JPH041919B2 - - Google Patents

Description

[Detailed description of the invention]

(a) Technical Field of the Invention The present invention relates to a switching device for speech recognition means intended for unspecified speakers and speech recognition means intended for specific speakers in a speech processing system. (b) Technical background With the development and spread of data processing technology in recent years, as part of the input/output means in data processing systems,
Speech recognition and synthesis technology, which was initially limited to voice-controlled classification devices and telephone guidance services, has been supported by advances in semiconductor integration technology and circuit design technology, and has expanded to include methods that require multiple levels of sophistication or large-capacity storage. As they became smaller and cheaper, data input/output means using Japanese language processing became widely used, taking advantage of the fact that they were suitable for interactive formats and were easy to use without requiring special training on the part of the operator. . (e) Prior Art and Problems Conventionally, speech processing devices have used voice recognition means for an unspecified number of speakers that is input via a telephone exchange, etc., or voice recognition that is performed by registering the voice pattern of a specific speaker. The means identifies the voice input, converts it into digital voice data, and sends it to the host computer (HOST). The voice processing device is a peripheral device for data input/output that sends a response voice to the caller using a voice synthesis section provided within the device according to the response data obtained by the data processing function of the HOST. In the above-mentioned recognition function, the recognition means targeted at unspecified speakers can be easily used because there is no need to register a speech pattern before use. However, it is necessary to maintain a dictionary for recognition in consideration of unspecified variations, and it is difficult to increase the number of words to be recognized or change the type of words, so it is subject to order specification. In order to deal with the large number of product names, etc., it has been necessary to code them using numbers and input them. On the other hand, with recognition methods that target specific speakers, it is necessary to register, for example, a speech pattern for each word before use, but since individual characteristics are captured and registered, variations among speakers are taken into account. Since this is not necessary, the number of words to be recognized can be increased or changed relatively easily. On the other hand, in systems using telephone lines, etc., many people use voice processing devices, so if the system has sufficient memory capacity, all users can be Although it is possible to provide a wide range of services to people, if the application is extended to all people, the storage capacity required for the dictionary will become enormous and cannot be accommodated. Furthermore, since not all users use the device frequently, unnecessary voice patterns are registered, resulting in a decrease in the utilization rate of the device. (d) Purpose of the Invention The purpose of the present invention is to eliminate the above-mentioned drawbacks. For example, for a certain period of one month or six months, a person who has a high usage rate can use a voice processing device to identify in advance the voice identification code that is used when inputting voice. The code is registered in the code table, and each time a user accesses the code, the code is checked against the table. Registered users will be recognized by a registered voice recognition method, while unregistered users will be recognized by an unspecified recognition method, even if it is somewhat inconvenient. For example, it is an attempt to provide a means for obtaining the highest possible utilization efficiency of the device with a limited storage capacity by taking advantage of both the advantages of processing by inputting a product name using a code. (e) Structure of the Invention The object of the present invention is to provide a first speech recognition means that targets speech by an unspecified speaker, and a first speech recognition means that targets speech by a specific speaker, in a speech processing device that identifies speech input and responds with a speech output. a second voice recognition means based on a registration formula, a switching means for switching input to both means under control of a control section, and the control section stores identification code data of a speaker using the second recognition means in its storage area. The controller includes an identification code table to be registered, and the control unit allows the speaker to input an identification code by voice upon access, and selects the first recognition means by the switching means, and uses the data of the identification code obtained for the identification code. When a match is not obtained by checking the code table, the first recognition means continues to recognize the speaker's voice, and when a match is obtained, the switching means is activated to recognize the second recognition means. Select the second option and notify the speaker of the selection.
This can be achieved by providing a speech recognition unspecified/specific switching device characterized in that the speech recognition is executed by the recognition means. (f) Embodiment of the invention An embodiment of the invention will be described below with reference to the drawings. The figure shows a block diagram of a speech recognition non-specific/specific switching device in an embodiment of the present invention. In the figure, 1 is a speech processing device, 2 is a network control unit (NCU), 10a is a control unit, 10b is a storage unit thereof, 11a is a speaker-independent speech recognition unit, 11b is a dictionary for non-specific speakers, 12a 12b is a speech recognition unit for a specific speaker; 12b is a dictionary unit that stores registered speech patterns; 13a is a speech synthesis unit; 13b is a voice message file unit; 14 selects audio signals to be input and output under the control of the control unit 10a. Switching section, 1
5 is a display, 16 is an input unit using a keyboard (KB) or light pen, etc., 10ba is a control program, 10bb is control data, and 10
bc is a recognition code table. In the configuration shown in the figure, the unit 10a accesses the control program 10ba and control data 10bb that store the voice signal inputted via the NCU 2 via the exchange when an external user makes a call from a telephone or the like in its storage unit 10b. Under continuous control, the switching unit 14 selects the recognition unit 11a or 12a and inputs the audio signal. The recognition unit 11a or 12a accesses each dictionary unit 11b or 12b, recognizes the audio signal, and outputs audio data in the form of a digital signal, and the control unit 10a sends the audio data to the host computer (HOST). performs data processing. Next, the control section 10a receives the processed data from the HOST and applies it to the speech synthesis section 13a, and the speech synthesis section 13a accesses the file section 13b and converts the processing data into a voice message. The control unit 10a causes the switching unit 14 to selectively connect the voice synthesis unit 13a, and responds to the caller by selecting a voice message based on the processed data via the NCU 2. The above is the same as the voice processing operation in the conventional voice processing system, but in the present invention, in order to increase the operational effect of the voice response device 1, the control unit 10a is a recognition unit 1 corresponding to unspecified/specific speakers in voice recognition.
The connection selection by the switching unit 14 of 1a or 12a is controlled while accessing the recognition code table 10. The recognition code table 10bc accumulates the past usage status of the voice processing device 1 for a certain period of time, for example, for each user code, and registers a plurality of users in the table 10bc in descending order of usage frequency. Registration may be done by the operator of the device 1 by displaying the user data on the screen of the display 15 and setting the registration using the input section 16, or by automatic processing of corrections such as deletions and additions every month. Also good. table 10
bc is constructed as shown in Table 1.

【table】

Claims (1)

[Claims] 1. In a speech processing device that identifies speech input and responds with speech output, a first speech recognition means targets speech by an unspecified speaker, and a second speech recognition means by a registration method targets speech by a specific speaker. means, a switching means for switching input to both means under control of the control section, and the control section includes an identification code table for registering identification code data of a speaker using the second recognition means in its storage area. , the control unit causes the speaker to input an audio identification code when accessing the device, and causes the switching means to select the first recognition means, and collates the obtained identification code data with the identification code table to find a match. If not, the first recognition means continues to recognize the speaker's voice, and if a match is found, the switching means selects the second recognition means and informs the speaker of the recognition. A speech recognition unspecified/specific switching device characterized in that the second recognition means executes speech recognition upon notification.