JPS59177599A - Voice registration system - 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] Technical Field of the Invention The present invention relates to a voice registration method for an appeasement recognition and response system that uses a telephone line or the like as an input medium, and specifically attempts to perform voice registration without making the speaker aware of the registration operation. 4. Prior Art and Problems There are two types of speech recognition devices: a registered type that targets a specific speaker, and a non-registered type that targets unspecified speakers. Speech recognition devices designed for non-specific speakers are easy to use because they do not require voice registration before use. However, a dictionary for recognition must be created taking into account variations among people (speakers), which makes the lexicon 1 etc. complicated, making it difficult to increase or change the number of words to be recognized. On the other hand, with registration-type speech recognition devices, it is necessary to register the speech before use, but once this is done, there is no need to take into account variations among speakers, which simplifies dictionaries, etc., and words to be recognized. Qin It is relatively easy to increase or change. As described above, the specific type and the unspecified type each have their advantages and disadvantages, so by combining both types of devices, a more flexible system configuration with a wider range of use can be expected.

Conventionally, Oak Entry (when attempting to perform unspecified voice recognition in an application such as accepting orders over the phone, the name of each product is coded in advance with numbers, etc., and the code is uttered by the speaker to be used by the device. The method used has been to call 811 and have it done manually.

Numerical codes were used because the large number of product names and their frequent changes made it difficult for unspecified voice recognition devices to create a dictionary for the large number of product names for recognition. It is. However, since numeric codes are difficult to memorize, it is necessary to prepare a correspondence table between product names and numeric codes, and then speak out the numeric code for the desired product while looking at the table, resulting in poor operability. If a registration type voice recognition device is used, it is possible to input the product name directly instead of the code, but voice registration is required prior to LRi&. In addition, considering the large number of product names and the fact that they change, it is difficult to register all product names, and it is common to use only a part of the product name. When you think about it, it's not even efficient. Furthermore, from the viewpoint of recognition rate, it is preferable to have a smaller number of registrations. However, even if you register only the necessary words, some words may still be registered.
Also, it is difficult to manage how the words should be registered. Furthermore, it is difficult to register a new word every time the user wants to use the word.

OBJECTS OF THE INVENTION The present invention aims to reduce the burden on the speaker regarding the above-mentioned voice registration, and also to naturally improve the ability of the registration type voice recognition system as required.

Structure of the Invention The present invention uses a telephone line or the like as an input medium, and includes a speech recognition section for unspecified speakers, a registered speech recognition section for specific speakers, and a speech synthesis section for output. In the voice registration method of the voice recognition response system provided, a hasher is provided to store the voice uttered by the speaker, and if the recognition result of the registration type voice recognition unit for the voice is denied by confirmation by the speaker. The unspecified voice recognition unit switches to the unspecified voice recognition unit, requests the speaker to input a code preset in the unspecified voice recognition unit for the word indicated by the voice, and after the input by the code is completed, the unspecified voice recognition unit The present invention is characterized in that the recognition result obtained by the method and the voice in the buffer are registered in the dictionary of the registration type voice recognition unit, and this will be explained in detail below with reference to the embodiments shown in Figs.

Embodiment of the Invention FIG. 1 is a block diagram showing an embodiment of the present invention, in which 1 is a line connected to a speaker, 2 is a network control side unit (NCU).
, 3 is a switching unit that switches input/output and the recognition unit, 4 is a speech recognition unit for unspecified speakers (hereinafter referred to as unspecified recognition unit), 5 is a dictionary for speakers, and 6 is a specific speaker Registered speech recognition unit (hereinafter referred to as specific recognition unit) for
, 7 is the dictionary memory, 8 is a voice processor added according to the present invention, 9 is a voice synthesis unit that sends back confirmation and response voices to the speaker, 10 is a control unit, 11 is a specific dictionary,
2 is a line connected to the host computer.

Figure 12 is an explanatory diagram showing the French ordering system, where 15 is a telephone used by a speaker (individual consumer, store clerk, etc.), 16 is a telephone exchange, 17 is a post computer, 20 is the voice recognition response system shown in FIG.

