board.

1 is a schematic top view of a vehicle to which a voice recognition control system is applied. It is a schematic diagram which shows the structure of a speech recognition control system. It is a flowchart for demonstrating the control content of a speech recognition control system.

  A speech recognition control system as an embodiment will be described with reference to the drawings. Note that the embodiment described below is merely an example, and there is no intention to exclude various modifications and technical applications that are not explicitly described in the following embodiment. Each configuration of the present embodiment can be implemented with various modifications without departing from the spirit thereof. Further, they can be selected as necessary, or can be appropriately combined.

[1. Device configuration]
The voice recognition control system of this embodiment is applied to the vehicle 10 shown in FIG. A driver's seat 14 and a passenger seat 15 are provided in the passenger compartment of the vehicle 10, and an instrument panel (instrument panel, dashboard) is disposed on the front side of the passenger compartment. A steering device and instruments are arranged in front of the driver's seat 14 and a glove box is arranged in front of the passenger seat 15 in the portion of the instrument panel facing the passenger compartment. A multi-communication type display device 16 that collects and provides user interfaces such as a car navigation function and an AV function is mounted in the center of the instrument panel in the vehicle width direction. The position of the display device 16 is diagonally forward left from the viewpoint of the driver sitting in the driver's seat 14 and diagonally forward right from the viewpoint of the passenger sitting in the passenger seat 15.

  The display device 16 includes a general-purpose video display device (display screen) having a touch panel and an electronic control device (computer) including a CPU (Central Processing Unit), ROM (Read Only Memory), RAM (Random Access Memory), and the like. An electronic device provided. The display device 16 is connected to in-vehicle devices such as the navigation device 11, the air conditioner device 12, the car audio device 13, and the multimedia system, and can function as an input / output device for various in-vehicle devices. For example, route information from the navigation device 11 to the destination, map information, traffic jam information, and the like can be displayed on the display screen of the display device 16. The display screen of the display device 16 includes a program received by an in-vehicle digital terrestrial broadcast tuner, DVD video content, video shot by a rear view camera, an interface for operating the air conditioner device 12 and the car audio device 13, and the like. Various visual information can be reproduced and displayed.

  In addition, the vehicle 10 is equipped with a voice recognition control device 1 that controls various in-vehicle devices using an occupant's voice as an input signal. The speech recognition control device 1 is an electronic device (ECU, electronic control device) in which a processor such as a CPU or MPU (Micro Processing Unit) and a ROM, RAM, nonvolatile memory, etc. are integrated. The processor here is, for example, a processing device (processor) including a control unit (control circuit), an arithmetic unit (arithmetic circuit), a cache memory (register), and the like. The ROM, RAM, and nonvolatile memory are memory devices that store programs and working data. The contents of the control performed by the voice recognition control apparatus 1 are recorded in ROM, RAM, nonvolatile memory, and removable media as firmware and application programs. When the program is executed, the contents of the program are expanded in the memory space in the RAM and executed by the processor.

  As shown in FIG. 2, a microphone array 21 and an indoor camera 22 as input devices are connected to the voice recognition control device 1. The microphone array 21 is an audio input device in which a plurality of microphones are arranged in a predetermined arrangement, and the indoor camera 22 is a wide-angle video camera capable of photographing the entire vehicle interior. On the other hand, as an output device, a navigation device 11, an air conditioner device 12, a car audio device 13, a display device 16, and the like are connected. The voice recognition control device 1 controls various in-vehicle devices based on the voice input from the microphone array 21 and the image taken by the indoor camera 22.

[2. Control configuration]
The voice recognition control device 1 has a function of identifying a person who has made a voice and identifying voice contents using an acoustic model suitable for the person. For example, when someone gets on the vehicle 10, the person who gets on the vehicle 10 is recognized and the position where the person is seated is stored. Also, when someone utters a sound in the passenger compartment, the person who uttered the sound is specified, and the sound is recognized using an acoustic model corresponding to the person. Different acoustic models are applied to persons with different voice characteristics. The acoustic model here includes a language model.

  As elements for performing the above control, the speech recognition control device 1 is provided with a specifying unit 2, a selection unit 3, a recognition unit 4, a database 5, and a control unit 6. These indicate some functions of a program executed by the speech recognition control apparatus 1 and are realized by software. However, some or all of the functions may be realized by hardware (electronic control circuit), or may be realized by using software and hardware together.

