JPS61130999A - Voice recognition equipment - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed Description of the Invention] [Field of Application of the Invention] The present invention relates to a device that recognizes human speech, and in particular, a device that compares speech information input by one person with pre-registered speech patterns. The present invention relates to an improvement of a speech recognition device equipped with a plurality of speech recognition means.

[Background of the invention]

For example, as disclosed in Japanese Unexamined Patent Publication No. 51-28701, the current speech recognition accuracy is significantly degraded for unspecified speakers, so it is necessary to register speech patterns in advance that have extracted features for each speaker. The mainstream method is to compare this with the voice input and recognize it based on the matching of those features.

By the way, there is a strong demand to use a speech recognition device with multiple speakers, and in this case, one speech recognition device can record the speech patterns for each word of multiple speakers (for example, 20 people). Memorize it in advance. Then, the words generated by the user (speaker) are compared with the stored speech pattern, and the word is recognized as the one word that corresponds to the speech pattern that is closest or within the expected error range.

However, 1. Memorizing many words from many speakers requires a large storage capacity, and recognizing matching speech patterns from among them requires a long recognition time. .

For this reason, a method has been proposed in which each speaker owns a cassette memory storing the speech patterns of each speaker, and the cassette is set when using the speech recognition device.

According to this method, although the total memory capacity is the same, there is an advantage that recognition accuracy and recognition speed are improved by matching only the voice pattern of a specific speaker.

However, this is inconvenient for the user, as the cassette memory or the like may be lost or stored.

[Purpose of the invention]

An object of the present invention is to provide a speech recognition device that does not impair recognition accuracy or recognition speed when multiple speech recognition devices are used by multiple speakers, and does not require removable memory. It is.

[Summary of the invention]

The present invention is characterized by: 1. means for identifying a speaker using any one of a plurality of speech recognition means; *
Storing the voice pattern for each i speaker) A common auxiliary storage means, and control for reading and storing the voice pattern corresponding to the identified speaker from the auxiliary storage means into the voice pattern storage means of the corresponding voice recognition means. It is necessary to provide means.

In a preferred embodiment of the present invention, the common auxiliary storage means contains (1) speech patterns for speaker identification (speaker change terms), and (2) speech patterns for tasks (command words and tasks). (terms, etc.) are memorized in advance. Immediately before each voice recognition means is used, a voice pattern for speaker identification is read out and stored in the voice pattern storage means in each voice recognition means.

In this state, for example, when Nitabe starts using the arbitrary voice recognition means, he utters a predetermined speaker change term that is encoded with a voice pattern for identifying the speaker, such as his name.

Accordingly, when a speaker is identified, the control means reads out a working speech pattern corresponding to the identified speaker from the common auxiliary storage means, and reads out a working speech pattern corresponding to the identified speaker from the speech pattern storage means in the corresponding speech recognition means. Store it in

Thereafter, voice recognition is performed by comparing only the voice pattern for work in one-on-one correspondence with the speaker and the voice input information of that speaker, and the necessary work is performed.

As a result, a plurality of voice recognition means can be used simultaneously by different speakers without requiring a removable memory, which may impair the recognition accuracy and recognition speed of each voice recognition means.

[Embodiments of the invention]

Hereinafter, an embodiment in which the present invention is applied to a voice input storage device will be described in detail. In this embodiment, audio guidance 9. This is a device that inputs and stores the worker's work and its results while providing answer back. Also,
It is also possible to freely register voice patterns that should be performed for the first time or for a new speaker.

FIG. 1 shows the configuration of an embodiment of a voice information input and recording device according to the present invention. In the figure, two voice recognition devices, one person and IB, are amplifiers 11A and 11B that amplify the voice signals of voice input micros A and 6B, A/D converters 12A and 12 that convert the voice signals into digital signals.
BS 14 voice pattern memories and 14B that store voice patterns in advance; and a voice recognition control circuit 13A that performs voice recognition by comparing input voice and voice patterns.
On the other hand, the voice output devices 2 and 2B are composed of synthesized voice memories 22A and 22B that store voices for outputting voices, and a synthesized voice memo IJ 22 according to the result of 'N voice recognition. A and 22B
Audio output control circuit 21A that selects and outputs the memory contents of
and a D/A converter 23A that converts the output signals of the 21B1 audio output control circuits 21A and 21B into analog signals.
And the 23B1 analog signal is amplified and the speaker ('!
It is composed of amplifiers 24A and 24B that generate answerback sounds from the earphones 7A and 7B.

