JPH0540498A - Voice recognizing device - Google Patents

Abstract

PURPOSE:To eliminate the need of operating a voice input switch whenever a voice is inputted, although the voice input switch is operated in order to prevent a malfunction caused by an ambient noise. CONSTITUTION:By a filter bank 34, a multiplexer 36 and an A/D converter 38, a voice inputted from a microphone 32 is converted to voice pattern data, and it is compared with standard pattern data registered in advance in a standard pattern table. When a voice input is allowed, a voice input switch 30 is turned on, and also, its turn-on time is set to a turn-on time timer of a memory 42. When an input voice is discriminated within a set time, an extension time is set to the turn-on time timer, and continuous voice inputs are allowed.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice recognition device,
In particular, the present invention relates to a voice recognition device that compares a voice pattern obtained by analyzing a voice input from a microphone with a preset standard pattern to recognize the voice.

[0002]

2. Description of the Related Art When a voice recognition device of this type is used in the vicinity of an audio device such as a stereo, the output sound from the audio device becomes ambient noise with respect to the voice recognition device, and there is a danger of erroneous recognition occurring frequently. In particular, for example, when controlling or operating such an audio device based on the recognition result of the voice recognition device, the voice or music output from the audio device is input to the voice recognition device in a considerable amount. So
There is an inconvenience that the voice recognition device operates undesirably. In order to prevent such a malfunction, a voice recognition device has been proposed in which, when voice input is performed to the voice recognition device, the output of the acoustic device is reduced only during the voice input period (Japanese Patent Laid-Open No. 63-29755). (See Japanese Patent Publication).

Further, in such a voice recognition device, generally, a voice pattern including a parameter representing a feature of a voice obtained by analyzing a voice input from a microphone is referred to as a preset standard pattern. The input voice is recognized by comparing and selecting the most similar standard pattern. In such a voice recognition device, even if the most similar standard pattern is selected, if the degree of similarity is extremely small, there is a high possibility that it is erroneous recognition. If it does not exceed the threshold of, recognition is generally rejected.

[0004]

In the former case, in order to set the voice input possible period, a complicated operation such as operating a switch each time voice input is required. In the latter case, if the threshold value of the similarity is too large, a recognition result is often not obtained due to a subtle ambiguity element of the voice, and if the threshold value is reduced, even noise may be erroneously recognized as voice. There is an inconvenience.

Therefore, the main object of the present invention is to prevent malfunctions due to ambient noise without complicated operations.
A voice recognition device is provided. Another object of the present invention is to provide a voice recognition device that can eliminate the inconvenience caused by setting the threshold value of the similarity. Still another object of the present invention is to provide a voice recognition device that can prevent erroneous recognition when a voice that is not a recognition target is input.

Still another object of the present invention is to provide a voice recognition device capable of preventing registration error as much as possible when registering a plurality of voices as a standard pattern for one item.

[0007]

SUMMARY OF THE INVENTION A first aspect of the present invention is a pattern creating means for analyzing a voice input from a microphone to create a voice pattern, and a recognition by comparing a voice pattern with a standard pattern registered in advance. A voice recognition device having a recognition means for setting a time setting means for setting an input time for allowing voice input from a microphone,
And a voice recognizing device further comprising extension means for extending the input time when the voice is recognized by the recognition means within the input time set by the time setting means.

A second aspect of the invention is a pattern creating means for analyzing a voice input from a microphone to create a voice pattern, a standard pattern setting means for presetting a plurality of voice patterns as standard patterns, and a pattern creating means. The selected voice pattern is compared with the respective standard patterns set in the standard pattern setting means, and the selecting means for selecting the standard pattern having the highest similarity, and the similarity of the standard pattern selected by the selecting means are predetermined. When a voice recognition device having a determination means for recognizing a voice by a standard pattern when the value is larger than a threshold value and rejecting the similarity when the similarity is smaller than a predetermined threshold value, the same voice is continuously rejected multiple times by the determination means. Characterized in that it further comprises means for making speech a recognition result. That is a voice recognition device.

