JP2007155986A - Voice recognition device and robot equipped with the same - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a voice recognition device and a robot equipped with the same, capable of fast-paced speaking, while preventing first part missing in voice recognition. <P>SOLUTION: The robot comprises: a control section 53 for composing words of speech; a voice synthesizing section 55 for generating an output voice signal based on the words; a speaker 18 for outputting an output voice based on an output voice signal; a microphone 14 for converting voice including at least user's voice to an input signal; an output voice elimination section 57 for generating an input voice signal by eliminating a signal component related to the output voice from the input signal; and a voice recognition section 59 for outputting a recognition result to the control section 53 by recognizing the user's voice based on the input voice signal. The control sections 53 and 61 control timing of starting recognition of the user's voice by the voice recognition section 59, after a predetermined period from start of outputting the output voice, and before finish of outputting the output voice. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a voice recognition device and a robot including the voice recognition device.

With the development of computer technology in recent years, robots using computers have been developed. Such robots include not only industrial robots used at manufacturing sites and the like, but also robots that are closely related to people, such as robots that take care of children (see, for example, Patent Document 1).
There are various methods for giving instructions to robots that are closely related to people. One of them is a method using voice dialogue.

In a voice dialogue performed with a robot as a partner, a method of repeating confirmation in order of a human instruction, a robot response, a human instruction, and a robot response is generally used. For example, in a system such as a telephone and a robot that does not use a headset, voice recognition is generally started after the end of the utterance of the robot. This is because if speech recognition is started while the robot is speaking, the robot's own speech may be recognized.
However, people tend to talk without waiting after the end of the utterance of the robot or before the end of the utterance. In particular, in the case of a person who has experienced the same conversation scenario, the above tendency appears remarkably. As described above, when a person speaks before the robot starts voice recognition, the robot cannot recognize the first part of the person's story, and thus cannot correctly recognize the person's story.

As a solution to the above problem, a system that informs a person that a robot has started speech recognition is generally known.
For example, a system is known in which a voice recognition start timing of a robot is transmitted to a person by turning on a lamp provided at the ear of the robot. Also known is a system that encourages human speech by sounding a short beep after the end of the robot's speech.
By using these systems, it is possible to obtain a certain effect that the robot can recognize a person's story from the beginning. However, when these systems are used, a person speaks at the timing specified by the robot. Thus, when the timing of speaking is restricted, people may feel stress.

For this reason, various speech recognition apparatuses have been proposed that deal with the above-mentioned tendency of people and accurately recognize people's stories and do not cause people to feel stress (see, for example, Patent Documents 2 and 3).
JP 2005-305631 A (page 8-11) JP 2003-345390 A (page 3, FIG. 1) JP 2004-333543 A (page 8-9, FIG. 1)

In the above-mentioned Patent Document 2, an inclination sensor that detects inclination, input preparation detection means that detects that a user is going to input based on the output of the inclination sensor, and a press talk button that is pressed at the start of voice input; A configuration of a speech processing apparatus including a buffer memory for temporarily storing speech and speech recognition processing means for performing speech signal recognition processing is disclosed.
According to this configuration, the tilt sensor detects that the user has picked up the voice processing device in order to press the press talk button, and the input preparation detection unit instructs preparation for voice input. Then, the audio signal is stored in the buffer memory before the user presses the press talk button. Thereafter, when the press talk button is pressed, the voice recognition processing means starts recognizing the voice signal stored in the buffer memory.
As described above, it is described that the speech signal stored before the press talk button is pressed is also recognized by the speech recognition processing means, so that the person's story can be recognized from the beginning.
However, in the above-described speech processing apparatus described in Patent Document 2, it is necessary to press the press talk button in order to start speech recognition by the speech recognition processing means, which may cause stress to the person.

In the above-mentioned Patent Document 3, a speech output unit that outputs system-side speech, a microphone that converts user speech into speech signals, a speech recognition unit that recognizes user speech, and a user's spoken dialogue proficiency level are determined. From the proficiency level determination unit, the audio output change unit that changes the output of the system side audio, and the audio signal input from the microphone, the signal corresponding to the output of the system side audio output by the audio output unit is subjected to correlation calculation and removed. A configuration of a voice interaction system including a voice response removing unit is disclosed.
According to this configuration, since the voice response removing unit is provided, it is described that the voice from the user can be recognized even when the voice dialogue system outputs a voice response.
However, in such a voice response removal unit (for example, an acoustic echo canceller, hereinafter referred to as AEC), as the nature of the processing, when reflection of sound in the environment is complicated, or other noise or distortion Due to factors such as the above, there is a possibility that the system side voice cannot be completely removed.
Also, immediately after the system side voice is output from the voice output unit, the system side voice removal processing in the voice response removal unit has not converged, and the voice recognition performance from the user may be degraded.

For example, if the system side voice cannot be completely removed and the voice recognition is performed while the system side voice is being output, the system side voice may be erroneously recognized as the user's voice. For this reason, the voice interaction system may misrecognize its own system-side voice without waiting for the user to speak.
Even when the voice response removal unit can completely remove the system side voice, if the voice recognition is performed from the beginning of the output of the system side voice, the external noise (other than the system side voice and the user's voice other than the system side voice) will be generated at the beginning of the output of the system side voice. If the sound is generated, this external noise may be mistakenly recognized as the user's voice.
When such misrecognition is performed, the voice dialogue system advances the conversation based on the misrecognition, and thus there is a problem that it is not possible to perform a conversation with accurate contents at a good tempo.

  The present invention has been made to solve the above-described problem, and is a voice recognition device capable of preventing speech recognition from being interrupted and realizing a conversation with a good tempo, and a robot including the voice recognition device. The purpose is to provide.

