JPS593494A - Voice input 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 (1) Technical Field of the Invention The present invention relates to a voice recognition device, and more particularly, to a voice input recognition device that switches the stored characteristics of a special number by voice input.

(2) Background of the technology Most speech recognition devices compare the input voice feature data with various pre-stored voice feature data, and select the feature data that is closest to the input voice feature data. The voice is recognized as the input voice. That is, features are extracted and pattern matching is performed.

Most of these features use formants in speech, that is, characteristics obtained by converting speech into a spectrum. Further, there are cases where pattern matching is performed on the formants as they are, and cases where the first formant frequency, second formant frequency, etc. are extracted and pattern matching is performed using those values.

(3) Prior art and problems Speech recognition involves speaker-independent recognition, which recognizes the input from an unspecified number of speakers by comparing it with a common standard pattern, and the speaker-independent recognition, which recognizes the input of an unspecified number of speakers by comparing it with a common standard pattern. There is specific speaker recognition, which recognizes input by comparing it with one's own registered patterns. The former method does not require registration work, but it is difficult to obtain a high recognition rate, and the latter method can obtain a high recognition rate, but requires registration work. Specific speaker recognition, which generally provides a high recognition rate, is often used, but pattern registration involves complicated key operations.

FIG. 1 shows a block diagram of the above-mentioned specific speaker recognition.

When registering or modifying a pattern, the input audio signal S is subjected to feature extraction in the audio input section 1, and is then converted into a registered pattern via the audio pattern registration/correction section 2, which performs initial pattern registration and modification of the pattern due to aging, etc. It is stored in the storage unit 3.

The above-mentioned registration or modification is performed by the control section 5 controlling the voice pattern registration/correction section 2 based on key operation data inputted from the operation section 4.

At the time of voice recognition, by inputting a user ID using key operation data from the operation unit 4 prior to input, the control unit 5 selects a registered pattern corresponding to the input 10 from the registered pattern storage unit 3, and performs verification during recognition. Data. The recognition section compares and matches the speech pattern obtained from the speech input section 1 and outputs the result to the control section 5. The above-described comparison and matching in the recognition unit performs recognition on the assumption that the pattern with the highest degree of similarity is the same voice. Further, the display unit 3 is used to monitor key input and voice input from operation operations, or to display messages from the system.

As is clear from the above description, the initial voice pattern registration work and the user's instruction to select the corresponding registered pattern are accompanied by key operations.

That is, the conventional voice input device requires a key operation unit 4 for performing key operations in an initial state and for specifying a registered pattern when the power is turned on.

For this reason, the conventional method does not free users from cumbersome key operations and does not take advantage of the advantages of voice input devices.

(4) Purpose of the Invention The present invention solves the above-mentioned problems, and its purpose is to provide a voice input recognition device that eliminates key operations by switching the stored characteristics by voice input. be.

(5) Structure of the Invention The present invention is characterized by a voice input means for extracting voice characteristics, a first storage means storing a standard voice pattern registration record, and a voice pattern registration record of a specific speaker. a second storage means in which the first . comprising a switching means for switching the output of the second storage means, and a recognition means for comparing the speech pattern data selected by the switching means and the speech pattern data obtained from the speech input means,
The voice input recognition device is characterized in that the first storage means and the recognition means perform initial settings, and then the second storage means and the recognition means perform voice recognition.

(6) Examples of the invention Hereinafter, the present invention will be explained in detail using the drawings.

FIG. 2 shows a block diagram of an embodiment of the present invention. The audio signal S is input to the audio input unit 1, and its output is connected to the recognition unit 6 and the audio pattern registration/correction unit 2. The output of the voice pattern registration/correction section 2 is input to the storage section 3'. The output of the recognition section 6 is input to the control section 5'. The control line of the control section 5' is connected to the voice pattern registration/correction section 22 displayed in Fig. 7. Storage section 3'
are connected to each. The output of the storage section 3' is input to the recognition section 6.

Further, the storage section 3' includes a standard pattern memory 3'-1, a registered Bakun memory 3'-2, . The switching section 3'-3 stores characteristic data obtained from the voice pattern registration/correction section 2 in a registered pattern memory in response to a control signal from a control section 5'. Furthermore, the contents of the standard pattern memory 3'-1 and the u-recorded pattern memory 3'-2 are selected by the switching unit 3'-3 and output to the recognition unit according to a control signal from the control unit 5'.

In the initial state, in embodiments of the present invention, the contents of the standard pattern memory are selected and used for matching with the input speech. The standard voice pattern referred to here is the number (0~
9) and several times as simple instructions can be considered. When a user inputs ID information (for example, a several-digit number) by voice input, the input voice signal has its characteristics extracted and enters the recognition unit. The recognition section compares and recognizes the data obtained from the voice input section 1 with the standard pattern memory stored in the initial state, and outputs the result to the control section 5'. These recognitions in the initial state are recognitions for selecting whether or not the voice input S to be inputted from now on is registered in the registered pattern memory 3'-2, and if it has already been registered, the control unit 5' uses characteristic data of a specific speaker, that is, the current user, stored in the registered pattern memory 3'-2 as verification data. In addition, if it is not registered, please start the registration process or
It is conceivable to request the re-input of D.

The display section 7 is a display section for conveying these operations to the operator, and the user proceeds with the operation based on the messages displayed on the display section 7.

When the above operations are completed, the user's own registered data is then compared and recognized with the feature data input manually from the voice input section, and the results are output to the control section 5'.

The display section 7 in the embodiment of the present invention displays characters, etc.
RT, plasma display, etc. are possible. It is also possible to output the display unit as audio, and the control unit 5
It is also possible to select and utter pronunciation data that is provided in advance with the voice corresponding to the message data to be uttered from ''.

(7) Effects of the Invention As mentioned above, in the initial state, the present invention provides a comparison between the characteristics of the input voice and the characteristics of the standard voice.
The system selects the characteristics of the user's voice, and then performs recognition using pattern data registered by the user. As a result, there is no need for conventional key operations, and speech recognition with a high recognition rate is possible after the initial state.

[Brief explanation of the drawing]

FIG. 1 shows a circuit configuration diagram of a conventional speech recognition system, and FIG. 2 shows a circuit configuration diagram of a speech recognition system according to the present invention. 1... Voice input section, 2... Voice pattern registration/correction section, 3'... Storage section, 3'-1... Standard pattern memory, 3'-2... Registration Bakun memory,
3'-3...Switching unit, 5'...Control unit, 6...
・Recognition Department Patent Applicant Fujitsu Limited

Claims (1)

[Claims] (1) A voice input means for extracting voice characteristics, a first storage means storing standard voice pattern data, a second storage means storing voice pattern data of a specific speaker, and the first storage means storing voice pattern data of a specific speaker; ．． The first storage means comprises a switching means for switching the output of the second storage means, and a recognition means for comparing the audio pattern data selected by the switching means and the audio pattern data obtained from the audio input means. A voice input recognition device characterized in that initial settings are performed by a recognition means, and then voice recognition is performed by the second storage means and the recognition means.