JPH0833741B2 - Voice recognition device - Google Patents

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice recognition device, and more particularly, to a correction process at the time of a rejection in which a recognition result is erroneous or cannot be recognized.

(Prior Art) FIG. 2 is a diagram showing a configuration example of a conventional speech recognition apparatus. This type of device is described, for example, in JP-A-57-62097 or JP-A-57-191698. In FIG. 2, 102 is a microphone for voice input, and 103 is a detection unit for detecting the voice section of the voice input from the microphone 102.
104 is a feature extraction unit that extracts a feature parameter from the voice section detected by the detection unit 103, 106 is a dictionary unit that stores the feature parameter, and 105 is a feature parameter that is stored in the dictionary unit and the feature parameter that is extracted from the input voice. A matching unit that performs matching with and calculates a matching distance, and
Reference numeral 107 denotes a determination unit that determines the matching distance calculated by the matching unit and outputs the recognition result. Reference numeral 100 denotes a display device, 101 denotes an auxiliary input device for instructing and inputting operations such as correction and registration, and 108 controls the display device and the input device, and the final recognition result together with the determination result from the determination unit 107 is output to the terminal 109. It is an output controller. In the conventional voice recognition device having such a configuration, as a correction method at the time of erroneous recognition in which a recognition result for a voice input is incorrect, or a rejection (unrecognizable determination), it is repeated until a correct recognition result is obtained by re-voice input. It is common to have a method and a method of inputting correction information from an auxiliary input device.

Various factors can be considered as the cause of erroneous recognition or rejection that is uttered during voice recognition. Misrecognition generally tends to occur when the feature parameters extracted from the voice uttered by the operator and the matching distances for the two or more different feature parameters stored in the dictionary are close to each other. . On the other hand, the cause of rejection is generally uncertain when the feature parameters extracted from the voice uttered by the operator have a very large matching distance with all the feature parameters stored in the dictionary, or the input voice is weak. There are two ways.

Considering this erroneous recognition and rejection from the operator's side, if erroneous recognition frequently occurs only with respect to input speech data of a specific classification, as a correction process, voice input is performed again until correct recognition results are obtained. Is often re-voiced several times. This phenomenon corresponds to the case where the difference in matching distance between the characteristic parameters of different classifications accumulated in the dictionary unit is small. If you have to enter such data frequently,
Change the characteristic parameters of the dictionary in the voice recognition device,
It is considered more efficient to eliminate the above phenomenon. This operation is called re-registration, and is a process of changing some or all of the characteristic parameters in the dictionary. However, the re-registration process generally requires more manpower than the simple correction process. Therefore, if the frequency of input data of the type in which misrecognition occurs frequently is low, rather than performing correction method by re-voice and re-registration processing, input the data, and when the misrecognition occurs, The number of operations is smaller when the correction process is executed by a data input method that does not rely on voice, such as an auxiliary input device. Another misrecognition,
The cause of the rejection is that the operator's utterance is vague or weak. Operators are strong against erroneous recognition and rejection due to such causes,
In many cases, it is recognized correctly if it is uttered clearly. If the operator is more or less familiar with the voice recognition device, it is possible to easily discriminate erroneous recognition and rejection due to the deficiency of the operator's voice. The most efficient correction means in this case would be correction processing by re-voice.

As described above, the most efficient correction methods for erroneous recognition and rejection that occur due to different performances are different. Therefore, when data is actually input using a voice recognition device, an efficient correction method for erroneous recognition and rejection often changes from time to time.

(Problems to be Solved by the Invention) However, in the above-described conventional voice recognition device, only one of the correction methods for erroneous recognition and reject is effective depending on the mode setting. For example, the correction information may be input from the auxiliary input device, or may be performed by re-voice. And the change of the correction method requires another operation such as correction mode switching in addition to the correction processing operation. Therefore, in the conventional voice recognition device, it is necessary to change the correction mode quite frequently when erroneous recognition or rejection occurs. In particular, an operator who is familiar with the voice recognition device can determine the optimum correction method instantly due to the nature of erroneous recognition and rejection, and perform the work of changing the correction mode accurately. The efficiency is reduced and the features of the voice recognition device cannot be fully exhibited.

An object of the present invention is to improve the erroneous recognition and the correction processing for rejects described above to improve the operability of the voice recognition device.

(Means for Solving Problems) In order to solve the problems, the present invention includes a detection unit that inputs a voice and detects a voice section thereof, and a feature extraction unit that extracts a feature parameter of the voice section. , A dictionary unit for accumulating a predetermined characteristic parameter, and a matching result between the characteristic distance of the input voice extracted by the characteristic extracting unit and the characteristic parameter accumulated in the dictionary unit are compared and determined to obtain a recognition result. In a voice recognition device having a determination unit for outputting, as a correction process when the recognition result is erroneous recognition or unrecognizable, a first correction means by re-voice input and a second input device by an auxiliary input device for inputting correction information. Correction means is provided, and the second correction is performed before a predetermined time elapses from the time when the correction processing is performed by the first correction means.
The correction process by the correction unit 1 is effective only when the correction process by the correction unit No. 1 is not performed.

(Operation) According to the present invention having such a configuration, the correction processing by the first correction is performed, and the correction processing by the second correction means is performed by the time when a predetermined time elapses from the time when the correction processing is performed. The correction process by the first correction means is valid only when the correction is not made. Therefore, when the operator's utterance is ambiguous or the utterance is weak and misrecognition or rejection occurs, the familiar operator immediately determines the deficiency and performs an appropriate correction process. By monitoring the corrective action on the device side, an appropriate correction mode can be selected, and extra scanning can be simplified. Therefore, the above problems can be solved.