In order to achieve the above object, the present invention provides the following means.
The speech recognition apparatus according to the present invention includes a control unit that assembles dialogue lines, a speech synthesis unit that generates output speech signals based on the assembled dialogues, and a speaker that outputs output speech based on the generated output speech signals. And a microphone that converts at least user speech uttered by the user into an input signal, and based on the output speech signal, a signal component related to the output speech is removed from the input signal to generate an input speech signal An output speech removal unit; and a speech recognition unit that recognizes the user speech based on an input speech signal and outputs a recognition result to the control unit, wherein the control unit recognizes the speech based on the dialogue. A timing of starting recognition of the user voice by a unit is controlled after a predetermined time from the start of output of the output voice and before the end of output of the output voice. To.

According to the present invention, the control unit controls the timing of the start of user speech recognition after starting the output of the output speech based on the dialogue and before the end of the output, thereby preventing the user speech recognition from being interrupted, A conversation with good tempo can be realized.
Since the user voice recognition start timing is controlled by the control unit before the user voice output is finished, the user voice recognition can be started before the output of the output voice is finished. Therefore, even if the user speaks immediately after the output of the output voice or during the output of the output voice, the voice recognition device can recognize the user's voice from the beginning, and can prevent the voice recognition from being interrupted. A good conversation can be realized.
Since the user speech recognition start timing is controlled based on the dialogue, even if the dialogue length changes, the user speech recognition can always be started before the output of the output speech is finished.
Since the user voice recognition start timing is a predetermined time after the output start of the output voice, the user voice can be recognized in a state where the process of the output voice removing unit is stable. Immediately after the output of the output voice is started, the process of the output voice removal unit is unstable, and in this state, there is a possibility that erroneous recognition of the user voice occurs. As described above, the user voice recognition start timing is set to be a predetermined time after the start of the voice output, thereby preventing erroneous recognition of the user voice and realizing a conversation with a good tempo.
Immediately after the output of the output voice is started, there is a high possibility that the user's utterance content is not valid for the speech related to the output voice. Therefore, by setting the user voice recognition start timing after a predetermined time from the output start of the output voice, it is possible to prevent voice recognition of the invalid answer and realize a conversation with a good tempo.

  In the above invention, an utterance time calculation unit that calculates the length of the utterance time of the output voice related to the dialogue based on the output voice signal is provided, and the control unit is based on the length of the utterance time. It is desirable to control the timing of the recognition of the user voice by the voice recognition unit.

According to the present invention, since the control unit controls the timing of the start of recognition of the user voice based on the length of the utterance time calculated by the utterance time calculation unit, it is possible to reliably prevent the user voice recognition from being interrupted. , Can achieve a conversation with good tempo.
Since the utterance time calculation unit calculates the length of the utterance time based on the output sound signal input to the speaker, it is possible to calculate the length of the utterance time of the output sound actually output from the speaker. it can. Since the control unit controls the timing of starting the recognition of the user voice based on the calculated length of the utterance time, the head of the user voice recognition can be surely prevented.
For example, even if a part of the dialogue includes an individual name, a nickname, etc., and part of the dialogue is changed by conversation, the utterance time calculation unit is based on the output audio signal related to the changed dialogue. The length of the utterance time can be calculated. Therefore, the voice recognition device can surely prevent the user voice recognition from being interrupted.

  In the above invention, a storage unit that stores in advance the length of the utterance time of the output speech related to the line, the control unit based on the length of the utterance time stored in the storage unit, It is desirable to control the timing of starting the recognition of the user voice by the voice recognition unit.

According to the present invention, the length of the utterance time of the output voice related to the dialogue is stored in advance in the storage unit, and the control unit controls the timing of starting the recognition of the user voice based on the stored length of the utterance time. Therefore, it is possible to surely prevent the user voice recognition from being interrupted and realize a conversation with a good tempo.
For example, since it is not necessary to calculate the utterance time based on the output audio signal input to the speaker, it is not necessary to calculate the utterance time. it can. Further, since it is not necessary to use the utterance time calculation unit, the configuration of the speech recognition apparatus can be simplified.

  In the above invention, the control unit calculates a start time obtained by subtracting a delay time of a predetermined length from the utterance time, and when the start time elapses from the start of output of the output speech, the control unit recognizes the speech recognition unit. It is desirable to start recognition of user speech.

According to the present invention, since the recognition of the user voice is started when the start time has elapsed from the start of output of the output voice, it is possible to prevent the user voice recognition from being interrupted and to realize a conversation with a good tempo. it can.
Since the start time is calculated by subtracting a predetermined delay time from the utterance time, the control unit can control the start of user speech recognition at a predetermined timing.
For example, it is possible to prevent erroneous recognition of the user's voice in the voice recognition device by controlling the predetermined timing to be immediately before the user speaks. That is, by shortening the interval from the start of speech recognition until the user speaks, the probability that an external sound is generated during that interval can be reduced. Therefore, it is possible to prevent the voice recognition device from erroneously recognizing the external sound as user voice.
The length of the delay time is preferably longer than zero and shorter than the length of the utterance time.

In the above invention, the control unit changes the length of the delay time,
It is desirable to control the timing of starting the recognition of the user voice by the voice recognition unit.

According to the present invention, the control unit controls the timing of the start of user speech recognition by changing the length of the delay time, thereby preventing the user speech recognition from being interrupted and realizing a conversation with a good tempo. be able to.
By changing the length of the delay time, the length of the start time obtained by subtracting the delay time from the speech time can be changed. Therefore, it is possible to change the start timing of user voice recognition.

  In the above invention, it is desirable that the control unit controls the timing of starting the recognition of the user voice in the voice recognition unit based on a sentence constituting the dialogue.

