JPS5855993A - Voice data input unit - 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 method and apparatus for reliably and correctly recognizing and inputting voice codes in voice keys, voice remote controls, and the like.

The problem with conventional systems that recognize voices and turn them on is the problem of erroneous recognition. Generally, these systems recognize a specific keyword when the user utters it, but if the recognition is made by mistake, the user can re-read the keyword from the beginning without understanding where the data input was inappropriate. The object of the present invention is to eliminate the above-mentioned drawbacks of the prior art and to provide an audio data input device for reliably inputting audio data without any unrelated information. There is something to do.

The feature of the present invention is that audio data consisting of a plurality of elements is inputted one element at a time in synchronization with a timing signal on the device side, and each time one element of data is inputted, the audio data is inputted! By notifying the user of the results determined at 1m@road, the voice data can be reliably input manually.

FIG. 1 shows a voice data human power device according to the present invention.

FIG. 2 is a block diagram of an embodiment of the present invention.

16 is a timing signal generation circuit, and in this embodiment, the signal is a "beep" sound, for example. The signal 012 is generated from a speaker 120 via an amplifier 18, and analyzes and recognizes the elements of audio data input manually from a microphone 19. This is a speech analysis 11g spear circuit. Reference numeral 13 denotes a speech synthesis circuit for notifying the contents recognized by the speech analysis and recognition circuit 12 by voice.

For example, the voice data is a number 5512 (san, go, etc.).

Figure 2 shows the system 70- when manually inputting audio data consisting of these four numerical elements, with the left side showing the work on the user's side and the right side showing the work on the device side. shows. In the voice input task 21, the four voice elements of 5 (san), 5 (go), 1 (ichi), and 2 (ni) are input in voice.The human input data is judged in the voice recognition routine 25. , the judgment results are voice-combined in the voice synthesis routine 26 and echoed to the user. ri22
be done. The user listens to the contents of the echo locust and makes a judgment. If the judgment is correct, go back to step 21 and enter the same data manually.

Correctly determined data is stored in the RAM 123 by the data storage routine 27.

FIG. 5 shows the tying for data input. In the data input of 5512 (san, go, one, ni), an example is shown in which the manual input of data 1 (one) is not performed correctly at the first time. The pulse signal sound of the Thai Deng, the circled Fumi is the voice data of the voice used, and the not circled Kana are the echo pack voice synthesized sounds determined by the device.In this example, 5.5°2 is This shows a case where 1 is correctly inputted once tp- and then correctly entered manually the second time.After the first input of voice data 34 of 1, the device inputs its contents t-2 (to) The user inputs the voice data 157 again after realizing that the recognition result is incorrect, but before that, as shown at 56 in the figure, the user places a silent blank 1''').

This blank means that the content determined just before is wrong, and the process is started again manually from the voice data. In this way, manually input 1 (ichi) 57 again, and if the result is correctly judged as 1 and the synthesized voice 58 is output, after use, manually input the next voice data 2 (ni) 59. After completing a series of data input operations, the end of data input is indicated at 5 in the diagram.
The device can be informed by providing two or more blank devices as shown at 311. 1st v! In i, MP
U (microprocessor unit)) 11 is ROM122
The system is controlled according to the system software stored in the interface 124, and the content of the recognized voice data or data accompanying it is transferred to the interface 124 as necessary.
Output to the outside via f. The MPU includes an interface 124 that can be configured using, for example, a 4-bit microprocessor HMC840 series manufactured by Hitachi.
In addition to the above functions, the function also controls connections with external devices.

For example, a distance calculation method based on PARCOR analysis is used for speech recognition.

The algorithms and methods of PARCO & analysis are well known, and the physical parameters (PARCOR coefficients,
Pitch information, amplitude information, etc.) are calculated. In this embodiment, audio data is composed of 10 elements, for example from 0 to 9, and these data are
(zero), 1 (ichi), 2 (ni), 5 (san) * 4 (
It is given by t, ), S (go), 6 (ro<), 7 (Lchi), 8 (stingy), and 9 (ku). The characteristics of these voices are PAR
It is expressed as a vector in an n-dimensional space composed of n physical parameters such as COR coefficients, and the above 10 audio features are stored in the ROM 122 as n-dimensional vector data.
7 arrangement data (zero, one, two, three) are pre-stored in the computer or recorded in advance before the audio data is manually processed.

The physical parameters are calculated by the analysis recognition circuit 12 by manually inputting the 10 voices (shi, go, roku, shi, hachi, ku) from the microphone 19 and stored in the RAM 125.

Therefore, 3 (zan), 5 (go), 1 (ichi).

2 (ni) When the data voice is manually recognized, the features of each human-generated voice are analyzed by the 9wt-circuit 12, and the obtained n-dimensional physical parameter vector -ak (k= 0 * 1 *...

9) and the vector b (J
-Q, 1, . . . , 9) and find which one of 0 to 9 it is closest to. Calculation is done by MPU
11. The specific calculation is as follows.
If the input audio data has n parameters (
RARCOR coefficient, etc.) is “J” (k = Os 1
* ""9* j-1, 2, ..., n), similar parameters of the prepared reference audio data bjj (
If j=O−1e ”·, 9°1.2...., n), the distance between the manually generated data and the reference data is 111d.
k and l are expressed as follows.

The suffix, j, represents the physical parameter of nf.t) The f suffix, α amount, is a coefficient for normalizing or weighting the eco physical parameter.

The MPU+1 calculates the formula (1) for all 1j of the input audio data with respect to the analysis result tk of =, and takes the smallest value of I=1. is the recognition result. In other words, +) = j.

When , the audio data has been input correctly.

The voice synthesis circuit 15 synthesizes and utters the echo back voice according to the above calculation result j-1O. The recognition result is
It is expressed as 4-bit data as shown in Figure 4,
The data [address data (described later) originally stored in the ROM 122 for voice synthesis is data bus 1]
25 and is sent to the speech synthesis circuit 15.

The speech synthesis circuit 13 includes a speech synthesis section 14 and a speech memory section 15.
It consists of The voice memory unit 15 stores PARCOR coefficients of voices to be synthesized (zero, one, . . . , <),
It stores data such as pitch information and amplitude information, and uses, for example, HI58882 manufactured by Hitachi. The speech synthesis unit 14 receives from the MPU++ the designation of the start address in the speech memory in which data necessary for speech synthesis is stored, and based on this, reads the data from the speech block memory 15 and synthesizes the speech signal. For example, HD38880, which is an LSI for speech synthesis manufactured by Hitachi, Ltd., is used.

FIG. 5 shows the starting address (4 digits in hexadecimal) of a block in which data for displaying the contents of the audio memory is stored. The synthesized voice is sent to the amplifier 18
The voice is emitted from the speaker 120 via the .

As shown in the embodiments above, by using the human voice input device according to the present invention, it is possible to input voice data reliably without making any mistakes.In this embodiment, input voice data is recognized.

Although this article describes a method of echoing back the recognized results using speech synthesis, other methods such as C
Even if you use the RT display for tigers, the effect remains the same.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing the configuration of the voice data input device according to the present invention, FIG. 2 is a diagram showing the operation flow of the voice data input device according to the present invention, and FIG. 3 is a diagram showing the manual timing of police voice data. Figure 4 is a diagram showing the BIC F pattern of data when human-generated voice is analyzed and the recognition results are converted into data, and Figure 5 is voice data R.
It is a figure which shows the address of OM. 12...Speech analysis recognition circuit, 15...Speech synthesis circuit 16...Timing signal generation circuit, 21...Speech input work, 22...Echo back, 26...Speech synthesis routine. Representative Patent Attorney Susuki 1) Yuki Toshi 4. -2, Sai1 Zug Z Figure 2 Sai + Figure MSF3 1-3B Sai, De eyes

Claims (1)

[Claims] 1. A voice recognition circuit for recognizing voice data composed of 11 or more elements, a means for echoing back the recognized content to the user according to the result of the discussion, and the voice data. It is equipped with a circuit for generating a signal for timing the input of the input and the generation of echo back by voice synthesis, and a means for controlling the above circuit, and the generated sound of the voice data is determined based on the content of the echo back. A voice data human-powered device, characterized in that if the content is different from the manually-generated data, the data can be manually input again, and the voice data can be reliably inputted by repeating the above procedure.