The auxiliary storage device 8 stores a plurality of voice recognition bags NIA and IB.
It is commonly used for storing voice patterns to be stored in the respective voice pattern memories 14A and 14B.

The control circuit 3 also controls the voice recognition control circuits 13A and 13B of the voice recognition devices IA and IB to receive voice recognition results, and controls the voice output control circuits 21A and 21B of the voice output devices 2A and 2B. In this case, guidance and answerback sounds are output from speakers 7A and 7B, and voice patterns from voice pattern memories 4A and 14B of voice recognition devices IA and IB are stored in auxiliary storage device 8. Voice recognition device A and IB or one person or I
This is a control computer that displays (or prints out) the control status, voice recognition results, etc. on the display (or printer) 5. This control circuit 3 is controlled not only by voice but also by a keyboard 4.

Next, an example of the audio words used in one embodiment of the present invention is shown in the second example.
As shown in the figure.

Spoken words consist of speaker change terms for changing speakers (speech patterns for speaker identification), working words for performing tasks, and command words used for tasks (sound patterns for tasks). .

First, the voice pattern is registered by the speaker reading out spoken words in sequence using the Micro A or 6B, and the voice is transmitted to the amplifier 11A or 11B.
, A/D converter 12A or 12B1 voice recognition control circuit 13A and voice pattern memory 14 via 13B.
A or 14B. The voice patterns stored in the voice pattern memory 14Af or 14B are stored in the auxiliary storage device 8 with addresses assigned for each speaker.

Regarding the address allocation for storing voice words in the voice pattern memories 14A and 14B, command words and working words are stored at the same address common to all speakers, and speaker change words are stored at different addresses for each speaker. Then, when starting to use the speaker change mode or when a change command is issued, only the voice patterns of the speaker change terms of all speakers are stored in the voice pattern memory 14A or 14B, In the work mode after completion, the voice pattern of command words and work words of one speaker identified in the above alternation mode is recorded in the voice pattern memo 1714.
Move to A or 14B and enter data by voice.

Next, an embodiment of audio information input according to the present invention will be described using FIG. 3.

In response to the voice guidance "What's your name?" from the speaker 7A, A-taabe utters "A-taabe" from the micro A, and the voice recognition control circuit 13A of the speech recognition device IA stores it in the voice pattern memory 14A. Select the word “A” that matches the input voice from among the voice words for speaker change
The CPU 11 searches for the "fat part" and outputs the memory address or the corresponding code to the control circuit 3.

The control circuit 3 issues a command to the voice output control circuit 21A to obtain an answer back, in addition to displaying the words received as a message on the display 5 by inputting the voice word code. The audio output control circuit 21A outputs the audio data in the synthesized audio memory 22A in response to the answerback command from the control circuit 3, and causes the speaker 7A to utter "A fat section" via the D/A converter 23A1 and the amplifier 24A. Here, when A-taabe utters and inputs rOKJ from micro A, the voice recognition control circuit 13A of the voice recognition bag ff1A selects from among the words for speaker change in the voice pattern memory 14A.
The word rOKJ that matches the input voice is searched and its address or code is output to the control circuit 3. As a result, the control circuit 3 identifies that the speech is A-bold, reads out and stores the working audio pattern for A-bold that was stored in the auxiliary storage device 8 in the audio pattern memory 14A, and stores A-broad. At the same time as switching to the work mode shown in the thick section, the voice output device 2A is controlled to emit voice guidance from seven speakers asking, ``What about the work?''.

When A-tabe inputs “in stock” by voice, the voice recognition results in “
What's the product number? ” is returned from the speaker 7A, so for example, if it is recognized correctly, the answerback will be returned as l-1, 2, 3J. If you input rAJ by voice in response to the guidance "?", the voice recognition will return rAJ and an answerback.

As described above, A-taibe transfers his working voice pattern from the auxiliary storage device 8 to the speech recognition device IA using A-taibe's voice, and then checks the accuracy by comparing it only with his own working voice pattern. Using high-speed and high-speed recognition,
This means that the data "3 and storage at storage area A" has been input.

