JPH036516B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION When the maximum similarity between the voice pattern of a voice command input from a microphone and a pre-registered standard voice pattern is greater than or equal to a comparison reference value, the voice command is determined as a word giving the maximum similarity. The present invention relates to a speech recognition device that performs recognition and output.

2. Description of the Related Art An example of a voice recognition device for an automobile that controls opening and closing of power windows, operation of a car radio, etc. using voice commands is as shown in FIG. 1.

In FIG. 1, reference numeral 2 denotes a command switch that is operated when inputting a voice command. When the command switch 2 is turned on, the voice recognition section 3 can receive a voice signal from the microphone 1. In this state, when the operator issues a predetermined voice command to the microphone 1, the voice recognition unit 3 generates a plurality of standard voice patterns and the voice command input from the microphone 1 using the operator's voice registered in advance. The degree of similarity with the voice pattern is calculated, and the value of the maximum degree of similarity is output to the comparison section 4. A predetermined rejiexion threshold TH is set in the comparator 4 as a comparison reference value by a setter 7, and when the maximum similarity from the speech recognition unit 3 is greater than or equal to the rejiexion threshold TH, the comparison result is output to the discriminator 5. The voice command is determined to be the word that gives the maximum similarity, the recognition output is given to the control unit 6, and the on-vehicle load corresponding to the recognition output is controlled.

On the other hand, if the maximum similarity in the comparison unit 4 is lower than the rejiexion threshold TH, it is assumed that erroneous recognition has occurred, and the discrimination unit 5 instructs the voice recognition unit 3 to repeat the request, and the operator receives a voice command. request re-input.

For this reason, the operator's pronunciation may be unclear,
Even if a noise such as an engine sound is input at the same time as a voice command and misrecognition occurs, the maximum similarity will not exceed the reset threshold for misrecognition.
Therefore, the on-vehicle load can be controlled only when correct voice recognition is performed.

However, like this speech recognition device, the resiexion threshold that determines the maximum similarity
When TH is a constant value, the maximum similarity is uniquely determined regardless of the type of control load.For example, if a voice command to open the power windows is misrecognized and the car radio is operated. However, if the power windows are opened due to a misrecognition of a voice command to operate a car radio, for example, when driving at high speed, the driver may be hit by a strong wind and the effect of misrecognition may be small. big.

In addition, when the noise is high due to engine sound, etc., the calculated maximum similarity value is small due to the inclusion of noise components, and in this case, a repeat request is made to prevent erroneous recognition, but the repeat request is On the other hand, even if the voice command is issued again, in a similarly high noise state, the maximum similarity does not exceed the relocation threshold TH, and the repeat request is repeated, even though the recognized word that gives the maximum similarity is correct. There is a problem that a state in which no recognition output is performed continues.

The present invention has been made by paying attention to such problems, and calculates the degree of similarity between the voice pattern of an input voice command from a microphone and each of a plurality of standard voice patterns stored in advance, and When the maximum similarity is greater than the comparison reference value, the above voice command is recognized as the command word giving the maximum similarity and output, and when the maximum similarity is less than the comparison reference value, the voice command is requested to be re-input. In order to improve the recognition rate of a voice recognition device for a vehicle, when re-inputting a voice command is requested, the above problem is solved by lowering the comparison reference value of the command word corresponding to the maximum similarity. The purpose is to

Hereinafter, the present invention will be explained based on the drawings.

FIG. 2 is a block diagram showing one embodiment of the present invention.

First, to explain the configuration, 1 is a microphone that converts a voice command from the operator into a voice command signal, 2 is a command switch (push button switch) that is turned on in response to a voice command, and 3 is a plurality of pre-registered voices. a voice recognition unit that calculates the degree of similarity between the standard voice pattern of the command word and the voice pattern of the voice command signal inputted from the microphone 1, and outputs the maximum degree of similarity and the word corresponding to this maximum degree of similarity as a recognition candidate; 5 is a comparison unit that compares the maximum similarity given by the speech recognition unit 3 with a rejiexion threshold that will be explained later; and 5 is a comparison unit that performs speech recognition when the maximum similarity is greater than or equal to the relocation threshold based on the comparison result of the comparison unit 4. The word giving the maximum similarity by the unit 3 is given to the control unit 6 as a voice command recognition output, while the comparison result of the comparison unit 4 when the maximum similarity is below the relocation threshold is sent to the voice recognition unit. This is a determination unit that outputs a repeat request for the 3.

