JPS6250840B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a device that operates a load by recognizing a word that provides the minimum distance when the minimum distance between an input speech pattern and a standard speech pattern registered in advance is less than a reference value.

As a vehicle voice recognition device that controls vehicle loads such as power windows and wipers using voice recognition,
For example, the one shown in FIG. 1 has been considered.

In FIG. 1, reference numeral 1 denotes a command switch that is turned on when inputting a voice command. When the command switch 1 is operated, a voice input request signal is given to the voice recognition section 2, which turns it into an operating state, and then input from the microphone 3. performs recognition processing of the audio signal. That is, the voice recognition unit 2 analyzes the frequency of the input voice signal, converts it into a voice pattern made of predetermined digital data, and calculates the distance between it and a standard voice pattern corresponding to a plurality of pre-registered command words. ,
Output the word that gives the minimum distance as a recognition candidate. On the other hand, the minimum distance obtained by the speech recognition unit 2 is determined by the comparator 4 as the reset threshold TH from the setter 5.
and the minimum distance is the resiexion threshold TH
Depending on the comparison results in the following cases, the judgment unit 6 provides a recognition output of the word that provides the minimum distance to the control unit 7, and operates the load corresponding to the voice command.

On the other hand, when the minimum distance is found in the comparison section 4 to be above the reset threshold TH, the judgment section 6 outputs a repeat request to the voice recognition section 2 according to the comparison result, and prompts the operator to re-input the voice command. I try to encourage them.

By the way, in a speech recognition device like this, where the rejigger threshold that ultimately determines the output of the recognition result is fixed, it is possible to set a different rejigger threshold depending on the type of word depending on the control load. However, it is difficult to unambiguously determine a reset threshold that takes into account all vehicle usage conditions.

Therefore, for example, if a specific reset threshold value is fixed for a word that commands the opening/closing of power windows, if you approach a toll booth while driving, give a voice command to open the power windows, pay the toll, and then open/close the power windows. However, this command to close the power windows is given while the vehicle is accelerating by pressing the accelerator, and since the windows are also open, the noise level in the cabin is quite high due to the engine sound, and this noise Due to the contamination, the minimum distance calculated for the voice input commanding the power window to close is large, exceeding the fixed reset threshold, and the repeat request is repeated over and over again without being recognized, resulting in incorrect recognition. Even though candidates have been obtained, the reset threshold is too strict (the reset threshold for the command to close the window is set to be strict to prevent hands, etc. from getting caught due to misrecognition), so the voice command is repeated many times. There was a problem in that the load could not be operated unless it was emitted.

The present invention has been made by focusing on such a problem, and calculates the distance between the input voice pattern and each of a plurality of standard voice patterns registered in advance, and when the minimum distance is less than a reference value, the minimum distance is calculated. In order to improve the recognition rate of a vehicle speech recognition device that activates a load by recognizing a word that gives The purpose of this invention is to solve the above problem by loosening the reference value for comparison with the minimum distance when calculating the minimum distance using the correlated word as a recognition candidate after recognition of .

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 command switch that is operated when inputting a voice command, 2 is activated by a voice input request signal when the command switch 1 is turned on, and recognition recognizes the voice command signal input from the microphone 3. Department.

The voice recognition in the voice recognition unit 2 calculates the distance between the standard voice pattern and the input voice pattern for a plurality of pre-registered command words, and selects the word that gives the minimum distance from among the calculated distances. Output as a recognition candidate. 4 compares the minimum distance of the recognition candidate given by the speech recognition unit 2 with a rejiexion threshold Tj predetermined corresponding to the recognized word, and when it is equal to or less than the rejexion threshold, outputs a comparison output that outputs the recognition candidate as a recognition result; A comparison section 6 generates a comparison output that rejects the recognition candidate when the rejiexion threshold is exceeded, and 6 outputs the recognition candidate from the speech recognition section 2 to the control section 7 for recognition based on the comparison output of the comparison section 4 or performs comparison. This is a determination unit that requests the speech recognition unit 2 to repeat the output of the recognition candidate rejected by the unit 4.

