JPH0228159B2 - - Google Patents

Description

[Detailed Description of the Invention] [Technical Field of the Invention] The present invention relates to a response confirmation method for the recognition status of a speech word recognition device, and in particular, when responding to a speaker with an audible sound, an audible sound is generated according to the recognition status. The present invention relates to a response confirmation method for a speech recognition device in which the frequency and duration of the speech are changed and special effects are added.

[Prior art and problems]

FIG. 1 shows an example of a response confirmation method of a conventional speech word recognition device, in which 1 is an electronic computer;
Reference numeral 2 indicates a speech word recognition device, 3 a speech synthesis device, 4 an earphone, and 5 a microphone. The words uttered by the speaker are input to the voice word recognition device 2 via the microphone 5 and are voice recognized. The computer 1 controls the speech synthesizer 3 according to the speech recognition result. The speech synthesis device 3 outputs the speech recognition result in the form of speech, and this output is transmitted to the vocalized sound via the earphone 4.

FIG. 2 shows another example of the response confirmation method of the conventional speech word recognition device, and 6 indicates a display device. Note that the same reference numerals as in FIG. 1 indicate the same parts. The words pronounced by the vocalizations are input to the speech word recognition device 2 via the microphone 5. The voice recognition result is input to the computer 1, and the computer controls the display device 6 according to the voice recognition result. The display device 6 displays the voice recognition results.

The disadvantages of the conventional examples shown in FIGS. 1 and 2 are that they both require wasted time even when the input is normal, and the former has the disadvantages of being complicated and expensive, and the latter It has the disadvantage that the line of sight of the vocalization moves to the display device each time.

[Purpose of the invention]

SUMMARY OF THE INVENTION It is an object of the present invention to provide a response confirmation method for a speech word recognition device that eliminates the above-mentioned drawbacks, has a simple and inexpensive structure, and is capable of efficiently confirming responses.

[Structure of the invention]

Therefore, the response confirmation method of the speech word recognition device of the present invention compares the standard pattern of registered words stored in the registered pattern memory with the recognition pattern for the speech word input, and compares the recognition pattern with the standard pattern. a voice word recognition device that generates a numerical value indicating the degree of similarity between words; a response confirmation device that conveys the recognition status of the voice word recognition device to the speaker by an audible signal; and a maximum value obtained by the voice word recognition device. A response confirmation method for a spoken word recognition device comprising a computer that notifies the response confirmation device of a numerical value of similarity, a difference between a numerical value of maximum similarity and a similar value of next-order similarity, and error information indicating the presence or absence of an error. The response confirmation device includes a synthesizer that generates an audible sound signal with a variable frequency, a timer that passes the audible sound signal generated by the synthesizer for a set time, and a new effect on the audible sound signal output from the timer. of the synthesizer, timer, and effect adding circuit according to the effect adding circuit to be added, the maximum similarity value sent from the computer, the difference between the maximum similarity value and the next similarity value, and error information. It is characterized in that it has a controller that controls one or more devices therein.

[Embodiments of the invention]

Hereinafter, the present invention will be explained with reference to the drawings.

FIG. 3 is a diagram showing an example of the speech word recognition device used in the present invention, FIG. 4 is a diagram showing one embodiment of the present invention, and FIG. 5 is a diagram showing the frequency and frequency of the audible sound signal used for response confirmation. FIG. 3 is a diagram for explaining duration time.

In FIG. 3, 11 is an automatic volume controller, 12 is a spectrum analysis section consisting of a group of bandpass filters, 13 is a multiplexer,
14 is an A-D converter, 15 is a sampling circuit,
Reference numeral 16 indicates a central processing unit, 17 a program memory, and 18 a registered pattern memory.

The automatic volume controller 11 controls the volume of audio input to a constant level. The output of automatic volume controller 11 is analyzed by spectrum analyzer 12. A plurality of output signals outputted from the spectrum analysis section 12 are selected by a multiplexer 13, and the output of the multiplexer 13 is converted into a digital signal by an AD converter 14. The output of the AD converter 14 is time sampled by a sampling circuit 15. The recognition pattern for speech recognition is, for example, In the above matrix, columns correspond to sampling times, and rows correspond to frequency bands. The recognition pattern of the voice input is stored in a working area of the program memory 17, for example. Needless to say, programs are also stored in the program memory 17. The registered pattern memory 18 stores standard patterns of words. The recognition pattern for voice input is
It is successively compared with the standard pattern in the registered pattern memory, and the distance to each registered word is determined. The difference in distance between the maximum distance setting value and the registered word is determined, and this is used as a score. The higher the score, the greater the possibility of normal recognition. Select the registered word with the maximum score and the registered word in second place. In this case, the greater the difference between the first and second place, the higher the possibility of normal recognition.

FIG. 4 is a block diagram of one embodiment of the present invention. In FIG. 4, 19 is a controller, 20
21 is a synthesizer, 21 is a timer, 22 is an effect adding circuit, and 23 is an amplifier.