A third aspect of the present invention is a pattern creating means for analyzing a voice input from a microphone to create a voice pattern, a standard pattern setting means for presetting a plurality of voice patterns as a standard pattern, and a pattern creating means. The selected voice pattern is compared with the respective standard patterns set in the standard pattern setting means, and the selecting means for selecting the standard pattern having the highest similarity, and the similarity of the standard pattern selected by the selecting means are predetermined. When the selected standard pattern is out of the identification target, the speech recognition apparatus includes a determination unit that recognizes the voice by the standard pattern when the standard pattern is larger than the threshold value and rejects when the similarity is smaller than the predetermined threshold value. Further comprising means for stopping the output of the recognition result from the means Wherein a speech recognition device.

According to a fourth aspect of the present invention, in a voice recognition device in which a voice pattern obtained by analyzing a voice input from a microphone is registered in advance as a standard pattern, different voice standard patterns are registered in different modes for one item. The voice recognition device is characterized by further comprising display means for indicating a mode to be registered.

[0011]

In the first aspect of the invention, the voice input switch is turned on to allow the voice input. At the same time, for example, an on-time timer sets an input time for allowing voice input. When the recognition unit recognizes the voice within the input time set in the timer, the extension unit sets the extension time in the timer again to extend the input time.

In the second aspect of the invention, when the similarity to the standard pattern selected by the selecting means is smaller than the threshold value, the standard pattern is rejected. When the standard pattern is rejected a plurality of times in succession, the standard pattern is used. Output the recognition result. Therefore, the rejected voice can be recognized by inputting it again several times.

In the third aspect of the present invention, the recognition result is not output when the standard pattern of the non-recognition speech shows the maximum similarity. Therefore, a malfunction does not occur due to a voice that is not a recognition target. According to the fourth invention, the mode in which the voice should be registered at that time is displayed by a different display mode such as an LED.

[0014]

According to the first aspect of the invention, since the voice input possible time is extended once the voice is recognized, it is not necessary to set the input time again in the case of continuous voice input. Therefore, in order to set the voice input possible period in order to prevent malfunction, it is not necessary to perform a complicated switch operation as in the conventional case. Also, since the voice is recognized only at the input time,
Ambient noise is less likely to be input to the microphone, and noise does not cause malfunction as in the conventional case.

According to the second aspect of the invention, it becomes easy to set a threshold value for discriminating the degree of similarity required for judgment or recognition. That is, in the prior art, if the threshold value is set small, it malfunctions due to noise such as surrounding voice, and if the threshold value is set large to prevent such malfunction, the probability of being rejected by an ambiguity element peculiar to the voice is large. Becomes
Although it was difficult to set the threshold value, according to the second aspect of the invention, if the same word is input by voice a plurality of times and the word is continuously rejected, the rejected word is recognized, resulting in malfunction. Even if the threshold value is set to a large value in order to prevent this, recognition can be performed by repeating the same voice input several times. Also, in the case of sudden sound or speech of conversation, the same word is rarely input repeatedly, so
It is possible to reduce malfunctions caused by such sudden sounds and conversation voices.

According to the third aspect, since the recognition result is not output when a voice that is not a recognition target is input, it is possible to prevent a malfunction due to the recognition result. If the user inputs a re-input instruction, the user can input the voice again. Also, if the voice is not input again within the predetermined time, if recognition is performed based on the most similar standard pattern in the target, the voice in the recognition target is input, but the target is outside the target. Even if it is determined that there is a voice input, it is recognized as a voice input in the target.

According to the fourth invention, registration errors can be reduced even when, for example, one control target is controlled by two or more recognition results. The above-mentioned objects, other objects, features and advantages of the present invention will become more apparent from the following detailed description of the embodiments with reference to the drawings.