The configuration of such a speech recognition device is the same as that of the conventional device, but in addition, in the present invention, a reset threshold THi (however, ,
i = 1, 2, ..., P, number of recognized words) were stored.
Resiexion threshold storage unit 10 using ROM etc.
is provided, and this reset threshold storage unit 10
The threshold value THi stored in is determined as a value according to the type of voice command word.

For example, the reset threshold for a voice command word that commands the opening and closing of power windows is set higher than the reset threshold for a voice command word that commands the operation of a car radio. The standards for comparing the similarity of speech recognition have been tightened. Reference numeral 12 denotes a buffer memory that temporarily stores the reset threshold THi stored in the reset threshold storage unit 10, and when a recognition request signal is output by turning on the command switch 2, each threshold data is stored from the reset threshold storage unit 10. Receive a transfer. Furthermore, when a recognition candidate corresponding to the maximum similarity is output from the speech recognition unit 3 after the transfer of the threshold value data, the buffer memory 12 outputs the rejiggering threshold corresponding to the word of this recognition candidate to the comparison unit 4. Furthermore, when the discriminating unit 5 outputs a repeat request, 14 receives from the buffer memory 12 the rejiexion threshold of the word corresponding to the maximum similarity that caused the repeat request to be made, and sets a predetermined value α to the rejiggering threshold corresponding to the recognition candidate. Subtract it from the threshold THk and use it as (THk-α) to calculate the buffer memory 1.
This is a subtraction section that returns the number to 2.

Note that although the speech recognition circuit section in the above embodiment is shown in a function block diagram separated by each function, the actual device includes a speech recognition section 3, a comparison section 4, a discrimination section 5, Each of the reset threshold storage section 10, buffer memory 12, and subtraction section 14 can be implemented by a microcomputer, and the microcomputer 1, command switch 2, repeat display, etc. can be externally connected to this microcomputer.

Next, the operation of the present invention will be explained with reference to the operation flow shown in FIG.

When controlling an on-vehicle load using a voice command, the command switch 2 is first turned on. By turning on the command switch 2, a recognition request signal is output to each of the voice recognition unit 3 and the buffer memory 12, and the voice recognition unit 3 enters an operating state in which it can recognize and process the voice signal from the microphone 1.
As shown in block a of FIG. 3, the relocation threshold THi stored in the relocation threshold storage section 10 is transferred to the buffer memory 12. Next, if the operator issues a predetermined voice command to the microphone 1, the voice recognition unit 3 receives the voice input shown in block b in FIG.
As shown in block d, the speech recognition process is executed, the input speech signal is converted into a speech pattern, and the similarity Si with the standard speech pattern of the voice command word registered in advance is calculated as shown in block d. The maximum similarity Sk out of Si is output to the comparison unit 4. (Block e) Also, the speech recognition unit 3 outputs the recognition candidate K corresponding to the maximum similarity Sk to the discrimination unit 5 and the buffer memory 12, respectively.
is the resiexion threshold corresponding to recognition candidate K
THk is outputted to the comparator 4 as a comparison reference value.

Next, as shown in the discrimination block f, the maximum similarity
Sk and the resiexion threshold THk are compared and judged, and the maximum similarity Sk is the resiexion threshold THk.
In the above case, based on the comparison result of the comparison unit 4, the discrimination unit 5 outputs the recognition candidate K giving the maximum similarity Sk to the control unit 6 as the recognition word of the voice command, as shown in block g of FIG. Operates the load according to the command.

On the other hand, if the maximum similarity Sk in the comparing section 4 is below the resiexion threshold THk,
Depending on the comparison result of the comparison section 4, the determination section 5 outputs a repeat request to the speech recognition section 3 and the subtraction section 14.

Upon receiving this repeat request, the subtraction unit 14
As shown in block h of the figure, the relocation threshold THk output to the comparator 4 at this time is read, and
By subtracting a predetermined value α,
It is returned to the buffer memory 12 again as (THk-α), and the relocation threshold THk that caused the repeat request to be made is changed to a smaller value by a predetermined value α. Next, as shown in block i, in response to a repeat request to the voice recognition section 3, a repeat display is displayed on a repeat display (not shown) to prompt the operator to re-input the voice command.