According to the present invention, since the control unit controls the start timing of user speech recognition based on the sentences constituting the dialogue, it is possible to prevent the user speech recognition from being interrupted and realize a conversation with a good tempo. it can.
For example, when the dialogue is composed of a plurality of sentences, by combining the timing of starting recognition of the user voice and the timing of outputting the output voice related to the second and subsequent sentences. Thus, it is possible to prevent the user's voice recognition from being interrupted and realize a conversation with a good tempo. That is, the user tends to listen to the output speech related to the first sentence without speaking, and to utter the output speech related to the second and subsequent sentences from the middle. For this reason, the user speech recognition is started in accordance with any one of the output timings of the output speech related to the second and subsequent sentences, thereby preventing the user speech recognition from being interrupted and reducing the tempo. A good conversation can be realized.
For example, it is possible to finely control the timing of starting user speech recognition according to the configuration of the dialogue, as compared with the case where the timing of starting user speech recognition is controlled by the time from the start of output of output speech. Therefore, it is possible to more reliably prevent the user voice recognition from being interrupted and realize a conversation with a good tempo.
Since the length of the utterance time is not used to control the timing of starting the recognition of the user voice, when it is difficult to calculate the length of the utterance time, or when it takes time to calculate the length of the utterance time, The user speech recognition start timing can be easily controlled.

In addition, the sentences constituting the dialogue include a sentence that is unlikely to be spoken by the user and a sentence that is likely to be spoken. Therefore, for example, the control unit may perform control so as to start recognition of the user voice from the time when the output voice related to the sentence that is likely to be spoken by the user is started. Or you may control to start recognition of a user's voice from the time when a dialogue changes from a sentence with a low possibility of being spoken by a user to a sentence with a high possibility of being spoken. By doing so, it is possible to prevent the user voice recognition from being interrupted and to realize a conversation with a good tempo.
Sentences that are unlikely to be spoken by the user include sentences such as a call to the user and a repetition of instructions from the user. Examples of the sentence that is highly likely to be spoken by the user include a sentence that requests an instruction to the user. The sentence described here is an arrangement of words in a certain format and method.

  A robot provided with a voice recognition device of the present invention is a robot provided with a voice recognition device that recognizes a user's voice, and the voice recognition device according to any one of claims 1 to 7. It is a device.

  According to the present invention, a robot equipped with a voice recognition device can use the voice recognition device of the present invention to prevent a user voice recognition from being interrupted and realize a conversation with a good tempo.

  According to the voice recognition device and the robot including the voice recognition device of the present invention, the control unit controls the timing of starting the recognition of the user voice after starting the output of the output voice based on the dialogue and before the end of the output. In addition to preventing the user's voice recognition from being interrupted, it is possible to realize a conversation with a good tempo.

Hereinafter, an embodiment of a robot according to the present invention will be described with reference to the drawings.
[First Embodiment]
FIG. 1 is a front view of the robot according to the first embodiment of the present invention, and FIG. 2 is a left side view of the life support robot shown in FIG.
As shown in FIGS. 1 and 2, the main body 1 of the life support robot includes a head 2, a chest 3 that supports the head 2 from below, a right arm 4 a provided on the right side of the chest 3, and a chest 3, a left arm 4 b provided on the left side, a waist 5 connected below the chest 3, a skirt 6 connected below the waist 5, and a leg connected below the skirt 6. 7.

One omnidirectional camera 11 is provided near the top of the head 2. A plurality of infrared LEDs 12 are arranged on the ring at predetermined intervals along the outer periphery of the omnidirectional camera 11.
In the vicinity of the center of the front surface of the head 2, as shown in FIG. 1, the front camera 13 for imaging the front is one on the right side when viewed from the front, and the microphone 14 is one on the left side when viewed from the front. Is provided.

  One monitor 15 is provided near the center of the front surface of the chest 3. One ultrasonic distance sensor 16 for detecting a person is provided above the monitor 15. One power switch 17 is provided below the monitor 15. Above the ultrasonic distance sensor 16, two speakers 18 are provided one on each side. In addition, as shown in FIG. 2, a backpack 33 that can store luggage is provided on the back of the chest 3. The school bag 33 is provided with an opening / closing door 33a that can be rotated around a hinge provided at the top. As shown in FIG. 1, one touch sensor 19 is provided on each of the left and right shoulders of the chest 3.

  A multi-joint structure is adopted for the right arm portion 4a and the left arm portion 4b. In the right arm portion 4a and the left arm portion 4b, side switches 20 are provided in the vicinity of the connection portion with the chest portion 3 to detect the pinching of a body or an object and stop the movement of the arm. As shown in FIG. 1, a handshake switch 21 that functions as a man-machine interface is built in the palm of the right arm 4a. For the side switch 20 and the handshake switch 21, for example, a pressure sensor is employed.

  In the vicinity of the center of the front surface of the waist 5, one ultrasonic distance sensor 22 for detecting a person is provided on each side. Below these ultrasonic distance sensors 22, a sensor region 24 in which a plurality of infrared sensors 23 are arranged is provided. These infrared sensors 22 are for detecting an obstacle or the like in the lower front of the robot body 1. As shown in FIG. 1 and FIG. 2, a total of four microphones 25 are provided below the waist 5 for detecting the sound source direction, one on the left and one on the front and back. As shown in FIG. 2, one handle portion 26 used for lifting the main body is provided on each of the left and right sides of the waist portion 5. The handle 26 is a recess so that the operator's hand can be inserted.

  Below the front surface of the skirt portion 6, a total of three infrared sensors 27 for detecting a step are provided in the center and on the left and right. As shown in FIG. 2, a charging connector 28 is provided on the back surface of the skirt portion 6.

As shown in FIG. 1, one infrared sensor 29 for detecting a lateral distance is provided on the front surface of the leg portion 7 on the left and right sides. These infrared sensors 29 are mainly used for level difference detection.
As shown in FIG. 2, a hook 30 for fixing the position of the robot body 1 to the charging stand is provided on the back surface of the leg portion 7. The leg portion 7 is a carriage provided with traveling wheels 31 and four ball casters 32.

  In the robot described above, the facial expression of the head 2 is variable by a drive mechanism (not shown). Also, a neck joint between the head 2 and the chest 3, a shoulder joint between the chest 3 and the right arm 4a, a shoulder joint between the chest 3 and the left arm 4b, an elbow joint in the right arm 4a and the left arm 4b, a wrist joint, etc. Can be driven by a driving mechanism (not shown), and the driving wheels 31 mounted on the legs 7 are driven by a driving mechanism (not shown) so that automatic steering and automatic driving are possible.