The controller 19 operates a synthesizer 20 and a timer 2 according to data sent from the computer.
1 and the effect adding circuit 22. The synthesizer 20 is a variable frequency oscillator, and changes its oscillation frequency according to control commands from the controller 19. The timer 21 passes the output signal of the synthesizer 20 for a specified period according to a control command from the controller 19. Effect addition circuit 2
2 intermittents the input signal according to a control command from the controller 19. The output of the effect adding circuit 22 is amplified by an amplifier 23 and sent to a speaker or earphone.

FIG. 5 illustrates the frequency and duration of the audible signal used for response confirmation. Third
The recognition status of the speech word recognition device shown in the figure is sent to an electronic computer (not shown). The electronic computer sends error information to the controller 19 indicating the maximum score for the input word, the difference between the maximum score and the next score, and whether there is an error in grammar or software processing.
send to The controller 19 controls the synthesizer 20 and the timer 21 according to the graph shown in FIG.
control. In Figure 5, I ₁ is the maximum score,
I ₂ indicates the runner-up score, and L indicates the maximum setting value. As can be seen from FIG. 5, the greater the maximum score _I1 , the greater the frequency of the audible signal, and the smaller the difference between the maximum score and the next score, the longer the duration of the audible signal. In the illustrated example,
When the maximum score I ₁ is close to the maximum setting value L, the frequency of the audible sound signal is 1000 Hz, and when the maximum score I ₁ is close to zero, the frequency of the audible sound signal is 300 Hz. Also, in the illustrated example, the maximum score I ₁
If the difference between and the next score _I2 is close to zero, the duration of the audible sound signal is set to 1 second, and if the above difference is close to the maximum setting value L, the duration of the audible sound signal is set to 0.1 seconds. be done. If the error information sent from the computer indicates that there is an error, the controller 19 operates the effect adding circuit 22. When the operation starts, the effect adding circuit 22 intermittents the input signal. This intermittent cycle is, for example, 20Hz.
Instead of using the effect adding circuit 22 to intermittent the audible sound signal, the synthesizer 20 can also be controlled to vibrate.

It is also possible to perform response confirmation more correctly and effectively by combining the conventional device shown in FIG. 2 with the device of the present invention. In this case, the speaker performs the following operation.

(b) If the audible tone for response confirmation is high and short,
Determine that the input was successful and proceed to the next step.

(b) If the audible tone for response confirmation is intermittent, it is determined that the input was not performed correctly, and the same input as the previous one is performed again.

(c) If the audible tone for response confirmation is low and long, look at the display device, and if the input is successful, proceed to the next step; if there is an error, enter the same input again to correct the error. conduct.

〔Effect of the invention〕

As is clear from the above explanation, according to the present invention, it is possible to notify the speaker of the recognition status of the spoken word input using a simple circuit, and in the case of normal input, it is possible to efficiently inform the speaker using a short sound. In the case of an abnormality, a long sound can be fed back to the vocalization sound accordingly, so that the efficiency of use of the speech word recognition device can be improved. Spoken word recognition devices can usually perform 95% to 99% recognition, and are of great significance when applied in cases where there are very few errors. If only about 50% recognition could be performed, the present invention would have little significance.

[Brief explanation of drawings]

1 and 2 are diagrams showing a response confirmation method of a conventional spoken word recognition device, FIG. 3 is a diagram showing an example of a spoken word recognition device used in the present invention, and FIG. 4 is a diagram showing an example of a spoken word recognition device used in the present invention. FIG. 5, which is a diagram showing one embodiment, is a diagram for explaining the frequency and duration of an audible sound signal used for response confirmation. 11... automatic volume controller, 12...
Spectrum analysis section consisting of a group of bandpass filters, 13...Multiplexer, 14...A-D converter, 15...Sampling circuit, 16...Central processing unit, 17...Program memory, 18...Registered pattern memory, 19...Controller , 20...Synthesizer, 21...Timer, 22...Effect addition circuit,
23...Amplifier.

Claims (1)

[Claims] 1. A spoken word recognition device that compares a standard pattern of registered words stored in a registered pattern memory with a recognized pattern for inputted spoken words and generates a numerical value indicating the degree of similarity between the recognized pattern and the standard pattern; a response confirmation device that conveys the recognition status of the speech word recognition device to the speaker by an audible sound signal; a maximum similarity value obtained by the speech word recognition device; the maximum similarity value and the next-order similarity; This is a response confirmation method for a voice word recognition device that includes a computer that notifies the response confirmation device of error information indicating the difference between numerical values and the presence or absence of an error, the response confirmation device transmitting an audible sound signal of variable frequency. a timer that passes the audible sound signal generated by the synthesizer for a set time, an effect adding circuit that adds a new effect to the audible sound signal output by the timer, and a maximum output signal sent from the computer. a controller for controlling one or more devices among the synthesizer, timer, and effect adding circuit according to a numerical value of similarity, a difference between a numerical value of maximum similarity and a numerical value of next-order similarity, and error information; A response confirmation method for a voice word recognition device characterized by: