JP2989195B2 - Voice registration method - Google Patents

Abstract

PURPOSE:To control a device producing a noise with a high precision by voice recognition by registering a voice for stop with the device operated and registering a voice for start of the operation with the device stopped. CONSTITUTION:When a user turns on a switch 1 of an air conditioner after speaking 'air conditioner on' to a microphone 13, correspondence between a pattern of 'air conditioner on' held in a memory and the signal of the switch to actually start the air conditioner is stored in the storage part of a voice recognizing device. When 'air conditioner off' will be registered, he starts the operation of the air conditioner and turns off the switch after speaking 'air conditioner off'. That is, the instruction for start is registered with the air conditioner stopped, and the instruction for stop is registered with the air conditioner operated. Thus, the recognition precision is improved when operation states of devices producing different noises are changed with a voice.

Description

Description: TECHNICAL FIELD The present invention is suitably applied to a voice registration system, more specifically, to voice recognition used for moving a noise emitting device by voice.

2. Description of the Related Art There are two types of speech recognition devices: a specific speaker speech recognition device that is used after a user has registered a speech in advance, and an unspecified speaker speech recognition device that does not limit the user. Is common. In particular, speech recognition poses a problem in situations where there is a lot of noise.Since the characteristics of the registered voice and the input voice are similar, the specific speaker method recognizes the voice better. Therefore, the recognition accuracy is good. However, voice recognition in noise has a drastic reduction in recognition accuracy compared to when there is no noise, and various measures have been considered. For example, there is a type in which the environmental noise is stored before a voice is input, and when the voice is input, the stored noise signal is subtracted before this signal (Ozeki: N
HK Giken Monthly Report 4p410 (Showa 36)). In this method, the voice input in the noise is corrected to the same state as the standard pattern registered when there is no noise, and then collation for recognition is performed. In addition, there is an attempt to process noise by calculation to make it noiseless (Tsuboi, Takebayashi: IEEJ 10
8, 10 p755 (Showa 63)), both of which require noise reduction together with voice input and extra calculations. As a result, there is a drawback that the time required for recognition becomes longer. On the other hand, on the other hand, there is also an attempt to add noise to the standard pattern for speech recognition to make it closer to the situation of the speech input at the time of noise. This is to register a voice in a state where there is environmental noise.

Consider a case where a target device is operated by voice.
When noise is produced by the operation of the device, the conventional method keeps the device operating and registers the sound under the environment.

(1) However, in practice, this is good when the recognition is performed while the device is moving, but at least the command for operating the device is uttered while the device is stopped. Therefore, the conventional method has a disadvantage that the recognition accuracy is deteriorated.

(2) In addition, when the voice recognition device does not have any function as a keyboard, there is a problem that it is not possible to notify the registered voice of a command to what kind of device.

(3) In addition, when there are a plurality of devices to be operated and the types of noise emitted from the devices are different, it is troublesome to operate each of them at the time of voice registration.

(4) Further, the input level of the microphone for voice extraction is adapted gradually according to the noise level of the environment, and is set to be sensitive to a sudden change in level such as voice. Therefore, if sudden noise is generated when operating individual devices, the sudden noise will cut out the voice section and capture the wrong signal into the device, and as a result, data that prevents recognition will be registered. There is a disadvantage that it is done.

Object The present invention has been made in view of the above situation, and has been made mainly for the purpose of accurately controlling a device that emits noise by voice recognition, and further has the above problem. The purpose of the present invention is to make it possible to instruct the purpose of the input voice without the input means such as a keyboard in the voice recognition device after the consideration.

Configuration In order to achieve the above object, the present invention is a method for registering a voice of a voice recognition device for starting or stopping the operation of a device by the voice recognition device, wherein (1) the voice is uttered and then targeted. The voice registered by changing the state of the switch of the device is registered in association with the change of the state of the switch, or (2) the voice input by changing the state of the switch of the target device is Registering the state in association with a signal that causes a change opposite to the change, or (3) sound generated by the environment;
When recording on a medium for each device by superimposing the sound generated by a device that changes the operation due to the uttered voice and the sound generated by the device, it gradually changes to the sound generated in a stable operation state when switching from device to another device When the sound is registered in such a manner that the reproduced sound from the recorded medium becomes stable, the sound is registered by an instruction to prompt the utterance. Hereinafter, a description will be given based on examples of the present invention.