  The database 5 is a storage device in which various general data relating to speech recognition are recorded and stored. Here, a large number of acoustic models used in speech recognition are recorded and stored in a state associated with a person corresponding to the acoustic model. One of the acoustic models possessed by the database 5 is created in advance based on the voice of a standard speaker. Other acoustic models are created, learned, and updated based on the voice of a person other than the standard speaker, and are suitable for recognizing each voice for each occupant on the vehicle 10 (optimized). ) For example, if the input speech features (for example, the acoustic spectrum) are far from the standard speaker speech features, a new acoustic model corresponding to the input speech is created and the speech is recognized. The acoustic model is learned and updated based on the result.

  The identifying unit 2 identifies a person who has emitted the sound based on at least the sound input from the microphone array 21. As a method for identifying a person, a method for identifying in real time when sound is detected, and a relationship between the person and the seating position when someone gets on the vehicle 10, and a sound source of the detected sound And a method of identifying a person based on the position.

  In the former case, it is possible to specify a person based on a waveform pattern, a frequency pattern, a voiceprint pattern, etc. included in the voice. Alternatively, it is also possible to extract a human face from an image photographed by the indoor camera 22 and specify a person whose lip movement matches the voice timing. In the latter case, a person may be specified by analyzing an image captured by the indoor camera 22 (for example, face authentication), or the person is uttered with some sound, and the same method as the former is used. The person may be specified. Information on the person specified here is transmitted to the selection unit 3.

  The selection unit 3 selects an acoustic model corresponding to the person specified by the specifying unit 2 from among a plurality of acoustic models recorded and stored in the database 5. For example, when the user A who is the owner of the vehicle 10 issues a voice command, the first acoustic model corresponding to the user A is selected. When a user B different from the user A issues a voice command, the second acoustic model corresponding to the user B is selected. The selection result here is transmitted to the recognition unit 4.

  The recognition unit 4 recognizes speech using the acoustic model selected by the selection unit 3. Here, the context of the utterance content is analyzed, the type of the in-vehicle device to be controlled is estimated, and the content of the voice command for the control target is recognized. A specific method of speech recognition is arbitrary, and a known speech recognition technique can be employed. For example, after a phoneme included in speech is analyzed based on an acoustic model, a word or phrase consisting of a series of phonemes is analyzed based on a language model, and its semantic content is recognized. The recognition result here is transmitted to the control unit 6.

  The control unit 6 actually controls the control target based on the result recognized by the recognition unit 4. Here, when the combination of the person specified by the specifying unit 2 and the control target is appropriate, the control target is actually controlled. For example, the user A who is the owner of the vehicle 10 can operate the navigation device 11, the air conditioner device 12, and the car audio device 13 with voice commands. On the other hand, the user C who is an acquaintance of the user A cannot operate these in-vehicle devices with voice commands. If the combination of the person and the control target is not appropriate, the voice command is canceled (cancelled) without being controlled.

[3. flowchart]
FIG. 3 is an example of a flowchart for explaining the control contents executed by the speech recognition control apparatus 1. First, voice information detected by the microphone array 21 and image information taken by the indoor camera 22 are input to the voice recognition control device 1 (step A1), and it is determined whether or not someone has boarded the vehicle 10 ( Step A2). When someone's boarding is detected, the person is recognized and the seating position of the person is specified (step A3). Thereafter, it is determined whether or not a voice is input (step A4).

Here, if any sound is input, the identifying unit 2 identifies the person who has emitted the sound based on at least the sound information (step A5). Further, it is determined whether or not the person is a user registered in the database 5 (step A6). If not registered, a new acoustic model is additionally registered in the database 5 (step A7). Thereafter, the selection unit 3 selects an acoustic model corresponding to the person (step A8). Further, the recognition unit 4 performs speech recognition based on the selected acoustic model (step A9).
When the speech recognition is completed, the acoustic model of the person is learned and updated based on the recognition result (step A10). Note that the control unit 6 determines a correspondence relationship between the specified person and the control target, and when corresponding, controls the control target. On the other hand, if the specified person does not correspond to the control target, the voice command is canceled and the control is not performed.

