JPS587697A - Voice recognizing system - 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 The present invention relates to a speech recognition system, and specifically to a method for speech recognition and response thereto that improves the efficiency of recognition processing.

Currently, the key is the input method for office computer♂1-evening? - Serial printers have been established as output means for touch-in, sensor output, etc., and floppy disks, magnetic memories, etc. have been established as storage file means.

By the way, due to technical innovations in voice synthesis and voice GW&, voice input/output means have been provided to office combinations and computers. Voice input depends largely on recognition technology.Normally, single W! The speech recognition device that performs recognition uses a dictionary (standard) of n words registered in the internal dictionary area.
is stored and the speech is recognized by referring to the dictionary.

V! processed in a conventional speech recognition device. An operational flowchart of the recognition command is shown in FIG. The outline of the operation is listed below.

(1) Input voice.

(2) Calculate the distance between the input speech and words that are constituent elements of the dictionary.

(3) The above (
Repeat operation 2).

(4) When the short distance is less than or equal to a predetermined value, it is assumed that the word has been recognized, and the name of the word is notified to the host computer (CPU).

By the way, according to the above subordinate example, in (2) and (3), distances are calculated for all words existing in the dictionary area, so only words belonging to a specific range (operator control commands, numbers, etc.) are calculated. If you do gw&, extra time will be spent, and the system will be burdened unnecessarily during that time.

In addition, if a word outside the required range is input to the system, the word name is notified to the CPH, and then the CPU
@ requires meaningless processing until it is determined that reception is impossible. During this time, the CPU is overloaded (increased burden)
Of course, the amount of time spent on this was not to be ignored.

The present invention has been made based on the above circumstances, and the dictionary name to be referred to during speech recognition is set as 79 rammeters along with normal dictionary information, and uR is set as a response to recognition.
By transmitting the word name and the dictionary name containing the word to the requested device, the load on the CPU is reduced.
Furthermore, it is an object of the present invention to provide a speech recognition method that speeds up processing.

Hereinafter, the present invention will be explained in detail using the drawings.

FIG. 2 is a block diagram showing an example of the configuration of a computer system to which the present invention is applied. In the figure, 1 is a main memory in which programs or various data are stored. 2 is a central processing unit (hereinafter referred to as CPU) that controls the entire system according to the program stored in the main memory 1; J is an input/output control device; −
Input/output devices such as a card, a CRT display, and a serial printer (none of which are shown) are connected. 4 is a speech recognition device in which the present invention is realized, and functionally it is a speech recognition device 1j! ! It consists of a texture control section 5 and voice dictionaries P and F. The speech dictionary Δ, fad stores some standard Δ turns of registered word groups for speech recognition, and has an address map as an extension of the main memory 1.

Details of the voice recognition control section 5 are shown in FIG. 2, where r is a file controller. A fixed magnetic disk device (8; external 7-isle memory) is connected to the file controller 7, and the fixed magnetic disk device 8 is connected to a voice language 4# (speech recognition (standard patterns of registered word groups) are stored in advance. 9 is a system path consisting of a plurality of lines for address/data control. Each of the above devices 1. :l, 3. j.

r is commonly connected to the system path 9 and constitutes a computer system.

FIG. 3 is a professional diagram showing the internal configuration of the speech recognition device in FIG. 2. In the figure, 11 is a path interface unit provided when connecting the speech recognition device to the computer system. This path interface,
-7 Ace Uni.

The controller 11 imports the Δ parameter accompanying the gR command issued from the outside (CPU 2) into the voice GM device, and as a result of the recognition, the word name and the accompanying dictionary name are sent to the CPU 2.
It is a hardware mechanism (data path) that controls the so-called interface part with the outside, which notifies the
2 is Mike U Grosse, 11 is ROM, 14 i
It is RAM.

BON IJ stores firmware routines to control this equipment, and includes microprocessors,
The processing flowchart of the firmware routine used in the present invention is shown in FIG. It is used for work when normalizing the characteristics of voice input data, or as a storage area for normalized data. 1j is the distance calculation method. This logic is an I-ware that automatically calculates (comparison calculation) the degree of similarity between a standard pattern and a voice-input Kanodata pattern, and it is an I-ware that automatically calculates (comparison calculation) the degree of similarity between a standard pattern and a voice-input pattern. This is the logic that calculates the absolute value of the difference between the spoken person's cursive turns. When the obtained value is less than or equal to a certain added value, the word is recognized and assumed to be true, and the CPU 2 is notified of the name of the word and the name of the dictionary to which the word belongs.