(Embodiment) Hereinafter, the present invention will be described based on an embodiment with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of the present invention. In FIG. 1, 200 is a display device, 201 is an auxiliary input device, and 202 is a microphone for voice input. Also, 203
Is a detection unit that detects the voice section of the voice input from the microphone, 204 is a feature extraction unit that extracts the feature parameter from the voice section detected by the detection unit 203, 206 is a dictionary unit that stores the feature parameter, and 205 is a dictionary. A matching unit that performs matching between the feature parameter stored in the unit 206 and the feature parameter extracted from the input voice and calculates the matching distance, and 207 determines the matching distance calculated from the matching unit 206, and the recognition result Is a determination unit that outputs
208 controls the display device 200 and the auxiliary input device 201, and is finally input to the voice recognition device from the voice section detection signal line 210 based on the input voice section detection signal and the recognition result from the determination unit 207. It is a controller that outputs data. 209 is its output terminal.

FIG. 3 shows a flowchart for voice input according to this embodiment. The operation of this embodiment will be described below with reference to this flowchart.

In this embodiment, before the voice input work is started, there are three modes as a selection mode of the correction means at the time of erroneous recognition and rejection, that is, a correction mode by re-voice, a correction mode by KEY input of the auxiliary input device, and the features of this embodiment KEY input ・
There is a simultaneous correction voice input mode. In the correction mode setting, the operator selects any one of these three modes (step 10; the step is hereinafter abbreviated as S). Here, the key input / correction voice input simultaneous mode, which is a feature of this embodiment, will be mainly described. The signal input from the microphone 202 in FIG. 1 is first input to the detection unit 203 (S11). The detection unit 203 detects the voice section in the signal, and when the voice section is detected, the feature extraction section 204 specifies the section for extracting the feature and sends a voice detection signal to the controller 208 via 210. (S12). The feature extraction unit 204 extracts the feature parameter in the voice section and sends it to the matching unit 205 (S13). The matching unit 205 performs matching between the feature parameter in the dictionary unit 206 and the feature parameter of the audio signal extracted by the feature extraction unit 204, and the matching distance calculated as a result is sent to the determination unit 207 (S1
Four). The determination unit 207 performs recognition processing or rejection determination based on the matching distance calculated by the matching unit 205, and sends the result to the controller 208 (S16).
If the result is reject, the controller 208 changes to the correction mode and instructs the display 200 or the like to display the mode change (S20). If the result is not rejected,
The controller 208 performs a recognition result confirmation operation (S17).
Here, in S18, when there is no determination of erroneous recognition from the operator, the controller 208 returns to the state of waiting for voice input from the operator. If all the voice inputs are completed here, the controller 208 ends the voice input mode (S19).
On the contrary, when the operator gives an instruction that the recognition is incorrect, the controller 208 switches to the correction mode as in the case of the reject (S21). The process can be divided into three depending on the setting of the correction mode. If the correction mode by re-voice is selected, voice input, voice section detection, feature extraction, pattern matching, and determination processing are processed in the same order as in the normal voice input mode (S22 to S26), reject, and erroneous recognition. If the error does not occur, the correction process ends (S27 to S29).
If the correction mode by the KEY input from the auxiliary input device 201 is selected (S32), the recognition result is corrected by the correction information from the KEY (S33), and the correction mode ends at this point.

When the KEY input / correction voice input simultaneous mode is selected, the voice recognition device is the microphone 202 and the auxiliary input device 20.
Both 1 are in the input waiting state (S30). The operator may input the correction information by an efficient correction means. The controller 208 determines which of the input information of the two is prioritized (S31). This determination method is as follows. That is, only when there is no correction input from the auxiliary input device 201 by the time when a predetermined time elapses from the time when it is confirmed by the voice section detection signal of the detection unit 203 that there is a correction input from the microphone 202, The correction information by the voice input is valid, and in other cases, the correction information from the auxiliary input device, that is, the correction information by the KEY IN input is valid. The processing in the KEY input / correction voice input simultaneous mode described above can be easily realized by changing a part of the processing procedure of the correction mode by re-voice and the correction mode from the auxiliary input device 201.

(Effect of the Invention) As described in detail above, according to the present invention, an input method which is always appropriate as a correction means for erroneous recognition / rejection that occurs in the work of inputting data etc. using a voice recognition device is a correction mode. It can be selected without extra operations such as changing, and work efficiency can be improved.

[Brief description of drawings]

FIG. 1 is a block diagram of an embodiment of the present invention, FIG. 2 is a block diagram showing a configuration example of a conventional voice recognition device, and FIG. 3 is an operation flowchart of this embodiment. 200: display device, 201: auxiliary input device, 202
...... Mike, 203 …… Detector, 204 …… Feature extractor, 205
…… Matching unit, 206 …… Dictionary unit, 207 …… Judgment unit, 20
8 …… Controller, 209 …… Output terminal.

Claims (1)

[Claims] 1. A detection unit for inputting voice and detecting the voice section, a feature extraction unit for extracting feature parameters of the voice section, a dictionary unit for accumulating predetermined feature parameters, and the feature extraction. In a voice recognition device having a determination unit that compares and determines the matching distance between the feature parameter of the input voice extracted by the unit and the feature parameter stored in the dictionary unit, the recognition result is erroneous. As the correction processing in the case of recognition or unrecognition, a first correction means by re-voice input and a second correction means by an auxiliary input device for inputting correction information are provided, and the correction processing by the first correction means is performed. Only when the correction processing by the second correction means is not performed by the time when a predetermined time has passed from the time when the correction processing is performed, the correction processing by the first correction means is validated. Speech recognition apparatus according to claim and.