JPS62119595A - Japanese language processor - 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. The present invention relates to a Japanese language processing device, and more particularly to a Japanese document creation device having a voice input device.

BACKGROUND ART In recent years, Japanese document creation devices (so-called word processors) have been increasingly used to create documents, but a certain amount of training is required to input from a keyboard, which has hindered their widespread use. There is. As a way to deal with this, word processors equipped with voice input devices have been developed, allowing users to enter text without having to search for letters on a keyboard. However, since ordinary spoken word processors use a monosyllabic input method, for example, to input ``r this invention'', [ha・n・ha・tsu・me・
It must be separated like this.

To eliminate such unnaturalness, input words.

Phrase input 9 It is necessary to input sentences, and attempts have been made to do so. However, the number of words required for a word processor is at least 500, and a pattern matching method for a specific speaker is currently a practical method for recognizing such a large number of words. In the specific speaker method, the user needs to register the +B language voice with his/her own voice, and in this case, the number of +B words is 500 or more.
〕of! The word processor cannot be used until the LT word has been uttered at least once, resulting in an enormous burden on the user.

Purpose The present invention was made in view of the above-mentioned circumstances.
In particular, this invention was developed with the aim of providing a 1-1 native language speech information processing device that can register Japanese sentence patterns with little burden on the user.

Configuration In order to achieve the objective 1-1 of ``1, the present invention includes a voice input section, an analysis section, a recognition section, a display section for displaying the results output from the recognition section, and a display section for displaying the results outputted from the recognition section. A Japanese language speech processing device is equipped with an instruction section that gives instructions to the result, and has a section that temporarily holds the input voice, and for the displayed result, outputs a message corresponding to the Japanese language instructed by the instruction section. The pattern in the holding section is registered as a standard pattern for speech recognition. below,
An explanation will be given based on an example of the present invention.

FIG. 1 is a configuration diagram for explaining one embodiment of the present invention, and in the figure, 1. is a microphone, 2 is a voice section extraction unit, 3 is a feature conversion unit, 4 is a register, 5 is a collation unit, 6 is a standard pattern unit, 7 is an ephemeral Kanji conversion unit, 8 is a display unit, 9 is a keyboard, and a standard pattern It is assumed that a small number of standard patterns are registered in advance in section 6. Now, if an unknown voice is input from the microphone 1, the part related to the voice is converted into a feature amount, such as a frequency spectrum, and is temporarily stored in the register 4, along with the characteristic pattern of the unknown voice. is displayed on the display unit 8 as a recognition result. The display section 8 may be a normal CRT, or a kana/kanji conversion section may be provided in front of the display section to display kanji characters. A keyboard is attached to the display section 8, and text can be corrected using the keyboard while viewing the CRT. For example, if you want to create a sentence like the one shown in Figure 2, ``It's a really nice day today...'', separate each word and write ``Thursday 1-1, Makoto, Ni...''
Suppose that when you enter "Makoto", "Makoto" is not registered as a standard pattern. At this time, as shown in the figure, the SIT word ``Marugoto J'', which is most similar to ``Makoto'', is output.

In this way, the human-generated speech starting from word +1 is converted into features and then stored in the register in order. Now, when you move the cursor to the part where the word +11-word that you made a mistake is displayed, "whole" is displayed. Move and use the keyboard to correct this as "Makoto." In this way, the third ``Makoto'' in the register becomes ``
It turned out that there was an error in the "whole", and the "
The "Makoto" pattern is newly registered as a standard pattern. With this method, the user can register a pattern without being aware that he or she is registering a standard pattern for a voice recognition neck wear. However, "Makoto" in the precedent
It is unclear whether the misrecognition was due to the pattern not being registered or whether it was due to poor pronunciation. Therefore, one possible method is to have a list of words registered in the standard pattern, and to register a new pattern only when the word modified from the keyboard is not included in this list.

FIG. 3 is a diagram showing an example of an electrical block diagram for implementing the above method, and this example is similar to the example shown in FIG. Same as example. However, if patterns are registered one after another in this way, the number of confusing words increases. Therefore, in order to prevent this, a method of introducing word similarity may be considered.

FIG. 4 is a diagram showing an example in which the above-mentioned word similarity is introduced. In this example as well, inputs are converted into feature quantities and recognized in the order in which they are input, as in the previous examples. , and create Japanese sentences by displaying them on the display. At this time, several input audio patterns can be stored in a register, and when it becomes full, the oldest patterns are erased and new patterns are saved. As in the embodiment shown in FIG. 2, after some speech recognition has been performed, R4 (R4) is corrected on the display section 2. With the intention of doing some work, I said,
If you uttered ``to...'' but it was incorrectly displayed as ``according to the rules of the association...'', Taikei and accounting are very similar, and inputting ``Taikei'' and ``accounting'' are incorrectly displayed. The degree of similarity with the standard pattern also increases. Therefore, if a standard pattern called "large system" is registered, errors between large system and accounting will increase, so the similarity is compared in the similarity recognition section 1], and this is the smaller one, that is, the standard pattern. Only when there is no confusing word in the pattern, and when the similarity determining unit 12 determines that the degree of similarity is smaller than a certain reference value, the incoming cover pattern is registered as a standard pattern.

This prevents registration of similar patterns.

As is clear from the explanation in Section 2, according to the present invention, it is possible to realize a 1-1 I/L processing device that can register voice standard patterns with less burden on the user.

[Brief explanation of drawings]

FIG. 1 is an electrical block diagram for explaining an embodiment of the Japanese language processing device according to the present invention, FIG. 2 is a diagram for explaining the operation of the present invention, and FIGS. 3 and 4 2A and 2B are diagrams for explaining other embodiments of the present invention, respectively. 1...Microphone, 2...Audio section cutting unit, 3...
・Feature value conversion unit, 4... register, 5... collation unit,
6... Standard pattern section, 7... Kana-Kanji conversion section, 8
. . . Display section, 9. Keyboard, 10. Registered word list section, 11.. Similarity recognition section, 12.. Similarity determination section. Patent applicant: Ricoh Co., Ltd. = 7 = Figure 1 Figure 2 Voice: "lmanjitsu""ha""makoto""ni" Register: Hanjifu IS Makoto ni Display 2 Today is the whole 1 word↓ Correction: Makoto Figure 3 Figure 4 1 ″ 4 !L# YES ++ O ?

Claims (3)

[Claims] (1) A Japanese voice that includes a voice input section, an analysis section, a recognition section, a display section that displays the results output from the recognition section, and an instruction section that gives instructions to the displayed results. The processing device has a part that temporarily holds input speech, and registers a pattern in the holding part corresponding to the Japanese language instructed by the instruction part as a standard pattern for speech recognition in response to the displayed result. A Japanese language processing device characterized by: (2) It is determined whether or not the character string instructed by the instruction section is registered in a speech standard pattern, and if it is not registered, it is registered. ) Japanese language processing device described in section 2. (3) Claim No. 1 characterized in that it has a part that temporarily holds the similarity of the input voice, and registers patterns in which the similarity of the recognition result is less than a certain value.
) Japanese language processing device described in section 2.