The present invention relates to a method for registering a voice for stopping a target device in operation while registering a voice of the voice recognition device for starting or stopping the operation of the device by the voice recognition device, In this case, a voice for starting the operation in a state where the air conditioner is stopped is registered. As an example, the operation of the air conditioner in a car will be described. In a normal voice recognition device, an instruction is given to register a voice so as to produce a voice while rotating a fan. However, the fan rotates when the air conditioner is operating, and the fan does not rotate when an instruction for activating the air conditioner is required. For this reason, it is clear that registering all sounds while the fan is rotating as in the usual method has an effect when stopping the rotation, but has an opposite effect when starting the rotation. Therefore, the command for starting is registered when the air conditioner is stopped, and the command for stopping it is registered while the air conditioner is operating, so that devices that emit different noises are voiced. In the case where the operation state is changed, the recognition accuracy is improved.

A first aspect of the present invention is a voice registration method for a voice recognition device for starting or stopping operation of a device by the voice recognition device, wherein a state of a switch of a target device is changed after the voice is uttered. The registered voice is registered in association with the change in the state of the switch.
As an example, a case where an air conditioner is operated in an automobile will be described.

FIG. 1 is a diagram for explaining an example of a case where an air conditioner is operated in an automobile. In the figure, 1, 2,... Switch for performing, 10 is a device switch status detecting unit for detecting which device switch is being operated, 11 is a correspondence table of device switches and word numbers, 12 is a control unit, 13 is a microphone, and 14 is a feature An extraction unit, 15 is a voice section detection unit, 16 is a dictionary pattern creation unit, 17 is a dictionary store unit, 18 is a dictionary. First, we say “air-con, on” to the microphone 13 and then switch on the air conditioning. Turn 1 on. Then, the “air-con, on” pattern held in the memory is associated with the signal of the switch that actually starts the air conditioning, and stored in the storage unit of the voice recognition device. According to this method, in order to register "air-con, off", the air-conditioning must be operated in advance, and then the switch must be turned off after "air-con, off" is spoken. Therefore, the voice is registered after generating the noise generated when "air-con, off" is uttered in practical use, so that the recognition rate in practical use is improved and the meaning of the registered voice is recognized without a keyboard. You can teach the device.

Also, if the switch is turned on and off after the utterance, an electrical noise may be generated and the registered voice may have to be uttered again. Therefore, if the order is reversed, in the voice registration method of the voice recognition device for starting or stopping the operation of the device by the voice recognition device, the state of the switch of the target device is changed and registered. It can be registered in association with a nearby change in the state of the switch. In this case, for example, the air conditioner needs to be operating when it is desired to register the turning off of the air conditioner switch. Therefore, as shown in FIG. 2, a final state storage unit 19 of the switch is provided, and even if the air conditioning is stopped, the switch is turned on once and then turned off, and then "switch off" is uttered. The air-conditioning switch-off, which is the operation closest to the "switch-off" voice, and this voice recognition pattern are stored in the storage unit of the voice recognition device in association with each other.

According to a second aspect of the present invention, in the voice registration method of the voice recognition device for starting or stopping the operation of the device by the voice recognition device, the state of the switch of the target device is changed and the input is performed. It is also possible to register the changed voice in association with a signal that causes the state of the switch to change in the opposite direction. In this case, if the same as in the previous example is explained, the air conditioner is turned on and "switch off" is uttered. At this time, it is stored in association with a command to turn off the air conditioning which is the operation performed immediately before this.

According to this method, when the switch is turned on, the device is in an operating state, that is, a state in which noise due to the operation of the device is uttered.
Therefore, the voice is registered in a noise environment where the voice must be uttered in practical use. For this purpose, it is necessary to have an instruction reverse to the operation that has occurred, and it is necessary to associate this with the instruction opposite to the operation.

Therefore, according to the first or second aspect of the present invention, when an actual voice recognition device is always used for registering a voice after uttering a voice and then switching on the operation or switching on and off the operation. Since the same noise environment is created and the voice is registered, it is possible to accurately control the noise emitting device by voice recognition. Further, since the voice is registered in association with the operation of the switch of the target device, the purpose of the input voice can be instructed to the voice recognition device without an input means such as a keyboard.

Further, the invention according to claim 3 uses a sound reproducing medium that is used together when registering a voice of the voice recognition device for starting or stopping the operation of the device by the voice recognition device. The sound generated by the operation of the device that changes the operation by the voice is also recorded, and the voice is registered while reproducing the storage medium at the time of the voice registration of the voice recognition device. Is shown in FIG. 3 and will be described below.