In addition, the robot according to the present embodiment is configured to move autonomously by supplying power from a battery built in the robot body, and coexists with a human being as a working space indoors. For example, it is used to assist, support, and care for the lives of users such as robot owners and operators in general households. Therefore, in addition to the conversation function for realizing the conversation with the user, the robot 1 has a function of watching the user's action, assisting the user's action, and acting with the user. Such a function is realized by, for example, a control device including a microcomputer or the like built in the main body of the robot 1. Various cameras and various sensors shown in FIGS. 1 and 2 are connected to the control device, acquire image information from the cameras and sensor detection information from the sensors, and execute various programs based on these information. By executing this, the various functions described above are realized.
The shape of the robot body 1 is not limited to the shape shown in FIGS. 1 and 2, and various shapes such as a model imitating an animal for pets can be adopted.

Next, the speech recognition function that is a characteristic part of the present invention will be described. The voice recognition function of the robot according to the present embodiment is realized by the voice recognition device provided in the control device described above. FIG. 3 shows a functional block diagram of the speech recognition apparatus according to the present embodiment.
As shown in FIG. 3, the speech recognition apparatus 51 includes a conversation scenario execution unit (control unit) 53 that generates a dialogue used for conversation with the user, a speech synthesis unit 55 that generates an output speech signal based on the dialogue, A speaker 18 that outputs output sound based on the output sound signal, a microphone 14 that converts sound including at least user's sound into an input signal, and an acoustic echo canceller (hereinafter referred to as AEC) that generates an input sound signal from the input signal (Output voice removal unit) 57, voice recognition unit 59 that recognizes user voice based on the input voice signal, delay control unit (control unit) 61 that instructs the voice recognition unit 59 to start recognition, and dialogue An utterance time calculation unit 63 that calculates the utterance time length of the output voice, and a proficiency level determination unit 65 that determines the proficiency level of the user's conversation.

The conversation scenario execution unit 53 selects a scenario for conversation with the user and generates a dialogue based on the selected scenario. The conversation scenario execution unit 53 includes a proficiency level determination unit 65 that determines a user's proficiency level, and a storage unit 67 that stores the determined proficiency level. The conversation scenario execution unit 53 is connected to the speech synthesis unit 55 and the delay control unit 61 so as to output an electrical signal to the speech synthesis unit 55 and the delay control unit 61. The conversation scenario execution unit 53 is connected to the voice recognition unit 59 so that an electric signal is input from the voice recognition unit 59.
The proficiency level determination unit 65 determines the proficiency level of the user's conversation, and is provided in the conversation scenario execution unit 53. The conversation scenario execution unit 53 changes the delay time length D based on the output of the proficiency level determination unit 65. For the determination of the proficiency level, for example, the number of days that have passed since the above-mentioned robot is turned on, the number of times the same content scenario is repeated, or the like is used.
The voice synthesis unit 55 generates an output voice signal based on the dialogue generated by the conversation scenario execution unit 53. The speech synthesizer 55 is connected to the conversation scenario execution unit 53 so that an electric signal is input from the conversation scenario execution unit 53. The voice synthesis unit 55 is connected to the speaker 18, the AEC 57, and the utterance time calculation unit 63 so as to output electric signals to the speaker 18, the AEC 57, and the utterance time calculation unit 63.

The speaker 18 outputs output sound to, for example, a user based on the input output sound signal. The speaker 18 is connected to the speech synthesizer 55 so that an electrical signal is input from the speech synthesizer 55. Note that a known speaker can be used as the speaker 18 and is not particularly limited.
The microphone 14 converts sound input to the microphone 14 including the user's utterance into an input signal that is an electrical signal. The microphone 14 is connected to the AEC 57 so that an electrical signal is output to the AEC 57. Note that a known microphone can be used as the microphone 14 and is not particularly limited.

The AEC 57 removes a signal corresponding to the output sound output from the speaker 18 from the input signal by performing a correlation operation between the output sound signal input from the sound synthesizer 55 and the input signal input from the microphone 14. The input audio signal is calculated. The AEC 57 is connected to the speech synthesizer 55 so that an electrical signal is input from the speech synthesizer 55. The AEC 57 is connected to the voice recognition unit 59 so that an electrical signal is output to the voice recognition unit 59.
The voice recognition unit 59 recognizes a user's utterance. Specifically, the user's utterance is recognized by matching an input voice signal input from the AEC 57 with a voice recognition dictionary. The voice recognition unit 59 is connected to the conversation scenario execution unit 53 so as to output an electrical signal to the conversation scenario execution unit 53. The voice recognition unit 59 is connected to the AEC 57 and the delay control unit 61 so that electric signals are input from the AEC 57 and the delay control unit 61.

The delay control unit 61 instructs the voice recognition start timing in the voice recognition unit 59. The delay control unit 61 is connected to the conversation scenario execution unit 53 and the utterance time calculation unit 63 so that electric signals are input from the conversation scenario execution unit 53 and the utterance time calculation unit 63. The delay control unit 61 is connected to the voice recognition unit 59 so that an electric signal is input from the voice recognition unit 59.
Specifically, the delay control unit 61 first performs speech recognition based on the speech duration length X calculated by the speech synthesis unit 55 and the delay time length D calculated by the conversation scenario execution unit 53. The start time (XD) is calculated. Thereafter, the delay control unit 61 outputs a speech recognition start signal to the speech recognition unit 59 when the start time (X-D) has elapsed from the output start of the output speech.