[0018]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A car audio system 10 of the embodiment shown in FIG. 1 includes a microcomputer 12, and an audio section 14 is controlled by the microcomputer 12.
The audio section 14 includes a tuner 18 and a tape deck 20.
And a stereo sound source 16 including a CD player 22 and the like, and a right signal R and a left signal L from this stereo sound source 16 are passed through amplifiers 24R and 24L, respectively.
It is given to speakers 26R and 26L arranged at appropriate positions in the interior of a vehicle (not shown). When the stereo sound source 16 is 4-channel stereo, a rear signal is further output. The audio unit 14 further includes a controller 28, and the controller 28 includes an operation switch (not shown) for manually operating the stereo sound source 16. However, when the audio unit 14, that is, the stereo sound source 16 is controlled by the control signal from the microcomputer 12, the voice input switch 30 provided in the audio unit 14 is operated. In this case, the control signal from the microcomputer 12 is replaced by the control signal from the microcomputer 12 instead of the operation signal from the operation switch.
6 is input. Note that the audio unit 14 is provided with a light emitting diode (LED) 31, and as described later, for example, a voice that is not a recognition target is input by the LED 31, and therefore voice input is required again.
Alternatively, the operator is notified of various procedures such as registration procedures.

On the other hand, a vehicle dashboard (not shown) has a microphone 32 for picking up a driver's voice for controlling the audio section 14.
Are placed. The audio signal from the microphone 32 is given to the filter bank 34. Filter bank 3
As well known, 4 includes, for example, an 8-channel band pass filter, and the band pass filter extracts the characteristic parameter of the audio signal input from the microphone 32. That is, the filter bank 34 includes a preamplifier, an AGC, a bandpass filter, a rectifier circuit, and a lowpass filter for each channel. Each characteristic parameter (analog signal) from the filter bank 34 is input to the multiplexer 36. The multiplexer 36 receives the 8 input from the filter bank 34.
The characteristic parameter signals of the channels are output in time sequence.
The audio signal output from the multiplexer 36 is converted into characteristic parameter data by the A / D converter 38.

The signal from the voice input switch 30 and the output of the A / D converter 38 are input to the microcomputer 12 via the input port 40. The microcomputer 12 compares the characteristic parameters input from the input port 40 with the respective standard patterns in the standard pattern table 42a (FIG. 2) formed in the memory 42, as described later, to input from the microphone 32. The recognized voice. Then, according to the recognition result, the aforementioned control signal is output to the audio section 14 through the output port 44.

Therefore, when the voice input switch 30 is operated, the audio section 14 is added to the microphone 32.
When a voice for controlling is input, a control signal is output from the microcomputer 12 according to the voice. In response to this control signal, the controller 28 controls the stereo sound source 16. As shown in FIG. 2, the memory 42 includes a standard pattern table 42a. In the standard pattern table 42a, a standard of each sound or word for recognizing a voice based on the characteristic parameters cut out by the filter bank 34. A pattern of characteristic parameters is registered in advance for each number. The standard pattern table 42a is, for example, a backup R
Composed of AM. A start end flag 42a is further formed in the memory 42, and the start end flag 42b is set when the voice data first exceeds the threshold as shown in FIG. 3, that is, when the start end of the voice indicated by "Fh" is detected. Turned on. The memory 42 further includes an audio data buffer 42c, and the audio data buffer 42c stores audio data from the A / D converter 38 fetched by the microcomputer 12. The audio data buffer 42c has a plurality of addresses so that a series of audio data from the start end (shown by "Fh") to the end (shown by "Ft") shown in FIG. 3 can be stored over a plurality of frames. Have. However, one frame is set to 5 milliseconds, for example. That is, the audio data buffer 42c is the A / D converter 3
The characteristic parameter data of the voice input from the microphone 8 and input to the microphone 32 is stored in frame order.

The memory 42 further includes a time table 42d having a setting area unique to each number of the standard pattern table 42a, and the time table 42d includes:
An "extended time" that is uniquely determined for each standard pattern set in the standard pattern table 42a is set. This extension time means the time when the on time of the voice input switch 30 should be extended. For example, in order to achieve one control with a series of two or more voices, it is necessary to keep the ON state of the voice input switch 30 after the previous voice is recognized, but how much the ON time is extended. The extension time indicating whether or not it should be set is set in the time table 42d. Then, as described later, the time read from the time table 42d is set in the on-time timer 42e which is also assigned to the memory 42.

Reject flag 4 included in the memory 42
2f is turned on when proper recognition cannot be performed (when recognition is rejected), and the reject number register 42g stores the number of the standard pattern table 42a indicating the word rejected in this way. The reject counter 42h counts the number of rejects, and is incremented each time it is rejected.