When A-tabe finishes his work, he says "takeover" and micro-A.
If you input from , it will switch from work mode to speaker change mode. That is, the control circuit 3 reads out the voice pattern for speaker change in the auxiliary storage device 8 and stores it in the voice pattern memo 17.
14 Store in A.

The above description has been made for the case where the A-thick section inputs audio from the micro A, but the case where the A-thick section inputs the audio from the micro B is exactly the same. In response to the voice guidance "What's your name?" from the speaker 7B, A-taabe utters "A-taabe" from the micro B. Then, the voice recognition control circuit (L3B) of the voice recognition device IB uses the voice pattern memo IJ.
14 Search out the word "A thick part" that matches the input from the audio words stored in B, and output its memory address ( ) or code to the control circuit 3.

The amplifier 2 of the audio output device 2B is controlled by the control circuit 3.
``A fat part'' is uttered from the speaker 7B via the speaker 4B. Here, when A fat part speaks and inputs "OK" from micro B, the voice recognition control circuit 13 of the voice recognition device IB
According to BK, the voice word "OK" that matches the input voice is searched from among the words in the registered voice memory 14B, and its address or code is output to the control circuit 3. As a result, the control circuit 3 transfers the audio pattern for the A-thick section stored in the auxiliary storage device 8 to the audio pattern memory 14B.
At the same time, the voice output device 2B is controlled and the speaker 7B asks, ``What about your work?''
”, says the voice guidance. 1 It works in exactly the same way as when inputting voice from Micro A.

Next time, when B thick inputs "B thick" from micro A (or 6B), as a result of voice recognition, the voice pattern memory 14 A (or 14B) has B thick from auxiliary storage device 8. The work voice pattern is entered, and the B bold section can now input voice data.

Thereafter, in the same way, the voice patterns for changing speakers and for working with multiple speakers stored in one set of auxiliary storage device 8 are derived to the multiple voice recognition devices IA and IB, and the voice patterns are freely spoken. Ability to change personnel and enter data. The voice pattern may be registered from one voice recognition device and transferred to another voice recognition device via an auxiliary storage device, or may be registered from each voice recognition device.

Here, the auxiliary storage device 8 may be an integrated circuit RAM or ROM, or may be a bubble cassette, cassette tape, floppy disk, or the like. However, in order to freely register the voice patterns of a new strong speaker, a storage means other than the ROM is used.

Auxiliary storage device 8 and registered voice memo IJ 14 A or 1
Reading and storing of the voice pattern between 4B may be performed from the keyboard 4 in addition to voice input. Furthermore, even if one voice recognition result is displayed on the display 5 and the two voice output devices and 2B are omitted, a plurality of speakers can alternately input voice information from a plurality of voice recognition devices.

@Figure 4 shows the configuration of another embodiment of the audio information input device according to the present invention, and the same reference numerals as in Figure 1 have the same functions. In the figure, radio mobile stations 30A and 30B receive input audio from micros A and 6B from transmitters 31A and 31B1 which emit radio waves from antennas 33A and 33B, receive radio waves from antennas 33A and 33B, and transmit audio from speakers 7A and 7B. Receivers 32A and 32B that generate guidance and answerback
It is made up of. Radio fixed station 20A and 2
0B is a receiver 21A and 21B1 which receives radio waves from radio mobile stations 30A and 30B via antennas 23A and 23B and inputs them to the voice input/output device 10A and the voice recognition devices IA and IB of IOB; voice input/output device 10;
The audio output device A and IOB consists of transmitters 22A and 22B that emit radio waves from the output audio of two people and 2B to receivers 32A and 32B of radio mobile stations 30A and 30B via antennas 23A and 23B. . The voice pattern is registered by the speaker reading out voice words sequentially from Micro A or 6B. The voice input from Micro A or 6B is emitted as a radio wave from transmitter 31A or 31B of radio mobile station 30A or 30B via antenna 33A or 33B. This radio wave is from antenna 23A or 2.3B
Receiver 21 of radio fixed station 20A or 20B via
At or 21B receives the speech recognition device IA or IB.
registered in the registered voice memory. The voice patterns registered in the registered voice memory are stored in the auxiliary storage device 8 with addresses assigned for each speaker. Further, the voice pattern stored in the auxiliary storage device 8 is transferred to the voice pattern memory of the voice recognition result IA or IB by operating the keyboard 4 or inputting voice to the voice recognition device IA or IB. When the speaker's voice data is input from Micro A or 6B, it is received by the receiver 21A or 21B of the radio fixed station 20A or 20B through one wave from the transmitter 31A or 31B of the radio mobile station 30A or 30B. and input into a voice recognition device or IB. The answer back of the voice recognition result is emitted from the voice output device 2 or 2B, and is emitted as a radio wave by the transmitter 22A or 22B. This radio wave is receiver 32A
or received by 323 and speaker 7A or 7B
is uttered from. When the speaker inputs data by voice from the micro A or 6B, an answerback or guidance is emitted from the speaker 7A or 7B, and the speaker inputs data by voice while opening the speaker.