Next, the operation of the voice recognition device 51 in the conversation between the life support robot having the above-described configuration and the user will be described.
First, the flow of conversation between the life support robot and the user in this embodiment will be described.
FIG. 4 is a schematic diagram for explaining the flow of conversation between the speech recognition apparatus of FIG. 3 and the user. In FIG. 4, the horizontal axis represents time, SP in the figure represents the utterance period of the life support robot, and RC represents the period during which the life support robot recognizes the voice.
When the user does not have a conversation, as shown in FIG. 4, the voice recognition unit 59 of the life support robot continues to recognize the voice (A), and waits for the voice input from the user. Yes.
In this state, when a voice instruction is input from the user, the voice recognition device 51 interrupts the voice recognition (B) and utters a speech in response to the input instruction (SP). The voice recognition device 51 starts voice recognition (RC) again at a time (C) when a predetermined time (X-D) has elapsed from the start of utterance, and starts to recognize the user's voice input. If the state where there is no voice input of a predetermined level or more for a certain period due to the user's voice input being terminated, the voice recognition device 51 delimits the voice recognition, and performs the following processing based on the result of the voice recognition. Do.
In this way, a catch ball type conversation in which the conversation is alternately repeated between the life support robot and the user is performed.

Next, the operation of the voice recognition device 51 when the above-described catch ball type conversation is performed will be described.
When a voice input instruction is input from the user while the life support robot is waiting for a voice instruction input from the user, the conversation scenario execution unit 53 of the voice recognition device 51, as shown in FIG. A stop signal for stopping voice recognition is output to the recognition unit 59. At the same time, the conversation scenario execution unit 53 assembles a conversation scenario according to an instruction input from the user, and generates a dialogue based on the assembled scenario. Alternatively, the dialogue stored in the recognition dictionary of the storage unit 67 provided in the conversation scenario execution unit 53 is called. At this time, the generated or called dialogue includes dialogue related to the user expected to answer the voice of the life support robot in addition to speech related to the voice output by the life support robot.
For example, when the input instruction from the user is “message”, the dialogue related to the robot in the conversation scenario assembled by the conversation scenario execution unit 53 is “I will convey the message. Further, the dialogue related to the user is the name of the other party to be notified, such as “Mr. A” and “Mr. B”.

Further, the conversation scenario execution unit 53 calculates the delay time length D based on the user's proficiency level determined by the proficiency level determination unit 65. The user's proficiency level is determined in advance by the proficiency level determination unit 65 and is stored in the storage unit 67. For example, the proficiency level determination unit 65 determines that the proficiency level of the user has increased when the number of days that have elapsed since the power of the robot is turned on, and generates a signal for increasing the length D of the delay time. Alternatively, when the number of repetitions of the scenario having the same content increases, it is determined that the user's proficiency level has increased, and a signal for increasing the delay time length D is generated.
The conversation scenario execution unit 53 outputs a speech signal related to the robot to the speech synthesizer 55, and delay-controls the speech recognition start signal to the speech recognition unit 59, the speech signal related to the user, and the signal of the delay time length D. To the unit 61.
The speech synthesizer 55 to which the speech signal related to the robot is input generates an output speech signal to be output to the speaker 18. The output audio signal is a signal having a waveform corresponding to the dialogue related to the robot. The output audio signal is output toward the utterance time calculation unit 63, the AEC 57, and the speaker 18.
The speaker 18 outputs a speech related to the robot as output speech to the user based on the input output speech signal.

On the other hand, the utterance time calculation unit 63 calculates the utterance time length X of the output sound actually output from the speaker 18 based on the input output sound signal. For example, the length X of the utterance time can be obtained by counting and dividing the number of bytes of the output audio signal. In addition, the calculation method of the length X of speech time can use a well-known method, and is not specifically limited. When the utterance time length X is calculated, a signal related to the utterance time length X is output from the utterance time calculation unit 63 to the delay control unit 61.
The delay control unit 61 calculates the length of the start time based on the input signal of the speech time length X and the signal of the delay time length D. Specifically, a value (X−D) obtained by subtracting the delay time length D from the speech time length X is the length of the start time. The length of the start time (XD) refers to the time from when the output sound is output from the speaker 18 until the start of speech recognition in the speech recognition unit 59. When the length of the start time (X-D) is calculated, the delay control unit 61 starts measuring time, and when the length of the start time (X-D) has elapsed, the conversation scenario execution unit 53 The input voice recognition start signal for the voice recognition unit 59 and the speech signal for the user are output to the voice recognition unit 59.

The length D of the extension time is desirably set so that the length of the start time (X−D) is equal to or longer than the time required for the processing in the AEC 57 to be stabilized.
By setting in this way, voice recognition can be started from a state where the processing of AEC 57 is stable, and erroneous recognition of user voice can be prevented. “AEC 57 processing is stable” means that processing for removing a signal component corresponding to an output audio signal from an input signal is stable. By stabilizing this process, it is possible to stably calculate the input audio signal from which the output audio signal has been removed.

  In addition, it is desirable that the range of the delay time length D is greater than or equal to zero and less than or equal to the speech time length X. That is, when the delay time length D is zero, the start time length (X−D) is equal to the speech time length X, and speech recognition is started simultaneously with the end of the speech time. When the length D of the delay time is equal to the length X of the utterance time, the length of the start time (X−D) becomes zero, and voice recognition is started as the utterance time starts.

On the other hand, the microphone 14 converts the sound input to the microphone 14 into an input signal that is an electrical signal, and outputs the input signal to the voice recognition unit 59 via the AEC 57. The input signal is an electric signal having a waveform corresponding to the waveform of the sound input to the microphone 14 and is a signal including a user's voice, a voice output from the speaker 18, and other external noise components. .
The AEC 57 removes a signal component corresponding to the output audio signal from the input signal based on the input signal and the output audio signal, and an input audio signal is generated. Specifically, a correlation function between the input signal and the output audio signal is obtained, and a signal component corresponding to the output audio signal is removed from the input signal based on the obtained correlation function. Note that a known echo canceller can be used as the AEC 57 and is not particularly limited.
When no output audio signal is input to the AEC 57, the input signal is output to the audio recognition unit 59 without being processed.
The above-described microphone 14 and AEC 57 are always processed regardless of the presence or absence of a speech recognition start signal output from the conversation scenario execution unit 53.