The re-input timer 42i of the memory 42
Is a timer for setting a time for allowing the operator to re-input a word that is not a recognition target. Also,
The blinking time timer 42j is a timer for setting a time interval for blinking the LED 31. The registration mode shown in FIG. 4 is set according to the operation of a registration key (not shown), and in the first step S1, a registration number is also set using a numeric keypad (not shown). This registration number is a number in the standard pattern table 42a, and a standard pattern of a word to be recognized is registered for each number. For that purpose, the user voice-inputs to the microphone 32 (FIG. 1) the word to be registered with the number. Depending on,
In step S2, sampling of voice input is started, and voice (parameter) data is input to the microcomputer 12 via the filter bank 34, the multiplexer 36 and the A / D converter 38 as described above. Therefore, in step S3, the microcomputer 12 takes in the audio data and temporarily stores it in a buffer (not shown). Next step S
At 4, the microcomputer 12 determines whether or not the beginning of the voice (which corresponds to "Fh" in FIG. 3) has already been detected. If the start point of the voice has not been input yet, it is determined in step S5 whether the voice data input in step S3 is the start point. In this step S5, "N
If "O" is determined, the process returns to step S3. If the input audio data is for the start end, the microcomputer 12 sets the start end flag 42b (FIG. 2), and "YES" at the previous step S4. In the same manner as when it is determined that the voice buffer 42 receives the voice data previously fetched in step S7.
Store in c (FIG. 2). Then, in step S8, it is determined whether or not the input voice data is the end (corresponding to "Ft" in FIG. 3).
If not, the process returns to the previous step S3. In this way, steps S3 to S8 are repeatedly executed, and the audio data from the start end to the end is frame-sequentially stored in the audio buffer 42c.

Then, in step S9, the microcomputer 12 normalizes the data stored in the audio buffer 42c (specifically, data compression). In step S10, the normalized voice data is saved in the area corresponding to the number set in step S1 of the standard pattern table 42a. In the next step S11, "extension time" is set in the time table 42d. That is, in this step S11, when the word for which the standard pattern is set is input to the standard pattern table 42a, the extension time indicating how much the voice input possible time (described later) should be extended is individually set. It Then, in step S12, it is determined whether or not to end the registration mode according to whether or not the registration key is operated again. If the registration operation is to be continued, the registration number is changed in step S13 and the process returns to step S2. In this way, the standard pattern data of the word to be recognized is registered in the standard pattern table 42a, and the data representing the extension time when each word is recognized is registered in the time table 42d in advance.

First step S of the recognition mode shown in FIG.
At 101, the microcomputer 12 receives the signal from the input port 40 (FIG. 1) and outputs the voice input switch 3
It is determined whether 0 is operated, that is, whether it is a voice input possible period. Then, step S101
When it is detected that the voice input switch 30 is turned on,
In the next step S102, the microcomputer 12 sets the ON time timer 42e (FIG. 2) to a predetermined time (for example, 10 seconds) for continuing the ON state of the voice input switch 30.

Thereafter, steps S103, S104, S
105, S106 and S108 are executed. These steps correspond to the steps S2, S3, S4, S5 and S6 described in the registration mode of FIG. 5, respectively, and thus duplicated description will be omitted here. Then, in step S107, it is determined whether or not the voice data input in step S104 is input within the voice input possible time set in the on-time timer 42e in step S102. If "YES" is determined in this step S107, the previous step S104
However, if "NO" is determined, the step S107a is executed.
At 12, the microcomputer 12 forces the voice input switch 30 to the off state and returns to step S101. That is, if there is no voice input within the predetermined time set in the on-time timer 42e after the voice input switch 30 is turned on, the microcomputer 12 turns off the voice input switch 30 and the recognition operation thereafter is not executed. ..