In the embodiments described above, once a voice pattern is registered in one set of voice recognition devices, the voice pattern can be registered in another voice recognition device using the auxiliary storage device without uttering a voice.

In the above embodiment, the voice pattern for speaker change is also
It is registered in the common auxiliary storage device R8, and is stored in the voice pattern memo 1J14A, 14B in each voice recognition device IA, IB only in the talk and person change modes. However, the voice pattern for speaker change is always stored in each voice pattern memory 4A.

14B may be stored.

In this case, a pattern corresponding to the identified speaker among the working voice patterns is selectively stored in another address of each voice pattern memory.

In addition, although we have described a method that uses voice recognition means to identify the speaker, it is also possible to input a code for each speaker using a keyboard or any other input means, and in this case, the control The speaker's mouth can be easily identified.

- [Effects of the Invention] According to the present invention, a plurality of speech recognition devices can be freely used by a plurality of speakers, and the speech pattern storage means of the speech recognition device corresponds to the speech pattern of the corresponding speaker by identifying the speaker. By storing the information in the memory, it is possible to provide a speech recognition device with excellent recognition accuracy.

[Brief explanation of the drawing]

FIG. 1 is a system configuration diagram showing the configuration of a voice information input device according to an embodiment of the present invention, and FIG. 2 is an example of a voice word used in the voice information input device shown in FIG. 1 and its memory. 3 is a diagram showing the contents, FIG. 3 is a procedure diagram of voice information input showing an example of speaker change and work, and FIG. 4 is a diagram showing the configuration of another voice information input device showing another embodiment of the present invention. It is a system configuration diagram. IA, IB... Voice recognition means, 2A, 2B... Voice output device, 3... Control means, 4... Keyboard, 5
... Display or printer, 6A, 6B... Microphone, 7A. 7B...Speaker, 8...Auxiliary storage means, IOA. 10B... Audio input/output device, 20A, 20B... Radio fixed station, 30A, 30B... Radio mobile station,
2nd part 30th

Claims (1)

[Scope of Claims] 1. A speech recognition device that matches speech input information with a pre-registered speech pattern, including a plurality of microphones, each of which has a speech pattern storage means provided corresponding to each microphone, and a speech recognition device that matches speech input information with a pre-registered speech pattern. a plurality of voice recognition means that recognize the voice by comparing the voice input from the microphone with the voice pattern; a common auxiliary storage means that stores the voice pattern of each of the plurality of speakers; and any of the microphones. means for identifying a specific speaker, and a voice pattern of the specific speaker among the voice patterns for each of a plurality of speakers stored in the auxiliary storage means, to the voice pattern storage means of the corresponding voice recognition means; A speech recognition device comprising a control means for reading and storing information. 2. The speaker identification means identifies the speaker by comparing the speech pattern for speaker identification stored in the speech pattern storage means of the speech recognition means with the speech input by the speaker. The speech recognition device according to item 1, which is a means. 3. The common auxiliary storage means also stores the voice pattern for speaker identification, and the control means stores the voice pattern for the specific speaker's work according to the speaker identification result by the voice recognition means. 3. The speech recognition device according to claim 2, wherein the speech pattern is read from the auxiliary storage means and stored in the speech pattern storage means. 4. The speech recognition device according to item 1, wherein the speaker identification means includes means for inputting a code for each speaker.