The voice recognition unit 59 recognizes the input voice signal input from the AEC 57 based on the recognition start signal input from the delay control unit 61. Specifically, the speech signal related to the user input together with the recognition start signal is matched with the input speech signal, and it is determined whether or not the input speech signal matches the speech signal related to the user. Yes.
When there is no voice input from the user for a predetermined period, the voice recognition unit 59 delimits voice recognition and outputs a determination result to the conversation scenario execution unit 53. The conversation scenario execution unit 53 assembles the next conversation scenario based on the determination result. Thereafter, the above process is repeated.
For example, if the dialogue related to the robot is “I will tell you a message. Who do you want to tell?”, When the user answers “Mr. A”, matching is performed in the voice recognition unit 59, and the user is “Mr. A”. Is recognized and determined. The conversation scenario execution unit 53 assembles the next conversation scenario based on this determination.
Note that the voice recognition unit 59 may divide the voice recognition and output the determination result to the conversation scenario execution unit 53 when the voice from the user is not input for a predetermined period, or may end the voice recognition. As a case where the voice from the user is not input for a predetermined period, a case where a voice of a predetermined level or higher is not input for a predetermined time can be cited.

If it is determined by the voice recognition unit 59 that the content answered by the user is not valid, the conversation scenario execution unit 53 selects a scenario with the content to be guided by the function of the life support robot. When the conversation scenario execution unit 53 selects a scenario related to guidance, when the user's voice input is not appropriate, or when the user does not answer the question from the life support robot Can be mentioned.
FIG. 5 is a schematic diagram for explaining the flow of conversation when the voice recognition apparatus of FIG. 3 provides guidance.
When the speech recognition apparatus 51 of the life support robot performs guidance, as shown in FIG. 5, the speech recognition unit 59 starts speech recognition (RC) simultaneously with speech (SP) of speech related to the guidance.
In this case, as shown in FIG. 3, the conversation scenario execution unit 53 outputs a signal having an extension time length D equal to the speech time length X to the delay control unit 61, and the speech signal related to the guidance is spoken. The data is output to the combining unit 55. The voice synthesizer 55 generates an output voice signal based on the dialogue related to the input guidance, and outputs it to the speaker 18 and the utterance time calculator 63. On the other hand, the signal of the length X of the utterance time is input to the delay control unit 61, and the start time (X−D) is calculated. The delay control unit 61 immediately outputs a speech recognition start signal to the speech recognition unit 59 based on the calculation result (zero) of the start time (X−D).

  By controlling in this way, the speech recognition device 51 can recognize the user's utterance when the speech recognition device 51 of the life support robot starts guidance or when the user speaks during the guidance. In particular, since the guidance is started after a predetermined blank period in the voice input from the user, there is a high possibility that the user's utterance time approaches the guidance start time. Even in such a case, since the guidance start time and the voice recognition start time are the same, it is possible to prevent the user voice recognition from being interrupted.

FIG. 6 is a schematic diagram for explaining another example in the flow of conversation between the voice recognition apparatus of FIG. 3 and the user.
Further, depending on the contents of the scenario assembled by the conversation scenario execution unit 53, the user may not speak until the life support robot finishes speaking the scenario.
In this case, as shown in FIG. 6, the speech recognition device 51 of the life support robot does not perform speech recognition in the speech recognition unit 59 while performing speech (SP) related to the robot. When speech of the robot is finished, the speech recognition device 51 starts speech recognition (RC) in the speech recognition unit 59.
In this case, as shown in FIG. 3, the conversation scenario execution unit 53 outputs a signal having an extension time length D with a time length of zero to the delay control unit 61, and outputs a speech signal related to the robot to the speech synthesis unit. To 55. The voice synthesizer 55 generates an output voice signal based on the dialogue related to the input guidance, and outputs it to the speaker 18 and the utterance time calculator 63. On the other hand, the signal of the length X of the utterance time is input to the delay control unit 61, and the start time (X−D) is calculated. Based on the calculation result (X) of the start time (XD), the delay control unit 61 outputs a speech recognition start signal to the speech recognition unit 59 after the speech of the speech related to the robot is completed.

  By controlling in this way, the interval between the speech recognition start time by the speech recognition unit 59 and the user's speech start time can be shortened, and erroneous recognition of the user speech in the speech recognition device 51 can be prevented. That is, by shortening the interval from the start of speech recognition until the user speaks, the probability that an external sound is generated during that interval can be reduced. Therefore, the voice recognition device 51 can be prevented from erroneously recognizing the external sound as a user voice.

According to the above configuration, the delay control unit 61 sets the timing of the user speech recognition start in the speech recognition unit 59 after the start of output of the output speech and before the end of output based on the start time (XD). Therefore, the user voice recognition can be prevented from being interrupted, and a conversation with a good tempo can be realized.
Since the user voice recognition start timing is controlled by the delay control unit 61 before the end of the user voice output, the user voice recognition can be started before the output of the output voice ends. Therefore, even if the user speaks immediately after the output of the output sound or during the output of the output sound, the voice recognition unit 59 can recognize the user's voice from the beginning, and can prevent speech recognition from being interrupted. A conversation with good tempo can be realized.
Since the user voice recognition start timing is after the start time (XD) from the output start of the output voice, the user voice can be recognized in a state where the processing of the AEC 57 is stable. Immediately after the output of the output voice is started, the processing of the AEC 57 is unstable, and in this state, there is a possibility that erroneous recognition of the user voice occurs. As described above, the user voice recognition start timing is set to be after the start time (X-D) from the start of the voice output, thereby preventing misrecognition of the user voice and realizing a conversation with a good tempo. .
Immediately after the output of the output voice is started, there is a high possibility that the user's utterance content is not valid for the speech related to the output voice. For this reason, the user voice recognition start timing is set to be after the start time (XD) from the output voice output start, thereby preventing voice recognition of the invalid answer and realizing a conversation with a good tempo. it can.