[4. Action, effect]
(1) As described above, by selecting the acoustic model corresponding to the person specified by the specifying unit 2, it is possible to improve the voice recognition accuracy in the cabin where a plurality of passengers can board. Thus, for example, even if the structure, utterance method, and social environment of the utterance organ of the speaker are significantly different from those of the standard speaker, the contents of the voice command can be accurately recognized, and various in-vehicle devices Controllability can be improved.

(2) Concerning the control contents in the specifying unit 2, not only the audio information acquired by the microphone array 21 but also the image information of the indoor camera 22 is used in a short time compared with the case where only the audio is used. The person can be easily identified. Therefore, the speech recognition accuracy and the controllability of various in-vehicle devices can be improved with a simple configuration.
(3) Further, by comparing the movement of the lips in the image and the timing of the voice, it can be surely specified that the person is a speaker.

  (4) By learning and updating the acoustic model of the specified person with the voice of the person, the subsequent voice recognition accuracy can be further improved. Further, since a new acoustic model is additionally registered for a user who has not been registered in the database 5, the expandability of the database 5 can be improved, and voice recognition accuracy and various in-vehicle functions can be improved with a simple configuration. The controllability of the apparatus can be improved.

  (5) By determining the correspondence between the person specified by the specifying unit 2 and the control target with respect to the control contents in the control unit 6, it becomes easy to set the usage authority of the in-vehicle device for each user. For example, when the user A who is the owner of the vehicle 10 asks the user C who is an acquaintance to temporarily drive, the user who is not familiar with the voice command by restricting the voice command by the user C An erroneous operation due to C can be prevented, and convenience can be improved.

[5. Modified example]
In the above-described embodiment, the acoustic model corresponding to each person specified by the specifying unit 2 is recorded and stored in the database 5, but the acoustic model is not necessarily set for each user. For example, if the input voice feature is not far from the voice feature of the standard speaker, no dedicated acoustic model is required for that person. Also, the characteristics of the input speech are classified into several types (for example, six types such as young men, young women, older men, older women, older men, older women), and an acoustic model is assigned to each type. It may be set. In this case, speech recognition may be performed using an acoustic model corresponding to the type associated with the specified person.

  Further, in the above-described embodiment, all processes from speech recognition to control of the control target are managed in an integrated manner by the speech recognition control device 1, but some or all of the functions of the speech recognition control device 1 are controlled by the vehicle. Relocation to the outside of 10 is also conceivable. For example, the voice recognition control device 1 can be connected to a network such as the Internet, a mobile phone wireless communication network, and other digital wireless communication networks, and a part or all of the functions of the voice recognition control device 1 are mounted on a server on the network. May be. Thereby, management and update of the database 5 become easy, and speech recognition accuracy and gesture recognition accuracy can be improved.

DESCRIPTION OF SYMBOLS 1 Voice recognition control apparatus 2 Identification part 3 Selection part 4 Recognition part 5 Database 6 Control part 10 Vehicle 11 Navigation apparatus 12 Air conditioner apparatus 13 Car audio apparatus 14 Driver's seat 15 Passenger's seat 16 Display apparatus 21 Microphone array 22 Indoor camera

Claims (5)

In a voice recognition control system that controls an in-vehicle device using the voice of a vehicle occupant as an input signal,
A specifying unit for specifying a person who has emitted the sound based on at least the sound;
A selection unit that selects an acoustic model corresponding to the person identified by the identification unit;
A recognition unit for recognizing the voice using the acoustic model selected by the selection unit;
A voice recognition control system comprising: It has an indoor camera that captures images in the passenger compartment,
The voice recognition control system according to claim 1, wherein the specifying unit specifies a person who has emitted the voice using the voice and the image. The voice recognition control system according to claim 2, wherein the specifying unit specifies the person by comparing the movement of the lip of the person and the timing of the voice. The voice recognition control according to any one of claims 1 to 3, further comprising a data update unit that creates an acoustic model corresponding to the person specified by the specifying unit using the voice. system. The control unit that controls the control target when the person specified by the specifying unit and the control target recognized by the recognition unit correspond to each other. The speech recognition control system according to item 1.

Images