Such a configuration is similar to the conventional device shown in FIG. A recognition correlation determining unit 10 that determines and outputs recognition of a voice command to close the power windows as a correlated word to a specific word commanding to open the power windows, for example, and a recognition correlation determining unit 10 that determines and outputs the recognition A switching circuit that switches and outputs the coefficient α of the coefficient setter 12 when the correlation of words is judged and output, and on the other hand, switches and outputs the coefficient of the coefficient setter 12 that becomes zero when the recognition correlation judgment unit 10 does not judge the correlation. 14, an adder 16 that adds the coefficient output from the switching circuit 14 and the resiexion threshold Tj from the threshold value setter 5 and outputs the result to the comparator 4 as a comparison reference; The recognition correlation determination unit 10 is provided with a memory 18 for temporarily storing and outputting recognition results corresponding to load control.

Note that the speech recognition section 2, comparison section 4, judgment section 6, recognition correlation judgment section 10, coefficient setter 12, switching circuit 14, adder 16, and memory 1 in the embodiment shown in FIG.
8 and the threshold value setting device 5 can be realized by program control of a microcomputer, and by externally connecting the command switch 1, the microphone 3, and the control section 7 to this microcomputer, the apparatus of the present invention is realized. be able to.

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

First, as a correlated word for a specific word to be set in the recognition correlation determination unit 10, for example, a power window opening command is set as a specific word, and a power window closing command that is expected to be input following this power window opening command is a correlated word. It is preset as . In this state, for example, if you turn on the command switch 1 and then input a voice command to open the power window into the microphone 3 as shown in block a of FIG. 3, the voice recognition will be performed as shown in block b. The distance between the pre-registered standard voice pattern for opening the power windows and the input voice pattern is calculated by unit 2. For example, as a distance calculation in the voice recognition unit 2, as a standard voice pattern registered in advance, (where j is an integer representing the type of command word such as 1, 2, etc.), whereas the input speech pattern is The distance Aj between the standard speech pattern Wj and the input speech pattern P is given by It is calculated by When the distance between the input speech data P and each of the plurality of standard speech patterns Wi (i=1, 2, . . . number of recognized words) is calculated in this way, the speech recognition unit 2 performs recognition that gives the minimum distance Aj. word
Wj is output as a recognition candidate. Speech recognition unit 2
The recognition candidate Wj outputted from the power window opening command, that is, the power window opening command
0, and the recognition correlation determination unit 10 determines whether the stored recognition result (Wold) of the previous voice input is a specific word, as shown in determination block c in FIG. Since the power window opening command in this case is not a specific word as a correlated word, the recognition correlation determining unit 10 proceeds to block d, outputs an L level output to the switching circuit 14, and outputs the coefficient from the coefficient setter 12 to 0 volts. is outputted to the adder 16, and the adder 16 adds the coefficient value zero given by the switching circuit 14 to the reset threshold Tj corresponding to the recognition candidate Wj given by the threshold value setter 5, and then adds the coefficient value zero given by the switching circuit 14 to the comparator 4. In this case, the reset threshold Tj from the threshold setter 5 is directly set in the comparison section 4. On the other hand, the comparison unit 4 is given the minimum distance Aj corresponding to the recognition candidate Wj by the speech recognition unit, and when the minimum distance Aj becomes less than or equal to the relocation threshold Tj in the judgment block e, the block f
As shown in FIG. 3, based on the comparison result of the comparison section 4, the judgment section 6 outputs the recognition candidate Wj giving the minimum distance Aj to the control section 7 as a recognition result, and at the same time stores the recognition result. The power window is controlled to open as shown in block g by driving the motor in response to the opening command. Although the explanation has been given here using a block diagram to make it easier to understand, in reality, the CPU can be operated by programming based on the flowchart shown in Fig. 3.

After opening the power window by such a voice command, the operator turns on the command switch 1 again,
When a power window closing command is issued to the microphone 3, the voice recognition unit 2 recognizes the input voice pattern of the power window closing command and the pre-registered standard voice as shown in blocks a and b in Figure 3. The distance to the pattern is calculated, and the voice recognition unit 2
outputs the registered word that gives the minimum distance A′j as a recognition candidate. In response to the output of this recognition candidate W′j, the recognition correlation determination unit 10 reads the previous recognition output Wold from the memory 18 and determines that the previous recognition output is a predetermined specific word since it is a command to open the power window. Then, the process proceeds from the discrimination block c to the discrimination block h, where the correlation with the currently recognized recognized word W'j is determined. Since the current recognition candidate W'j is a power window closing command, it is determined that it is a correlated word W'j with respect to a preset specific word Wj, and an H level output is generated to the switching circuit 14.