16 is an audio input section. The audio input unit 16 analyzes the frequency of an audio analog signal input from a microphone (not shown), samples and holds the peak value of each frequency band at each firmware operation time, and outputs an 8-bit,
The path interface unit y) 11, ROM 12*R, which has the function of converting into a digital value
AM13-distance calculation unit 15. The audio input section 16 is the microphone.

It is commonly connected to the internal path 17 of the bus 12.

6 is the speech dictionary buffer shown in FIG. 2, which is connected to the system path 9 via the path interface unit 18. This speech dictionary buffer 6 includes a dictionary area DA where a plurality of dictionaries are stored and a dictionary information area (
DIA). The number of words in the dictionary is set in this dictionary information area.

Figure 4 (,) Cb) are parameters accompanying the voice recognition command used in the present invention, (a) tf command + parameters passed from CPUJ to voice recognition device l at the start, 佽) are recognition After processing, the difference between the voice recognition device and the CPU
O indicates 749 meters passed for x.The operation of the present invention will now be described in detail using FIGS. 2 to 5.

First, the CPUJ issues a voice command to the voice recognition device 4 via the system bus 9, and also sends the parameters shown in FIG. 4(,).

Upon receiving the command, the voice recognition device 1 starts up its own microgrocer 12 and starts processing from the beginning of the firmware routine shown in FIG. Of course, the firmware routine iROMJjlk shown in FIG.
Assume that it is stored in 3.

Hereinafter, the flow of speech recognition operations inside the speech recognition device 1 using the microlog tube will be enumerated sequentially.

(1) According to the former shown in Figure 4 (-), V! The number of dictionaries and dictionary names to be referred to at the time of recognition are read via the path interface unit 11.

(2) Receive the input voice at the voice input unit 16 and capture the input voice according to the above operation.

(3) According to the dictionary information of the dictionary name read in (1),
The distance between a certain word stored in the dictionary buffer C and the voice captured in the above (2) is compared and calculated. The calculation here is performed by the distance calculation logic 1j according to the procedure described above.

In the embodiment of the present invention, hardware is used to speed up the distance calculation, but of course a microgross processor 12 may also be used instead of this function.

(4) Repeat the above distance calculation for all words in one specified dictionary.

(5) The above (
3) #Repeat the operation in (4).

(6) If the minimum distance obtained by distance calculation is less than or equal to a predetermined value, it is assumed that *g* has been performed, and the dictionary name and word name that include that word are shown in Figure 4 (b). 7 Ohm, according to T, pass Inter 7 A Suuni, T 11 to 9
and sends it to CPU 2.

As explained above, the present invention is expanded to set the dictionary name to be referred to during speech recognition as a parameter together with normal dictionary information, and to transfer the recognized word name and the dictionary name containing the word to the requesting device as a response to the recognition. As a result, meaningless processing is omitted, the processing time required for recognition is reduced, and the CP
By adopting the present invention, which reduces the burden on U, an efficient speech recognition system can be realized.

[Brief explanation of the drawing]

Figure 1 shows the lli processed in a conventional speech recognition device.
! FIG. 2 is a block diagram showing an example of the configuration of a computer system to which the present invention is applied; FIG. 3 is a block diagram showing the internal configuration of the speech recognition device in FIG. 2; 4(a) and 4(b) are diagrams showing parameters associated with voice recognition commands used in the present invention, and FIG. 5 is a firmware flowchart schematically showing the operation of the present invention. 2...CPU, 4-speech recognition device, 6-speech dictionary pa, fa, 11-Δsinterf, -suyu=. 12.-Microprocessor, J J-ROM. 15-Distance Calculation Rosic.

Claims (1)

[Claims] In a speech input computer system that stores a plurality of dictionaries in a storage device that can be referenced by a speech recognition device, recognizes input speech tg using the dictionaries, and treats recognized words as input data,
The speech recognition device includes a first means for receiving input speech, and performs a computation to compare registered words, which are constituent elements of the dictionary, with the input speech according to the dictionary name and dictionary information obtained from the outside and referred to during speech recognition. and a second means to perform Kg discrimination based on the value obtained by the second means, and when it is within a certain range, recognize the word name and the dictionary name containing the word. A voice recognition method characterized by notifying an external device that has received a request.