Following step S108, steps S109 and 110 shown in FIG. 6 are executed, which are steps S7 and S8 in the previous registration mode.
The same description is omitted here, and the duplicated description is omitted here. Then, in step S111, the microcomputer 12 calculates the degree of similarity between the audio data stored in the audio buffer 42c and each of the standard patterns registered in advance in the standard pattern table 42a. Then, the standard pattern indicating the maximum degree of similarity is selected in step S112.
In step S113, the first threshold value for discriminating the degree of similarity is set, and in step S113
Proceed to 114. The first threshold set in step S113 is relatively large, and if the similarity in the case of perfect identity is "100", this first threshold is, for example, "9."
It is set to "0". Then, in step S114, it is determined whether the similarity of the standard pattern selected in step S112 exceeds the first threshold value set in step S113. When the maximum similarity is larger than the first threshold value, the word indicated by the standard pattern that gives the maximum similarity is output as the recognition result (step S11).
5).

In the following step S116, extension time data is read from the area of the number corresponding to the word in the time table 42d, and the extension time is set in the on-time timer 42e in the same manner as in step S102. That is, when the input voice is identified by the standard pattern registered in advance in the standard pattern table 42a in step S115, the on-time timer 42e is reset in step S116 to allow the voice input continuously. , And returns to step S103 (FIG. 5) to wait for the subsequent voice input. As described above, when the input voice is recognized, the voice input possible time is extended, and therefore, even when the voice is continuously input, it is not necessary to operate the voice input switch 30 again. For example, by controlling the tape deck 20 of the car audio system 10,
When you want to "fast forward", "fast forward", "play",
It is only necessary to continuously input the voice by "fast forward", ... "Play", but even in this case, the voice can be continuously input only by first turning on the voice input switch 30 once. Further, in steps S107 and S107a, after the time set in the on-time timer 40e has elapsed, voice input cannot be performed, so that malfunction due to ambient noise can be prevented.

Note that step S116 is step S11.
If it is executed only when the word indicated by the specific number in 5 is recognized, that is, if the voice input possible time is extended only when the specific word is recognized, malfunction due to ambient noise is possible. The property can be further reduced. If it is determined in the previous step S114 (FIG. 6) that the similarity of the standard pattern indicating the maximum similarity selected in step S112 is smaller than the first threshold value, the process proceeds to step S117 shown in FIG. That is,
In step S117, the reject flag 42f
To determine if is turned on. If the reject flag 42f is turned off, step S
At 118, the reject flag 42f is set, the number of the rejected word (standard pattern) is stored in the reject number register 42g, the reject counter 40h is incremented, and then the process returns to step S103 (FIG. 5).

When it is detected in step S117 that the reject flag 42f is already turned on, in the next step S119, the microcomputer 12 is detected.
Refers to the reject number register 42g, and determines whether the number of the standard pattern rejected immediately before is the same as the number of the standard pattern rejected this time, that is, whether the same word is continuously rejected. .. If the previously rejected word is different from the currently rejected word, that is, if “NO”, in step S120, the reject number register 42
g is updated with the number of the standard pattern rejected this time, the reject counter 42h is incremented, and the process returns to step S103.

If the previously rejected number and the currently rejected number are the same, that is, step S
If “YES” is determined in 119, the microcomputer S121 sets a second threshold value, such as “80”, which is slightly smaller than the first threshold value, and the step S
At 122, it is determined whether the maximum similarity selected at step S112 (FIG. 6) exceeds the second threshold set at step S121. If the maximum similarity exceeds the second threshold, the recognition result is output based on the standard pattern. However, if the maximum similarity is equal to or less than the second threshold value, in step S123, the microcomputer 12 refers to the reject counter 42h and determines whether the number of rejects has reached a predetermined number n (for example, 3). to decide. Step S
When it is determined to be "YES" in 123, the microcomputer 12 outputs the number loaded in the reject number register 42g as a recognition result in step S124. If the number of rejects has not reached the predetermined number, in step S125,
The reject counter 42h is incremented, and a third threshold value of, for example, "70" smaller than the second threshold value is set, and the process returns to step S103.

In this way, when consecutive voice inputs are identified as the same standard pattern and are rejected in the same manner, the threshold value of the similarity is set to be gradually smaller, so that the voice input is performed again. It can be recognized. Therefore, even if the first threshold value that is initially set is set to be relatively large to reduce erroneous recognition as much as possible, there is no possibility that voice input cannot be continued due to rejection. Further, when the voices are rejected a predetermined number of times (for example, three times) in the same manner, the voice identified by the standard pattern indicated by the rejected number is identified (step S124), and therefore voice input is performed several times in the same manner. By repeating, the voice is surely input. In the case of sudden sound or conversation, the same word is rarely repeated, so there is no malfunction due to sudden sound or conversation.