Therefore, the switching circuit 14 switches to output the coefficient α in the coefficient setter 12, and the adder 1
In step 6, a coefficient α is added to the relocation threshold T'j corresponding to the recognition candidate W'j set by the threshold setting device 5, and the recognition candidate is set as shown in block i.
The reset threshold value T'j for W'j is changed to T'j+α and output to the comparator 4.

In other words, the resiexion threshold value obtained by adding the coefficient α to T′j increases the value of the resiexion threshold value that allows the recognition output and relaxes the comparison reference value, and as a result, when the power window closing command is issued, the vehicle Even if the cabin noise level is high due to the engine sound due to acceleration, and this noise level is mixed into the voice command and the minimum distance A'j calculated by the voice recognition unit 2 has a value larger than that in the steady state, the comparison Since the relocation threshold T'j set in section 4 is relaxed, the relocation threshold T'j (=
Based on the comparison result of the comparison section 4, the judgment section 6 outputs the recognition candidate W'j corresponding to the power window closing command to the control section 7 as a recognition result, and Close the power window as shown.

Note that if the minimum distance Aj or A'j from the speech recognition section 2 in the comparison section 4 exceeds the resiexion threshold Tj or T'j from the adder 16, the comparison section at this time Judgment unit 6 based on the comparison result of 4.
As shown in block j, a repeat request is output to the voice recognition unit, prompting the voice command to be input again.

In this way, when the input of a correlated word corresponding to a predetermined recognized word is recognized, the reset threshold for determining the similarity of the correlated word is relaxed, so for example, a toll gate, etc. If you first open the power windows with a voice command when passing, and then give a voice command to close the power windows while the engine is accelerating after paying the toll, the minimum distance of the recognition candidate for the command to close the power windows. Since the reset threshold that determines the distance is relaxed, engine noise during acceleration mixes noise with the voice command, and even if the minimum distance calculated by voice recognition becomes large, recognition candidates are rejected and a repeat request is made. It is possible to output recognition candidates as recognition results without noise, and to ensure that the power windows are closed under conditions of high noise levels. Note that a one-touch type power window is suitable as the power window.

In addition, although the above embodiment takes the opening and closing of power windows as an example, the present invention is not limited to this, and appropriate correlated words that are expected to be input after the load is activated by specific word recognition. Similarly, the reset threshold may be relaxed.

As explained above, according to the present invention, the distance between an input speech pattern and each of a plurality of pre-registered standard speech patterns is calculated, and when the minimum distance is less than a reference value, a word giving the minimum distance is calculated. In a vehicle voice recognition device that recognizes and activates a load and requests re-input of a voice command when the minimum distance exceeds a reference value, an input is expected after the load is activated by recognizing a specific word. Correlation words are determined in advance, and when the minimum distance between the recognition word of this specific word and the correlation word as a recognition candidate is calculated, the reset threshold for comparison with this minimum distance is relaxed, so that the correlation word corresponding to the specific word is Regarding voice commands, it is possible to reliably recognize voice commands even in situations where the noise level is high and the degree of similarity to the standard pattern is low, and repeat requests are repeated many times due to rejection of recognition candidates. Since there is no voice command, operations using voice commands are easy to perform, and the effect is that the voice commands can be reliably recognized even in environments with high noise levels, and the recognition rate can be greatly improved.

[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...Command switch, 2...Voice recognition unit, 3...
... Microphone, 4 ... Comparison section, 5 ... Threshold value setting section, 10 ... Recognition correlation judgment section, 12 ... Coefficient setting device, 14 ... Switching circuit, 16 ... Adder, 18
……memory.

Claims (1)

[Claims] 1. Calculates the distance between the input voice pattern of the voice command and each of the standard voice patterns of a plurality of pre-registered command words, and when the minimum distance among the calculated distances is less than a reference value, In a vehicle voice recognition device that recognizes the above-mentioned voice command as a word that gives the minimum distance and outputs it, and operates a load according to the recognition output, the correlation that is expected to be input after the load is activated by recognizing a specific word. A reference value changing means is provided for predetermining words, and changing a reference value to be compared with the minimum distance to a larger value by a predetermined value when calculating a minimum distance using the correlated word as a recognition candidate after recognizing the specific word. A vehicle voice recognition device characterized by: