JPH0235988B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a voice input control device, and more particularly to an improvement in a voice input control device that controls controlled equipment such as in-vehicle equipment in accordance with input voice.

FIG. 1 is a schematic block diagram of a voice input control device which is the background of the present invention. First, a conventional voice input control device will be briefly explained with reference to FIG. Voice is input to the microphone 1 and converted into a voice signal, and this voice signal is sent to the voice recognition circuit 10.
It is input to 0. In the voice recognition circuit 100, predetermined voice keywords for selecting the functions of each controlled device and voice characteristic parameters corresponding to device names are registered. Then, the voice recognition circuit 100 analyzes the waveform of the voice input from the microphone 1, extracts characteristic parameters, compares the characteristic parameters with the characteristic parameters registered in advance, and identifies the corresponding keyword. . The control circuit 200 generates a control signal corresponding to the keyword as a recognition result of the speech recognition circuit 100, and supplies it to the corresponding drive circuit 300. Accordingly, the drive circuit 300 controls the controlled device 400.

FIG. 2 is a schematic block diagram of the speech recognition circuit 100 shown in FIG. Next, the speech recognition circuit 100 will be explained with reference to FIG. A voice signal from the microphone 1 shown in FIG. 1 is given to a voice analysis circuit 101. This voice analysis circuit 101
is composed of a plurality of bandpass filters, smoothing circuits, A-D converters, etc., and analyzes the waveform of an input audio signal to extract characteristic parameters. The feature parameters extracted by this speech analysis circuit 101 are stored in the feature parameter register 10.
2. Registered parameter memory 104
stores characteristic parameters serving as keywords for all voices required to control the controlled device 400. Similarity calculation circuit 103
is to calculate the degree of similarity between the input voice feature parameter stored in the feature parameter register 102 and any parameter registered in the registered parameter memory 104. The determination circuit 105 determines whether the degree of similarity calculated by the degree of similarity calculation circuit 103 is as high as possible, and compares the degree of similarity at that time with a predetermined threshold Th. This is to identify which registered pattern the recorded voice corresponds to. Then, the determination circuit 105 encodes the keyword number as the identification result and outputs it as numerical data. Note that the timing signal generation circuit 106 generates a timing signal for determining the procedure of the series of recognition processing described above, and transmits this timing signal to the speech analysis circuit 101, registered parameter memory 104, and determination circuit 105. give.

By the way, the speech recognition circuit 100 configured as described above has the ability to sufficiently recognize input speech when used in a low-noise environment. However, in places where the engine noise is loud, such as inside a car, the recognition rate of input speech decreases, and the noise itself is often mistakenly recognized as keyword speech. For this purpose, such a speech recognition circuit 1
When 00 was used in a voice input control device, sufficient performance could not be obtained. In particular, in environments where a radio is playing or people are talking, some of the waveforms of these sound sources may be similar to the sounds of the registered keywords, and the probability of misrecognition is particularly high. There was a problem.

Therefore, the main object of the present invention is to provide a voice input control device that can reduce erroneous recognition as much as possible and prevent malfunctions of controlled equipment even in a relatively noisy environment. It is.

To summarize the invention, a predetermined voice for controlling a controlled device is emitted twice, and the characteristic parameters of the first input voice and the characteristic parameters of the second input voice are made to be the same. If it is, the controlled device is controlled in response to the input voice, and even if the characteristic parameters of the first input voice and the next input voice do not match, the controlled device is controlled according to the input voice. If the voice is of the first type, the device to be controlled is configured to control the controlled device in response to the voice.

The above objects and other objects and features of the present invention will become more apparent from the detailed description given below with reference to the drawings.

FIG. 3 is a schematic block diagram of a device for recognizing twice-pronounced speech, which is the premise of the present invention. In general, when humans pronounce the same word consecutively, the waveforms of those sounds have extremely high similarities.
If a person using this device recognizes each voice produced with the intention of controlling the device, the probability that the recognition results will be inconsistent is extremely small. Focusing on this point, in this embodiment, a new recognition result register 110, an old recognition result register 111, and a comparison circuit 112 are newly provided in the voice input control device shown in FIGS. 1 and 2. . The new recognition result register 110 stores the recognition result of the second voice pronounced by the speech recognition circuit 100, and the old recognition result register 111 stores the recognition result of the first voice pronounced by the speech recognition circuit 100. Remember. The comparison circuit 112 then recognizes the new recognition result register 11.
It is determined whether the recognition result of the voice pronounced the second time as the content of 0 matches the recognition result of the voice pronounced the first time stored in the old recognition result register 111. When the comparison circuit 112 determines that the two match, it activates the control circuit 200.

Next, the operation will be explained. The speaker produces twice the necessary sound to turn on the power of, for example, the radio 401 as the controlled device. When the first voice is produced, the microphone 1 provides a voice signal corresponding to the voice to the voice recognition circuit 100. As explained in FIG. 2 above, the speech recognition circuit 100 analyzes the waveform of the speech, extracts the feature parameters, determines the degree of similarity with the feature parameters of multiple types of speech registered in advance, and extracts the feature parameters. Perform voice recognition. Then, the recognition result of the first voice pronounced is stored in the new recognition result register 110. When the speaker continues to pronounce the same sound,
The speech recognition circuit 100 recognizes the speech. At this time, the recognition result of the first voice stored in the new recognition result register 110 is stored in the old recognition result register 110.
will be forwarded to. Then, the new recognition result register 11
0 stores the recognition result of the second voice produced. Comparison circuit 112 recognizes new result register 1
The contents of 10 and the contents of the old recognition result register 111 are compared, and if they do not match, the control circuit 200 is disabled, and if they match, the control circuit 200 is enabled. Therefore, the control circuit 200 generates a control signal and supplies it to the drive circuit 300 when the recognition results of the twice-pronounced voices match each other. In response, drive circuit 300 powers on radio 401.

In addition, in order to control the radio 401, approximately 20 words are required as voices for turning on the power, selecting the frequency, switching the volume, etc., but the voice recognition circuit 1
The recognition result coded with 00 is approximately 5 bits of information. Therefore, the recognition new result register 110
The old recognition result register 111 can be constructed relatively easily by using some flip-flops, and the comparator 112 can be constructed relatively easily by using some exclusive OR and AND elements.

In this way, in the embodiment shown in FIG. 3, when the same voice is pronounced twice in succession, the radio 401 can be controlled only if the recognition results of the two voices are the same. Particularly in relatively noisy environments such as the inside of a car, the noise level is high and there are a variety of noises, but it is generally rare for the same word's speech waveform to occur twice in succession. It is. Therefore, the voice input control device according to this embodiment can prevent malfunctions caused by noise.

FIG. 4 is a schematic block diagram of one embodiment of the present invention. In the example shown in FIG. 3 above, each time the user attempts to control the radio 401, the same voice must be pronounced twice in a row.
You may feel some inconvenience. In addition, from the time when the intention to control the radio 40
1 also takes a little longer to operate. The embodiment shown in FIG. 4 attempts to solve these problems.

That is, in this embodiment, sounds as keywords are classified into at least two types of hierarchy. One voice is a group of keywords required at the start of a series of control operations, such as turning on the power of a controlled device, and the other is a group of keywords that are frequently used in a short period of time, such as selecting a station on the radio 401 or adjusting the volume. . In response to this, the speech recognition circuit 100
is provided with registered parameter memories 104 and 154 that store feature parameters of at least two types of keywords. Furthermore, the speech recognition circuit 1
A classification circuit 107 is provided for classifying which of two types of hierarchy the keyword recognized by 00 belongs to. This classification circuit 107 demultiplexes the memory address signals of the registered parameter memories 104 and 154 output from the timing signal generation circuit 106, and can be constructed by combining several logic gate elements. Then, the output of the recognition result recognized by the determination circuit 105 and stored in the recognition result register 110 is given to the classification circuit 107 . The classification circuit 107 operates to force the comparison circuit 112 to output a matching signal when the determination circuit 105 determines that the input speech is the second keyword group. Conversely, if the input voice is of the first keyword group, the comparison circuit 112 is caused to perform the same operation as in the embodiment shown in FIG. 3 above. For this purpose, the comparator circuit 112 can be modified by simply replacing a 2-input gate element with a 3-input gate element, for example.

By configuring as described above, when starting an operation such as turning on the power, the same voice can be heard twice.
The control action is achieved only after the voice is pronounced once, and the control action can be achieved only by pronouncing the voice once for the second keyword group. Here, the operation of this embodiment will be explained using an example of the voice input procedure. The first keyword group is "Nishiki" as a keyword for calling, and "radio" is a keyword for powering on radio 401.
A keyword corresponding to an operation such as channel selection, for example, "1 channel" is classified into the second keyword group. The user first pronounces the same keyword twice in a row, such as "ninshiki" and "ninshiki." This keyword is recognized by the speech recognition circuit 100, the most recent recognition result is stored in the new recognition result register 110, and when the second speech is recognized, the contents of the new recognition result register 110 are changed to the old recognition result register 110. The recognition result of the voice that was transferred to and pronounced the second time is stored in the new recognition result register 110. At this time, the classification circuit 107 classifies the pronounced voice as belonging to the first keyword group,
Comparison circuit 112 is not forced to output a match signal. However, when the comparison circuit 112 determines that the contents of the new recognition result register 110 and the contents of the old recognition result register 111 match, it provides a match signal to the control circuit 200. Accordingly, control circuit 2
00 outputs a control signal for operating the drive circuit 300. As a result, drive circuit 300 is activated, and radio 401 can be controlled from now on.