Since the start time length (X-D) is calculated by subtracting the delay time length D from the utterance time length X, the delay control unit 61 starts the user speech recognition at a predetermined timing. Can be controlled.
Since the parameter for determining the length of the start time (X-D) includes the length X of the speech time related to the dialogue, even if the length of the dialogue changes, before the output of the output speech related to the dialogue is finished. In addition, user voice recognition can always be started.
It is desirable that the user voice recognition start timing is immediately before the user speaks. By doing so, erroneous recognition of the user voice in the voice recognition device 51 can be prevented. That is, by shortening the interval from the start of speech recognition until the user speaks, the probability that an external sound is generated during that interval can be reduced. Therefore, the voice recognition device 51 can be prevented from erroneously recognizing the external sound as a user voice.

Since the utterance time length X is calculated by the utterance time calculation unit 63 based on the output sound signal input to the speaker, the utterance time length X of the output sound actually output from the speaker is calculated. be able to. Since the delay control unit 61 controls the user speech recognition start timing based on the calculated utterance time length X, it is possible to reliably prevent the user speech recognition from being interrupted.
For example, even when a part of a line includes an individual name, a nickname, and the like and a part of the line is changed by conversation, the utterance time calculation unit 63 is based on the output audio signal related to the changed line. Thus, the length X of the utterance time can be calculated. Therefore, the voice recognition device 51 can surely prevent the user voice recognition from being interrupted.

As described above, the proficiency level determination unit 65 may determine the proficiency level of the user based on the number of days that have elapsed since the robot was turned on, the number of times the scenario has the same content, or the conversation. The time from when the speech is output from the scenario execution unit 53 to when the speech recognition result is input from the speech recognition unit 59 may be measured, and based on this time, the user's proficiency level may be determined. It is not a thing.
By determining the proficiency level in this way, it is possible to determine the proficiency level of the user more reliably and finely, prevent the user voice recognition from being interrupted, and realize a conversation with a good tempo.

As described above, the length of the utterance time of the output voice that is output from the speaker 18 one by one may be calculated by the utterance time calculation unit 63, or the length of the utterance time when a predetermined dialogue is output in advance. The measured value may be stored in the storage unit 67 in association with a predetermined dialogue.
By doing in this way, the calculation of the utterance time length by the utterance time calculation unit 63 can be omitted, and the calculation load during the utterance can be reduced. Further, since it is not necessary to use the utterance time calculation unit 63, the configuration of the speech recognition apparatus can be simplified.

[Second Embodiment]
Next, a second embodiment of the present invention will be described with reference to FIGS.
The basic configuration of the life support robot of this embodiment is the same as that of the first embodiment, but the timing of utterances in the speech recognition apparatus is different from that of the first embodiment. Therefore, in this embodiment, only the utterance timing in the speech recognition apparatus will be described with reference to FIGS. 7 and 8, and description of the main body of the life support robot will be omitted.
FIG. 7 is a block diagram illustrating functions of the speech recognition apparatus according to the present embodiment.
In addition, the same code | symbol is attached | subjected to the component same as 1st Embodiment, and the description is abbreviate | omitted.
As shown in FIG. 7, the speech recognition apparatus 151 includes a conversation scenario execution unit (control unit) 153 that generates dialogue used for conversation with the user, a speech synthesis unit 55 that generates an output speech signal, and outputs output speech. A speaker 18 that converts the input sound into an input signal, an AEC 57 that generates an input voice signal from the input signal, and a voice recognition unit 59 that recognizes the user voice based on the input voice signal. .

  The conversation scenario execution unit 153 selects a scenario of conversation with the user and generates a dialogue composed of a plurality of sentences based on the selected scenario. The conversation scenario execution unit 153 is connected to the speech synthesis unit 55 and the delay control unit 61 so as to output an electrical signal to the speech synthesis unit 55 and the delay control unit 61. The conversation scenario execution unit 153 is connected to the voice recognition unit 59 so that an electric signal is input from the voice recognition unit 59.

Next, the operation of the voice recognition device 151 in the conversation between the life support robot having the above-described configuration and the user will be described.
FIG. 8 is a schematic diagram for explaining the flow of conversation between the speech recognition apparatus of FIG. 7 and the user. In FIG. 8, the horizontal axis represents time, SP1 and SP2 in the figure represent the utterance period of the life support robot, and RC represents the period during which the life support robot recognizes the voice.
When a voice instruction is input from the user to the voice recognition device 151, the voice recognition device 151 is configured with a dialogue (for example, two sentences of a first sentence and a second sentence) for the input instruction as shown in FIG. 8. (SP1, SP2). Here, SP1 indicates an utterance related to the first first sentence, and SP2 indicates an utterance related to the next second sentence.
The speech recognition apparatus 151 does not perform speech recognition while performing the utterance SP1 related to the first sentence, and performs speech recognition RC simultaneously with starting the utterance SP2 related to the second sentence.

Next, the operation of the voice recognition device 151 when the above-described catch ball type conversation is performed will be described.
As shown in FIG. 7, the conversation scenario execution unit 153 of the speech recognition apparatus 151 assembles a conversation scenario according to an instruction input from the user, and dialogues based on the assembled scenario (for example, the first sentence and the second sentence). That consists of two sentences).
For example, when the input instruction from the user is “message”, the dialogue related to the robot in the conversation scenario assembled by the conversation scenario execution unit 53 is “I will convey the message. The sentence is “Tell the message.” The second sentence is “Who do you want to tell?”

The conversation scenario execution unit 153 sequentially outputs the first sentence signal and the second sentence signal to the speech synthesis unit 55. On the other hand, the conversation scenario execution unit 153 outputs a voice recognition start signal to the voice recognition unit 59 in synchronization with the second sentence signal.
The voice synthesizer 55 to which the first sentence and second sentence signals are input generates an output voice signal to be output to the speaker 18. The output audio signal is output toward the AEC 57 and the speaker 18. The speaker 18 outputs the lines related to the first sentence and the second sentence as output sound to the user based on the input output sound signal.

