JPH0749699A - Audio device responding to voice input - Google Patents

Abstract

PURPOSE:To obtain a device which recognizes words inputted by a voice input and replies the answer to the words in voiced words. CONSTITUTION:A voice input is divided into components for every frequency band by plural filter circuits. The signal waves are converted to digital numerical values by A-D converters 7 provided for each frequency band and each higher harmonic content peculiar to the voice is discriminated from the numerical values for voice recognition. The recognized voice is compared and detected against the voice data of the words beforehand registered in a memory 9 by a CPU 8 and the words of inputted voice are recognized. Then, the CPU 8 generates detected signals and a voice IC 9 replies beforehand selected words which are suitable for the reply to the inputted voice in a voice.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention identifies short words such as greetings in daily life, instructions, commands, and inquiries in simple words, and gives a voice reply or turns on an electric switch.
It relates to the development of a device that can respond by turning it on and off.

[0002]

2. Description of the Related Art Research on human speech recognition has been carried out in a wide range of fields such as translation of a speech language by a large computer and development of a voiceprint device used for discrimination of a specific individual by voice. At present, research that attempts to identify words with a short degree using a simple device is lagging behind.

[0003]

DISCLOSURE OF INVENTION Problems to be Solved by the Invention It is difficult to identify a voice because its vibration waveform changes complicatedly in an extremely short time.
Moreover, if the vibration changes subtly, the meaning of the word changes greatly, and even if the vibration waveform is different due to individual differences, the meaning is the same, so there are very complicated elements.

The present invention is intended to provide a device for identifying a voice having a complicated vibration waveform in a short time by the method described in detail below and immediately responding to the voice.

[0005]

[Means for Solving the Problems] The means for solving the problems will be described below with reference to the drawings. When a person examines the output voltage waveform of the microphone 1 when vowels a, a, u, e, o, and vowels are generated at a scale of 330 Hz, one example of the waveform is shown in FIG. As shown in the voice, it can be seen that it is made up of synthesized waveforms up to the fifth harmonic component. A and O, and I and U are not shown in the figure because they are very similar to each other, but the harmonic content of each sound is significantly different, so it can be easily distinguished. The scale of everyday conversation is 200-1000Hz
Because it is the band of 200 to 5000H
It is understood that it is sufficient to analyze the voice signal wave within the frequency band of z.

Japanese is thought to be composed of five vowels and some consonants. For example, the sound of "ka" is pronounced in the Roman alphabet by KA followed by A followed by A. It feels like you're being done. This is millisecond (ms)
When we analyze the waveform on the unit time axis, it seems that the sound of A is transformed into the sound of K during the period of more than 10 ms at the beginning of the rising of the sound, but most of it is after that. It is the sound of A. As is clear from pronouncing the car for a long time, the sound of A cannot be felt unless the sound of A continues for a certain length of time. From the above, the Japanese word “iroha”
It can be said that the 48 characters are made up of all 5 vowels, and these are made up of about 10 consonants and are transformed for a very short time in the initial stage to form 48 sounds. Therefore, it is possible to identify to a certain extent only by analyzing which vowels are included in a short word such as "Good morning, right now, etc."

The identified word is compared by the CPU 8 with the word registered in its memory 9. If there is a matching word in the registered words, the CPU 8 operates the voice IC 10 with this output. The voice IC 10 pronounces a word designated by the CPU 8 as a voice and sets it as a response to the input voice. This is the basic concept of the present invention.

[0009]

FUNCTION AND EXAMPLE Next, the constitution of the present invention, its example, and one example of experimental results will be described. The voice input is transformed into an electric signal (for example, a voice waveform signal shown in FIG. 1) by the microphone 1 and then amplified by the amplifier 2 to be further converted into an impedance-matched signal wave, which is distributed to each subsequent element. . Three
The input signal wave distributed to each of the 4, 5 and 5 filter circuits is discriminated into components for each frequency band of the filter circuit and then converted into digital signals by the subsequent analog-digital (A-D) converter 7 to output digital signals. Dl,
Obtain Db, Dh. The gate 6 controls each A-D converter 7 to operate in synchronization with the input voice.

Table 1 shows an example of this experimental data. The microphone 1 is a commercially available electromagnetic type, the amplifier 2 is a logarithmic amplifier, and the filter circuits 3, 4, 5 are crossovers. When the frequencies are 700 Hz and 1500 Hz and the attenuation characteristics are set to -18 db / oct, and the AD converter 7 uses the three commercially available frequency counters to embody the circuit configuration of FIG. Is the experimental value of. The scale of each word is approximately 250 Hz, and the duration of each is less than 1 second.

[0011]

[Table 1]

Although the count value for each filter is less than 200 in all, the difference in each word can be sufficiently discriminated. However, if the experiment is repeated for a plurality of men and women to obtain the values of Dl, Db, and Dh, the individual difference gradually increases and it becomes difficult to discriminate. When the accuracy of the high frequency band is increased as a strong logarithmic characteristic and the component ratios Db / Dl and Dh / Dl of each harmonic are calculated, this value has very little variation due to individual differences, and the accuracy of discrimination is significantly improved. To do.

