JPS5876890A - 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 The present invention relates to a voice recognition device, and particularly to a terminal capable of communicating with a center machine via a telephone line using a human voice. The present invention relates to a speech recognition device suitable for.

In a conventional voice recognition terminal via a telephone line (for example, SR series voice recognition device (source: large project pattern information processing system lecture paper results, pages 135 to 143)), a voice recognition device is installed in the center machine, The voice uttered by a human is put on the telephone line, and the center side
It takes the form of recognition, and has the following drawbacks.

■ S/N due to noise (crosstalk) in each episode used, signal distortion in the line, and differences in the line that is switched each time it is used.
Due to fluctuations in the number, the quality of voice input to the voice recognition device at the center is poor, and rounding to achieve sufficient recognition performance requires creation of standard patterns and other software processing, which increases the price of rounding. ing.

■ In order for the center machine to recognize the person using the telephone, it is essential that the speaker is unspecified.
Since it is necessary to use m technology, the 11ffi rating is high.

Therefore, the present invention solves the above-mentioned problems with known speech recognition devices, namely, the influence of the voice that exists when recognition is performed on the center machine side through a telephone and the unspecified speaker.
! ! The aim is to solve both problems in knowledge.

The present invention was developed as a result of studying the above-mentioned problems that are unavoidable when installing a voice recognition device in a center machine.
This was born out of consideration of the progress of the 1W technique2 and the method of system operation, and in Bihon's opinion, a non-gynecological time axis collation unit (absorbing the IIF contraction of the input leech voice) is installed on the terminal side. (Pattern matching is performed.) An ultra-compact voice recognition system mainly based on LSi is installed, voice input is performed on the terminal side, that is, a telephone, and further recognition J! By recognizing the input by fit, the i1g iron connection code is sent to the center til
The feature is that the configuration is such that it is possible to transfer data to 1 and t.

Hereinafter, the present invention will be described in detail based on Examples.

FIG. 1(A) shows a block diagram of a first embodiment of the present invention. In FIG. 1, data is transferred between a center machine 1 and a terminal 2 via a line 3. It is assumed that line 3 is a digital line. /The movement of the stem is 1 stuffing machine 20
After receiving all the voice inputs, the cocoon processing unit 21 changes the r analog signal At to the digital signal B, and the voice recognition unit 22 converts it into a digital signal B.
Verify the internal standard pattern and find the best match! IIl & result.

By transmitting the recognition result Ct to the line 3 through the communication interface 24, the recognition result C (coded) corresponding to the internal nature of the voice uttered by the speaker is obtained from all center machines l.
can be sent to. For example, if the present invention is used in systems such as Shinkansen seat reservations and product ordering systems, which conventionally required keyboard input using a butsuhon, voice input can be performed and voice recognition can be performed on the terminal side, so it is possible to Difficulties in recognition can be avoided.

FIG. 2 shows an example of installing the voice recognition section 22 in a telephone.

This corresponds to the embodiment shown in FIG. 1(A).

LSI'f! The progress of i-techniques is very impressive, and the number of nests is only increasing by 2 per year.Even at present, the WRFT technology included in the speech recognition unit 22, for example, features extraction, distance calculation, etc., are each implemented on one-chip LSI. It is possible to create a hybrid LS I'! on a ceramic substrate. By using the I-configuration, it is easy to realize the audio-m section on one ceramic substrate.

In FIG. 2(A), a feature extraction chip 221, a distance calculation chip 222, and a nonlinear matching chip 22 are mounted on the ceramic substrate 4.
3. A control chip 231t is mounted and interconnected with wiring 41 to realize a portion including the voice recognition section 22 and control section 23 shown in FIG. 1(A).

Therefore, as shown in FIG. 2(B), it is possible to realize the entire system shown in FIG. Fully equipped with LSI7゜8...
Hybrid L8I mounted on the flexible lamic board 4
6 will also be installed. Here, the preprocessing section 21 of the present invention is an analog circuit or the like, and it is avoided to use the same hybrid LSI due to the technical difficulty thereof.

The power to the telephone 20 may be used to supply power to the circuit.

Although the details of the speech recognition section 22 are not shown in FIG. 1, it has internal sections as shown in FIG. 2(A), and also has a memo 1 for storing standard patterns. (For the configuration of the voice recognition device, please refer to Ogan No. 54-83773.) 'In voice input via ML speech synthesis, there are no specific speakers who may be using telephones, so Since a speaker certification course is required, a pattern for unspecified speakers is created by using FLU M k in the memory for the standard pattern kakukasu.
Set.