  The voice recognition unit 59 recognizes the input voice signal input from the AEC 57 based on the voice recognition start signal input from the conversation scenario execution unit 153. Here, the speech recognition start signal is output from the conversation scenario execution unit 153 simultaneously with the output of the second signal. Therefore, the speech recognition in the speech recognition unit 59 is started substantially simultaneously with the output of the speech related to the second sentence from the speaker 18.

According to the above configuration, since the user speech recognition in the speech recognition unit 59 is started at the timing when the output speech related to the second sentence is output from the speaker 18, the user speech recognition is prevented from being interrupted and the tempo is prevented. A good conversation can be realized.
That is, the user has a tendency to listen to the output voice related to the first sentence without speaking, and to utter the output voice related to the second sentence from the middle. Therefore, by starting the recognition of the user voice at the timing of outputting the output voice related to the second sentence, it is possible to prevent the user voice recognition from being interrupted and realize a conversation with a good tempo.

According to said structure, compared with the case where the timing of the user voice recognition start is controlled by the time from the output start of the output voice, the timing of the user voice recognition start is finely controlled according to the configuration of the dialogue. Therefore, it is possible to more reliably prevent the user voice recognition from being interrupted and realize a conversation with a good tempo.
Also, since the length of the utterance time is not used to control the user speech recognition start timing, it is difficult to calculate the length of the utterance time, or when it takes time to calculate the length of the utterance time. In addition, the user voice recognition start timing can be easily controlled.

  As described above, the dialogue generated by the conversation scenario execution unit 153 may be composed of two sentences of the first sentence and the second sentence, or may be composed of more sentences. Well, not particularly limited. In this case, the start of speech recognition in the speech recognition unit 59 may be any of the timings of speech output after the second sentence, and is not particularly limited.

Note that, as described above, the timing for starting speech recognition in the speech recognition unit 59 may be determined based only on the order of the sentences constituting the dialogue, or the speech recognition may be performed based on the content of each sentence. The start timing may be determined and is not particularly limited.
That is, the sentences constituting the dialogue include a sentence that is unlikely to be spoken by the user and a sentence that is likely to be spoken. Therefore, the conversation scenario execution unit 153 may perform control so as to start recognition of the user voice from the time when the output voice related to the sentence that is likely to be spoken by the user is started. Or you may control to start recognition of a user's voice from the time when a dialogue changes from a sentence with a low possibility of being spoken by a user to a sentence with a high possibility of being spoken. By doing so, it is possible to prevent the user voice recognition from being interrupted and to realize a conversation with a good tempo.
Here, sentences that are unlikely to be spoken by the user include sentences such as a call to the user and a repetition of instructions from the user. Examples of the sentence that is highly likely to be spoken by the user include a sentence that requests an instruction to the user.

The technical scope of the present invention is not limited to the above embodiment, and various modifications can be made without departing from the spirit of the present invention.
For example, in the above embodiment, the voice recognition device of the present invention has been described as applied to a life support robot. However, the present invention is not limited to a life support robot, and other instructions from a person can be input by voice. It can be applied to the equipment.

It is a front view explaining the structure of the robot which concerns on the 1st Embodiment of this invention. 2 is a left side view illustrating the configuration of the life support robot shown in FIG. It is a block diagram explaining the function of the speech recognition apparatus which concerns on this embodiment. It is a schematic diagram explaining the flow of the conversation between the speech recognition apparatus of FIG. 3 and a user. It is a schematic diagram explaining the flow of the conversation when the voice recognition apparatus of FIG. 3 performs guidance. It is a schematic diagram explaining the other example in the flow of the conversation between the speech recognition apparatus of FIG. 3 and a user. It is a block diagram explaining the function of the speech recognition apparatus which concerns on the 2nd Embodiment of this invention. It is a schematic diagram explaining the flow of the conversation between the speech recognition apparatus of FIG. 7 and a user.

Explanation of symbols

14 Microphone 18 Speaker 51, 151 Speech recognition device 53, 153 Conversation scenario execution unit (control unit)
55 Speech synthesis unit 57 AEC (output speech removal unit)
59 Speech recognition unit 61 Delay control unit (control unit)
63 Utterance time calculation unit 67 Storage unit

Claims (7)

A control unit that assembles dialogue lines;
A speech synthesizer that generates an output speech signal based on the assembled dialogue;
A speaker that outputs output sound based on the generated output sound signal;
A microphone that converts voice including at least user voice uttered by the user into an input signal;
An output sound removal unit that generates an input sound signal by removing a signal component related to the output sound from the input signal based on the output sound signal;
A voice recognition unit that recognizes the user voice based on an input voice signal and outputs a recognition result to the control unit;
The control unit controls the timing of starting recognition of the user voice by the voice recognition unit after a predetermined time from the start of output of the output voice and before the end of output of the output voice based on the dialogue. A voice recognition device characterized by the above. An utterance time calculation unit that calculates the length of the utterance time of the output sound related to the line based on the output sound signal;
The speech recognition apparatus according to claim 1, wherein the control unit controls the timing of starting recognition of the user speech by the speech recognition unit based on the length of the utterance time. A storage unit for storing in advance the length of the utterance time of the output voice related to the line;
The speech recognition according to claim 1, wherein the control unit controls the recognition start timing of the user speech by the speech recognition unit based on the length of the utterance time stored in the storage unit. apparatus. The control unit calculates a start time obtained by subtracting a predetermined delay time from the utterance time,
The speech recognition apparatus according to claim 2 or 3, wherein the speech recognition unit starts recognition of the user speech when the start time has elapsed from the start of output of the output speech. The control unit changes the length of the delay time,
The speech recognition apparatus according to claim 4, wherein timing for starting recognition of the user speech by the speech recognition unit is controlled.   The speech recognition apparatus according to claim 1, wherein the control unit controls the recognition start timing of the user speech in the speech recognition unit based on a sentence constituting the dialogue. A robot equipped with a voice recognition device for recognizing a user's voice,
A robot having a speech recognition device, wherein the speech recognition device is the speech recognition device according to any one of claims 1 to 7.

Images