From step S118, S120 or S125 in FIG. 7, the process returns to step S103 in FIG. 5, but the input time set in step S102 is still valid at that time. When the same voice is repeatedly input and rejected, the process shown in FIG. 7 becomes effective. If there is no re-voice input within the input time, it ends as rejected.

In another embodiment, step S1 shown in FIG.
Subsequent to step 13, step S201 shown in FIG. 8 is executed. In this step S201, similarly to step S114, it is determined whether or not the maximum degree of similarity shown in step S112 exceeds the first threshold value determined in step S113. When the maximum similarity does not exceed the first threshold value, that is, when the rejection is performed, the process may proceed to step S117 of FIG. 7 as in the previous embodiment, or may be finished as it is. Good.

If the maximum similarity exceeds the first threshold, the microcomputer 12 determines in step S202 whether the word giving the maximum similarity is a recognition target. That is, assuming that there is a cassette tape mode and a tuner mode in the embodiment of FIG. 1, in each mode, as shown in Table 1, words to be recognized are preliminarily limited for each mode. ..

[0037]

[Table 1]

In this case, the microcomputer 12
For example, in tuner mode, registration number "1" ~
If any of "5" gives the maximum similarity, or if any of the standard patterns of registration numbers "6" to "13" gives the maximum similarity in the cassette mode, in step S202, the voice at that time is output. It is determined that the input is outside the recognition target. When it is determined that the object is not a recognition target, that is, when “NO” is determined in step S202, in step S203, the microcomputer 12 sounds, for example, a buzzer (not shown) or the LED 31 (FIG. 1). For example, by turning on the light, the user is notified that the unrecognized word has the maximum degree of similarity and thus needs to be re-input. At the same time, in step S204, the re-input timer 42i
A predetermined time, for example, 3 seconds is set in (FIG. 2). And
If there is no voice input within the time set in the re-input timer 42i, the process goes through step S205 and then step S20.
At 6, the microcomputer 12 determines the standard pattern that gives the maximum similarity within the recognition target. For example, when there is a voice input of "rewind" in the cassette mode, the voice is vaguely uttered, so step S11
If it is determined in 2 that it is the most similar to the standard pattern of "band change" and the next most similar is the standard pattern of "rewind", step S20.
In 6, the word indicating the maximum similarity in the recognition target, that is, “rewind” is determined, and whether the similarity exceeds the first threshold value is determined in the same manner as in step S201 described above.
The determination is made in step S207.

However, if there is voice input within the time set in the re-input timer 42i, step S in FIG.
The operation from 103 is executed, and the re-input voice is judged. Next, a modification of the registration mode shown in FIG. 4 will be described with reference to FIG. In this modification, as shown in Table 2, a so-called "multi-function" in which one key or switch has a plurality of functions
This is a registration method for achieving the above.

[0040]

[Table 2]

In order to achieve such a multi-function effect, it is necessary to register in advance two or more voices for one display, but it is difficult to distinguish between them, and therefore misregistration and misrecognition may occur. It was the cause. In order to solve such a problem, the embodiment shown in FIG.
When registering a voice for controlling a device controlled by the above voices, the user is informed of the fact according to a specific display, and as a result, false registration and false recognition are reduced. Is. That is, step S301
In, the microcomputer 12 determines whether it is a case of registering different words for each mode in one switch such as “1 / AMSS” and “2 / RPT” shown in Table 2. For example, the "1 / AMSS" switch is used to set one channel of AM broadcast in the AM radio mode, is used to set one channel of FM broadcast in the FM radio mode, and is set to the beginning in the cassette tape mode. Used for settings. Therefore, in this case, "YES" is determined in step S301. If not, microcomputer 1
In the next step S302, the LED 2 always lights up the LED 31 (FIG. 1). If “YES” is judged,
That is, when a plurality of voice registrations are performed for one switch, the microcomputer 12 sets the blinking mode of the LED 31 in the next step S303. Then, in step S304, it is determined whether or not it is necessary to register three or more voices with respect to one switch, such as "1 / AMSS". If only two voice registrations are required for one switch, that is, if "NO" is determined, step S305.
At the microcomputer 12, the blinking timer 4
Set the first timer time to 2j (Fig. 2), and conversely "YE
When S "is determined, the microcomputer 12 sets the second timer time in step S306. The first timer time and the second timer time are predetermined so that the blinking speed and interval of the LED 31 are different. Therefore, the user can register the voice pattern suitable for each mode by judging the lighting state of the LED 31 (that is, constant lighting, blinking 1 and blinking 2), and eliminate erroneous registration. ..