A cassette tape 20 is used as a recording medium, and the noise of the device is recorded in the tape 20. Here, an example in which a device in a car is operated by voice will be described. In the case of a car, it is dangerous for the driver to register the voice while driving, so the voice is registered when the vehicle stops. Third
In the case of the figure, the cassette tape 20 records an engine sound, a sound obtained by adding a sound of an air conditioning fan to the engine sound, and the like. This cassette tape is inserted into a cassette player of a car radio. First, an engine sound of a car is reproduced, and a command necessary for signaling is uttered a required number of times to register a voice. Next, the sound of the fan plus the sound of the engine is reproduced. This is for operating the air conditioner in the vehicle, and the sound of the fan of the air conditioner is added.
When the sound of the fan is added, an air conditioning control command, for example, "strong" or "weak" is registered. In the voice recognition data thus registered, noise superimposed on voice uttered when using the recognition device,
Since the information is registered in advance, the recognition accuracy in actual use can be improved.

However, when the noise recorded on the tape changes suddenly, the acoustic characteristics change suddenly, so that the recognition device may mistakenly input the voice as data. For this purpose, as shown in FIG. 4, the level of the noise recorded on the tape is gradually changed. In a voice detection method of an ordinary voice recognition device, a noise level of a portion where voice is not uttered is measured, and an operation is performed while setting a voice cut-out threshold value to such an extent that noise is not determined as a voice signal. Thus, there is no adaptability to sudden noise changes. Therefore, by gradually changing the recording level and gradually following the threshold value, an erroneous audio signal is prevented from being captured.

Although the speech recognition method is not described here, this registration method does not depend on the recognition method, and therefore any recognition method can be used. Also,
In the above embodiment, the start point of the noise is a signal of utterance, but it is more preferable to record words to be uttered in a tape on a tape. However, care must be taken at this time if recording with too loud voice will hinder registration.

Therefore, according to the invention shown in FIG. 3, when various devices are operated by voice in noise, the noise uttered by the operation is recorded, so that registration with higher recognition accuracy than the conventional method is performed. It becomes possible.

Effects As is clear from the above description, according to the present invention, when the operation state of a device that emits different noises is changed by voice, recognition accuracy is improved. In addition, to register the voice to switch on or turn off the operation after uttering the voice, always create the same noise environment as when using an actual voice recognition device and register the voice, and generate noise Control by voice recognition of equipment can be performed with high accuracy. Further, since the voice is registered in association with the operation of the switch of the target device, the purpose of the input voice can be instructed to the voice recognition device without an input means such as a keyboard.

Furthermore, when various devices are operated by voice in the noise, since the noise uttered by the operation is recorded, there is an advantage that registration with higher recognition accuracy than the conventional method can be performed. .

[Brief description of the drawings]

FIGS. 1 and 2 are configuration diagrams for explaining an embodiment of the present invention, respectively, and FIGS. 3 and 4 are diagrams for explaining another embodiment of the present invention. 1 to n switches 10 device switch state detectors
11: Device switch word number correspondence table, 12: Control unit, 13: Microphone, 14: Feature extraction unit, 15:
Voice section detection section, 16: Dictionary pattern creation section, 17: Dictionary storage section, 18: Dictionary, 19: Final state storage section, 20 ...
…cassette tape.

Continuation of the front page (56) References JP-A-2-32105 (JP, A) JP-A-2-305047 (JP, A) JP-A-62-91999 (JP, A) JP-A-59-132000 (JP, A) JP-A-62-65088 (JP, A) JP-A-59-122822 (JP, A) JP-A-2-64600 (JP, A) JP-A-2-83596 (JP, A) 2-244096 (JP, A) JP-A-2-154299 (JP, A) JP-A 63-149000 (JP, U) Utility model registration 2559790 (JP, Y2) JP-B 61-30939 (JP, B2) JP-B-63-67197 (JP, B2) JP-B-4-30040 (JP, B2) JP-B-4-49715 (JP, B2) (58) Fields investigated (Int. Cl. ⁶ , DB name) G10L 3/00 521

Claims (3)

(57) [Claims] 1. A method of registering a voice of a voice recognition device for starting or stopping an operation of a device by the voice recognition device, wherein the voice is uttered and then registered by changing a state of a switch of a target device. A voice registration method, wherein voice is registered in association with a change in the state of the switch. 2. A method for registering a voice of a voice recognition device for starting or stopping the operation of a device by the voice recognition device, wherein the voice input by changing the state of a switch of a target device is used. A voice registration method characterized in that a state is registered in association with a signal that causes a change opposite to the change. 3. A voice registration method for a voice recognition device for starting or stopping an operation of a device by the voice recognition device, wherein the device whose operation is changed by a sound generated by an environment and a uttered voice is used. When recording on a medium for each device by superimposing the sound that is generated, when switching from one device to another, record it so that it gradually changes to sound that is generated in a stable operation state, and when registering sound, A sound registration method characterized by registering a sound in response to an instruction to prompt a utterance when a reproduced sound from a recorded medium becomes stable.