If the operator issues a voice command to microphone 1 again after receiving this repeat display, then
As shown in blocks a to e of FIG. 3, the maximum similarity of the input speech patterns is determined again, and in the discrimination block f, the relocation threshold is lowered by a predetermined value α.
A comparison is made with THk, and if the value of the similarity Sk at this time is equal to the value of the first similarity, then the rejiexion threshold THk is lower by α than the first, so there is extremely large noise. Unless the maximum similarity value is lowered due to the addition of
The recognition candidate K from the speech recognition unit 3 that is equal to or higher than THk and gives the maximum similarity Sk is given to the control unit 6 from the discrimination unit 5 as a recognition output of the voice command, and the load corresponding to the voice command is operated.

On the other hand, if the second voice command in response to the repeat request is different from the first voice command, the rejiexion threshold THj corresponding to the second different voice command must not have been modified at all during the first voice recognition. The second maximum similarity Sj is not affected by the change in the first resiexion threshold TKk.

Incidentally, when inputting a voice command in response to a repeat request, there is no need to turn on the command switch 2 again. Furthermore, when the command switch 2 is turned on again after changing the reset threshold in response to a repeat request, the reset threshold changed by the subtractor 14 returns to its original value. Furthermore, in the embodiment described above, the value of α in the subtraction section is kept constant, but in another embodiment, the value of α (change rate) for changing the reset threshold depending on the recognized word may be set separately. Also good. As described above, it is desirable to set the value of α to be different for each reset threshold, and to reduce the value of α for those that are highly influenced by misrecognition.

As explained above, according to the present invention, the configuration is calculated by calculating the degree of similarity between the voice pattern of the input voice command from the microphone and each of a plurality of standard voice patterns registered in advance. When the similarity is above the comparison standard value, the above voice command is recognized as the command word that gives the maximum similarity and is output, and when the maximum similarity is below the comparison standard value, the voice command is requested to be re-input. In a vehicle voice recognition system, when a voice command is requested to be re-inputted, the comparison standard value of the command word corresponding to the maximum similarity is lowered to loosen similarity discrimination, so the noise level such as engine sound is lowered. Even if a voice command is issued in a state where the maximum similarity is lower than the comparison reference value and a repeat request is made, the comparison reference value will be lowered when the same voice command is input again in response to the repeat request. This allows the voice command to be reliably recognized and the load to be activated even under high noise levels, and even if the correct voice command is issued, the repeat request is repeated and the load is activated due to the voice command. It is possible to reliably prevent the continuation of an uncontrolled state. In addition, the maximum similarity rejiexion threshold for identifying recognition candidates is set separately depending on the type of voice command word, that is, the type of load to be controlled by the voice command. By setting a high reset threshold for power window voice commands, you can reliably prevent voice commands to operate the car radio from being misrecognized and activating the power windows. Even if erroneous recognition occurs, the effect is that a load that has a large impact on driving (large resiexion threshold) will not be driven due to incorrect recognition of a voice command for a load that has a small effect on driving (small resiexion threshold). It will be done.

[Brief explanation of the drawing]

Fig. 1 is a block diagram showing an example of prior art by the applicant, Fig. 2 is a block diagram showing an embodiment of the present invention, and Fig. 3 is an operation flow diagram showing the operation of the present invention. . 1...Microphone, 2...Command switch, 3...
Speech recognition section, 4... Comparison section, 5... Discrimination section, 6... Control section, 10... Resiexion threshold storage section, 12... Buffer memory, 14... Subtraction section.

Claims (1)

[Claims] 1. When the degree of similarity between the utterance pattern of the voice command signal input from the microphone and each of a plurality of pre-registered standard voice patterns is calculated, and the maximum degree of similarity obtained by the calculation is greater than or equal to a comparison reference value. In a vehicle voice recognition device that recognizes and outputs the voice command as a word that gives the maximum similarity, and requests re-input of the voice command when the maximum similarity falls below a comparison reference value, the voice command When re-input is requested, only the comparison reference value corresponding to the word that gives the maximum similarity is changed to a lower value by a predetermined value among the comparison reference values predetermined for each of a plurality of voice command words, and A voice recognition device for a vehicle, characterized in that a reference value changing means is provided that does not change the comparison reference value corresponding to words other than words.