In the above-mentioned embodiment, the voice input switch 30 is provided to allow the voice input. However, without providing such a special switch, for example, "input" is used. The voice input enable state may be set by various voice inputs.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention.

FIG. 2 is an illustrative view showing the memory of FIG. 1 in more detail.

FIG. 3 is a waveform diagram showing the beginning and end of a recognized voice.

FIG. 4 is a flowchart showing a registration mode in the embodiment of FIG.

5 is a flowchart showing a part of a recognition mode in the embodiment of FIG.

6 is a flowchart showing a part of a recognition mode in the embodiment of FIG.

FIG. 7 is a flowchart showing a part of a recognition mode in the embodiment of FIG.

FIG. 8 is a flowchart showing a modification of the recognition mode in the embodiment of FIG.

9 is a flowchart showing a modification of the registration mode in the embodiment of FIG.

[Explanation of symbols]

 10 ... Car audio system 12 ... Microcomputer 14 ... Audio part 16 ... Stereo sound source 30 ... Voice input switch 31 ... LED 32 ... Microphone 34 ... Filter bank 36 ... Multiplexer 38 ... A / D converter 42 ... Memory 42a ... Standard pattern table 42c ... voice buffer 42d ... time table

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Koji Araki 2-18 Keihan Hondori, Moriguchi City, Osaka Prefecture Sanyo Electric Co., Ltd. (72) Inventor Koji Dejima 2-18 Keihan Hondori, Moriguchi City, Osaka Sanyo Electric Co., Ltd. Within the corporation

Claims (4)

[Claims] 1. A voice recognition device comprising pattern creating means for analyzing a voice input from a microphone to create a voice pattern, and recognition means for recognizing by comparing the voice pattern with a standard pattern registered in advance. In, a time setting means for setting an input time for allowing voice input from the microphone, and an extension for extending the input time when voice is recognized by the recognition means within the input time set by the time setting means. A voice recognition device, further comprising means. 2. A pattern creating means for analyzing a voice input from a microphone to create a voice pattern, a standard pattern setting means for presetting a plurality of voice patterns as a standard pattern, and a pattern creating means. Selecting means for comparing the voice pattern with each of the standard patterns set in the standard pattern setting means and selecting the standard pattern showing the highest degree of similarity;
And voice recognition provided with a judging means for recognizing the voice by the standard pattern when the similarity of the standard pattern selected by the selecting means is larger than a predetermined threshold value and rejecting when the similarity degree is smaller than the predetermined threshold value. The speech recognition apparatus, further comprising means for, when the same speech is rejected a plurality of times in succession by the determination means, using the speech as a recognition result. 3. A pattern creating means for analyzing a voice input from a microphone to create a voice pattern, a standard pattern setting means for presetting a plurality of voice patterns as a standard pattern, and a pattern creating means. Selecting means for comparing the voice pattern with each of the standard patterns set in the standard pattern setting means and selecting the standard pattern showing the highest degree of similarity;
And voice recognition provided with a judging means for recognizing the voice by the standard pattern when the similarity of the standard pattern selected by the selecting means is larger than a predetermined threshold value and rejecting when the similarity degree is smaller than the predetermined threshold value. The voice recognition device, further comprising means for stopping the output of the recognition result from the determination means when the selected standard pattern is out of the identification target. 4. A voice recognition device in which a voice pattern obtained by analyzing a voice input from a microphone is registered in advance as a standard pattern. When registering different voice standard patterns in different modes for one item, registration is performed. The voice recognition device, further comprising display means for indicating a mode to be performed.