Next, a second example will be described.

This is the system shown in No. 11M+ (B), and is characterized by the addition of speech synthesis s and the provision of a function to confirm the recognition result by its output.

In the first embodiment, although it is not shown in the drawings, in accordance with the voice guidance sent from the center machine, the first person Kawaba enters a tedious input into each part by voice, so before transferring it to the center machine, the tgmb If the confirmation cannot be made, the confirmation voice will be sent again by the central machine and the overhead will be large.

In the example of &42, the output of the back voice synthesizer 25 is 1LH1! Since the M effect can be ensured and the force can be transferred, it can be reduced to a negative percentage on the center machine side.

Audio wise, "fJ!i, 825, even though it has a bow shape, is made into an LSI and is easy to miniaturize, so there is no problem with the aesthetics shown in Figure 2.

In Fig. 2 (1B), the voice climbing auxiliary memory 10 is connected to the voice recognition section 22 via the lead set 9 and the Hirayama assistant memory interface 2b, but in this case, the telephone line of the U-axis station is connected to the voice recognition unit 22. The purpose is to make it possible to receive praise and praise for the person who spoke, and for the time being, there will be a small number of salaries (Sakuame, et al. 100 [
In this system, it becomes a southern effect. If Mfllfi (inffc) is installed on the center machine side, it will be difficult to register a standard pattern and to be a reasonable speaker. This is because the memory must be equipped with a light on the center machine side, and when creating a pattern of vertical words, the eJ ability to illuminate the center machine is desirable.Therefore, the center machine 9411 Approximately pupil device i
1'? If you are going to write a story, it is essential to use the futokukenwa name anshin.

In the present invention, the telephone 20 is provided with an external storage memory 10, so that the user can use his/her own electronic device to record the Kusunoki pattern on the terminal side. Since it can be registered, the overhead of the center machine is reduced. As mentioned above, in the γ application where a small amount of money is paid, the auxiliary memory 10 for registering the sound may be a simple auxiliary memory such as a magnetic cart.
All you have to do is replace each historical pot.

In order for the speaker to register the standard pattern, as shown in Figure 2, a lead wire 9 and an auxiliary memory interface 26 are required.
Through the pre-processing unit 21, it is converted into a digital '@ number,
It is guided to the speech recognition section 22.

In a feature extraction step 221 inside the speech recognition unit 22, the percentage features that are a-colored in the speech are extracted δ. % image extraction chip 22
1 is composed of, for example, a filter puncture, and at each point in time of the input voice, outputs a percentage parameter corresponding to the voice of 7.

Therefore, the voice voiced by the speaker eventually becomes a feature parameter output from the image extraction chip 221, and is stored in the auxiliary storage memory and becomes a standard pattern for the corresponding speaker.

In accordance with another embodiment of the present invention, either the sound leg synthesis unit 25 in the 11th station (1-1) or the combination of the auxiliary memory 10 for voice registration and the auxiliary memory interface 2by is configured to have a configuration in which there is no difference in length. It is also possible.

When emitting the sound of the center machine, you can use a normal dial or push button, or as an example of encoding the audio magazine RWIi, you can send out the dial output or the multi-frequency output of Punyuhon. A voice dialing method may also be used.

In another example of the H method, the data transfer cycle/11111 with the center machine may be formed by determining the transfer rule (protocol) 7 with the center machine using the recognition result code. As explained above, it is possible to recognize the error on the terminal side and transfer the entire code to the center machine, so there is no influence on the line and Ansen 1
It is possible to realize a very sophisticated system.

[Brief explanation of the drawing]

FIG. 1 shows a configuration diagram of one embodiment of Honjamei, and FIG. 2 shows an implementation diagram of one embodiment of Motojamei. Representative Patent Attorney Hajiyuki Hachita...=B"-;-13
``Te-・4

Claims (1)

[Claims] A preprocessing means for digitizing the output signal corresponding to the input audio of the 4-phone machine, and a preprocessing j! A voice recognition means that extracts features from the received signal and performs two inquiries with the standard pattern, and a non-code generation means that does not generate a code signal indicating the recognition result are placed on the telephone side, and the code signal is transmitted to the telephone line. Transferring to the center machine through the t-featured audio IIgA!
! Device.