The identified words, for example, the measured values of "Imaima" 90, 110, 70, 1.2, 0.8 are CPU8.
Is detected by comparison with that of the word pre-registered in the memory 9, and the address output is applied to the voice IC 9, so the voice IC 9 is the voice preset as this reply,
For example, the audio signal "Welcome back" is output. This output is uttered by the loudspeaker 13 as a voice having a sufficiently large volume and becomes a reply.

Registration of each count data and the like in the memory 9 is performed by the controller 12 as follows. FIG. 3 shows a simple 2-bit configuration as an example. Reference numeral 12 is a case body of the controller, a panel with numbers 12-1 to 12-4 is a keyboard, and each of the four keys is a CPU 8 and a voice IC by signal lines L1, L2, L3.
9 in parallel. The title is displayed for each key underneath the number of the key, but for the sake of convenience, this is for predetermining the relation between the memory 9 and the voice IC 10. Now press the key 12-1 and the signal line L
When 1 is turned on and number output 1 is built in beforehand, the word "Ohayoyozaizai" which is preferable as a reply to the word "Ohayoyo" that is written in the title is selected and this voice is selected. Set the voice IC so that can output. Next, when the experimenter inputs his voice "Ohayoyo", the CPU 8 inputs Dl, Db, Dh obtained by discriminating this voice and inputs Db / Dl, Dh /
Dl, Dh / Db, etc. are calculated and the respective numerical values are stored in the memory 9
It is stored in the designated address 01 of. Then, when the key 12-2 is pressed, the signal line L2 is turned on and the number output 2 is transmitted.
The voice IC 10 selects "Welcome" as a reply to the title "I'm home" of this key, and sets the voice IC 9 so that the voice of this word is output. Next, when the experimenter inputs "I'm home", the CPU 8 uses the key 12-
As in the case of 1, each value such as count data is stored in the memory 9
It is stored in the designated address 02 of. Below, 12-3, 12-
By performing the same operation as the number of 4 and keys, the data of each word by the voice peculiar to the person is registered in the designated address of the memory 9.

In the case of targeting an unspecified number of people, the memory 9 is a read-only memory (ROM), which is obtained as data common to each person and is Db / Dl, Dh / D.
If the values of l, Dh / Db, are recorded in the ROM 9, it is not necessary for each person to register. Also, D
In order to calculate the component ratios of b / Dl, Dh / Dl, etc., it is necessary that the counting bit of the AD converter 7 be 8 or more. The processing capacity of the CPU 8 and the capacity of the memory 9 can be reduced by reducing the number to 10.

[Effects of the Invention] Most of the words that people use everyday such as conversations, greetings, instructions, etc. are made up of at most 5 or 6 sounds, and their responses are also common sense. There are many. Even if you decide in advance that you will reply to the same words as “today” when you are called “today” or “return home” to the person who came back “just now”, it would be a bit unnatural Is not something. Therefore, if the number of words stored in the device of the present invention is set to about 16 bits in 4 bits, the effect of application is great in developing a doll that can talk with a person or a robot that responds to instructions and operates according to the instructions. Is.

[Brief description of drawings]

FIG. 1 shows voice signal waveforms of vowels a, a, d.

FIG. 2 is a configuration diagram showing the principle of an apparatus for identifying a voice and replying with a voice in response to the identification.

FIG. 3 is a configuration diagram of a controller for registering voice in advance.

[Explanation of symbols]

 1 Microphone 2 Magnifier 3, 4, 5, Filter for each frequency band 6 Gate circuit 7 A-D converter 8 Microcomputer 9 Memory 10 Voice IC 11 Loudspeaker 12 Controller

Claims (4)

[Claims] 1. A plurality of filter circuits (3, 4, 5) provided for each frequency band to identify spoken words.
After analyzing the input voice signal wave into the signal wave component for each band, the digital signal for each frequency band is accumulated for the duration of the voice by the same number of AD converters (7) as the filter circuit. The obtained count values Dl, Db, Dh and the correlation ratios Db / Dl, Dh / Dl, Dh / of the count values
A voice device which responds to a voice input characterized by identifying the voice from Db 2. The count value shown in claim 1 of a word input by voice and its correlation ratio (hereinafter referred to as count data etc.)
And the count data of words registered in the memory (9) in advance, the microcomputer (CPU)
If (8) is compared and verified, and if a match is determined, the CPU (8) transmits as an address number output of this memory (9), and an IC for voice generation (hereinafter referred to as voice IC) (10). A) A voice device which responds to a voice input which causes the voice programmed as the most suitable word to be returned as a response from the) to be pronounced as a reply or which controls another device with this address output. 3. Pre-registering count data etc. in the memory (9) and pre-programming the words contained in the voice IC (10) and the CPU (8) address output described in claim 2. A voice device that responds to voice input with a controller (12) 4. The registration operation of the memory (9) and the voice IC (10) so that the comparison test is performed using only the correlation ratio for each frequency band among the count data and the like, and the reply of claim 2 is uttered. Device for responding to voice input completed in the process of factory production with relation program between CPU and CPU (8)