In this order entry system, first, the speaker speaks the product name as it is pronounced or in an arbitrarily abbreviated form over the telephone, and the voice response recognition system receives this and stores it in the voice response sofa 8. Recognize with. For recognition, an area corresponding to the speaker is read from the identification dictionary 11, and this is stored in the dictionary memory 7 for use (identification of the speaker will be described later). Then, the voice synthesis unit 9 selects the closest recognition result.
send it back to the speaker for confirmation. If affirmative at this time, the code of the product name and the quantity input by voice at the same time are sent to the post computer to process the order, and then prepared for the next input.If negative, the switching unit 3 Switch to unspecified recognition unit 4. The unspecified recognition unit 4 inputs the product name not as its reading but as the reading of a numerical code assigned in advance to the product name. In other words, unspecified dictionary 5
Since the reading of the numerical code assigned to each product name is registered in advance, the speaker can recognize the desired product name by uttering the numerical code of the desired product name. At this time as well, 1.1# recognition is obtained using the voice from the voice synthesis fl139, but when the result is affirmative, the voice in the voice buffer 8 and the code used by the non-specific recognition unit 4 at that time are used in the specific dictionary. Register on 11. In this way, when the same speaker orders the same product from now on, they can input the product name. Moreover, during this time, the speaker does not have to be aware of voice registration at all.

Note that if the same word has already been registered, it will be replaced. This is effective in optimizing the contents of the dictionary, such as when the same pronunciation or different pronunciations are recognized depending on the telephone line.

An example of ordering orange juice is shown below. However, it is assumed that no word corresponding to orange juice is registered on the device side, and the word ``ohheal'', which corresponds to a large bottle of beer, is registered as the closest word. The coat is 121, and the orange genose coat is 111.

The following is the interaction between the causer (speaker) and the device.

User: ``Orange'' (Stores this audio in the buffer.

Also, it uses a specific dictionary to recognize speech) device -゛Product number 2, ichi, ni shi-)-1i It's a human bottle. P" User: "Yes" Device: P Please tell me the product number. (switches to non-specific recognition) User: ``Ichi'' Device: ``Bee'' User 2-in-1 Device: ``Pii'' User 2-in-1 Device - Pa Product Number Ichi, Ichi, Ina Orange juice, isn't it?゛User: ``Yes'' (At this time, the previous ``Orange'' is automatically registered.

) From then on, if you say "orange" using a registered recognition device, the orange juice with product number 111 will be recognized.

Order acceptance processing in order entry is generally performed in the following procedure. First, the speaker dials on the telephone 15 to call up the voice response recognition system 20, notifies the subscriber's subscriber number by voice, and also notifies the password for identity verification. In addition to ordering, in addition to ordering,
Since there are inquiries, changes, etc., it is common to notify them by a code to distinguish them. For unspecified recognition methods, the product number and quantity are required after this, so call 111.
05 (111 products 5 (solid), 15603 C15
The order is read as follows: ``3 pieces of item 6'' and finally ``End''.The order is now complete, and the device now reads out the entered information and asks for confirmation.

If affirmative, the order details are registered in a file on the host computer side, or control is passed to the next process.

When ordering by product number in this way, the user has a list that associates the number with the product name.

There are over 1000 product names listed in this table, so
It is difficult to register the pronunciation of the product ``old'' in a dictionary, and since the products handled change over time, it is also difficult to change the contents of the dictionary each time. Therefore, in general, non-specific dictionaries deal with a wide variety of products that are rarely used, while specific dictionaries deal with a small number of frequently used products. Of course, it is more convenient for the user to be able to input the pronunciation of the product name that is easier to intuitively identify.

In this respect, according to the present invention, the user can treat it as a registration type, and if something is not registered or needs to be changed, it will be automatically registered by inputting a code when prompted by the device. The system will be complete and easy to use for users.

Effects of the Invention As described above, the voice registration method of the present invention has the advantage that voice registration is performed without the user being aware of it, resulting in a voice recognition system that gradually becomes easier to input.

[Brief explanation of the drawing]

1 and 2 are block diagrams showing an embodiment of the present invention and its installed state. In the figure, 1 is a line, 3 is a switching unit, 4 is an unspecified recognition unit, 5 is a dictionary thereof, 6 is a specific recognition unit, 8 is a voice recognition unit, 9 is a voice synthesis unit, and 11 is a specific dictionary. Applicant Fujitsu Limited

Claims (1)

[Claims] A voice recognition response system that uses a telephone line etc. as an input medium and includes a voice recognition unit for unspecified speakers, a registered voice recognition unit for specific speakers, and a voice synthesis unit for output. In the registration method, a buffer is provided to store the voice uttered by the speaker, and if the recognition result of the registration type voice recognition unit for the voice is rejected by the speaker, the unspecified voice recognition unit , the speaker requests the input of a code preset by the unspecified speech recognition unit for the word indicating the voice, and after the input using the code is completed, the recognition result by the unspecified speech recognition unit and the data in the buffer are requested. A voice registration method characterized in that the voice of is registered in a dictionary of the registration type voice recognition unit.