Next, when the user pronounces the words "radio" and "radio" of the second keyword group, the classification circuit 107
determines that the voice belongs to the first keyword group, and the comparison circuit 112 outputs a matching signal. In response to this coincidence signal, the control circuit 200 controls the radio 401.
The drive circuit 30 sends a control signal to turn on the power.
Give to 0. Accordingly, the drive circuit 300
Turn on the power of 01.

Next, the user pronounces a voice for selecting channels such as "1 channel" or "5 channel" only once. The classification circuit 107 selects the pronounced voice as the second
The comparison circuit 112 is forced to output a matching signal. The control circuit 200 recognizes new result register 110
Based on the recognition result stored in , for example, a control signal for receiving one channel on radio 401 is generated. Accordingly, the drive circuit 301 causes the radio 401 to receive one channel.

As mentioned above, according to this embodiment, radio 4
At the start of a series of control operations such as powering on the 01, the same voice is emitted twice, and after the start of the control operation, for things that require frequent control such as channel selection, the same voice is emitted only once. Therefore, the user does not feel any inconvenience, and the time from when the user intends to control until the radio 401 actually performs the desired operation can be shortened.

Note that the speech recognition circuit 10 in the above-mentioned embodiment
0, control circuit 200, recognition result register 110,
If the circuit 111 and the comparison circuit 112 are formed on the same semiconductor chip, the entire device can be made smaller.

As described above, according to the present invention, a voice for controlling a controlled device is emitted twice, and when the characteristic parameters of each voice match, the target controlled device is controlled. By controlling the controlled device if the voice is of the first type even if the characteristic parameters of the voice do not match, malfunctions of voice recognition due to ambient noise etc. can be prevented. can do. Moreover, since the controlled device can be controlled by voice input, no manual operation is required and practical functions can be provided. Furthermore, the same sound is emitted twice at the start of the control action, and for items that require frequent control after the start of the control action, the sound only needs to be emitted once. It can save time.

[Brief explanation of drawings]

FIG. 1 is a schematic block diagram of a voice input control device which is the background of the present invention. FIG. 2 is a schematic block diagram of the speech recognition circuit shown in FIG. 1. FIG. 3 is a schematic block diagram of a device for recognizing twice-pronounced speech, which is the premise of the present invention. FIG. 4 is a schematic block diagram of one embodiment of the present invention. In the figure, 1 is a microphone, 100 is a speech recognition circuit, 101 is a speech analysis circuit, 102 is a feature parameter register, 103 is a number similarity calculation circuit,
104, 154 are registered parameter memories, 105
106 is a determination circuit; 106 is a timing signal generation circuit;
107 is a classification circuit, 110 is a recognition new result register, 111 is a recognition old result register, 112 is a comparison circuit, 200 is a control circuit, 300 is a drive circuit, 4
01 indicates a radio.

Claims (1)

[Scope of Claims] 1. A voice input control device that controls a controlled device according to the input of a first type of voice and a second type of voice, comprising: audio input means for outputting audio signals corresponding to the respective voices in response to input of first and second types of audio; Feature parameter extraction means for outputting feature parameters of speech; first feature parameter storage means for storing feature parameters of the first type of speech; second features for storing feature parameters of the second type of speech. parameter storage means; recognition means for recognizing whether the feature parameter of the voice outputted from the feature parameter extraction means is similar to any of the feature parameters stored in the first or second feature parameter storage means; , a first recognition result storage means for storing the recognition result of the recognition means, a second recognition result storage means for storing the previous recognition result by the recognition means, stored in the first recognition result storage means. Comparing means for comparing the current recognition result and the previous recognition result stored in the second recognition result storage means to determine a match; a recognition result by the voice recognition means and a matching output from the comparing means; control means for controlling the controlled device according to the input voice; and the recognition means recognizes that the characteristic parameter of the input voice is similar to the characteristic parameter of the first type of voice, and recognizes that the characteristic parameter of the input voice is similar to the characteristic parameter of the first type of voice, Depending on what the first recognition result storage means stores,
A voice input control device, comprising means for causing the control means to control the controlled device according to the input voice even if the comparison means derives a non-coincidence output signal.