JPS6243213B2 - - Google Patents

Description

[Detailed Description of the Invention] (1) Technical Field of the Invention The present invention relates to foreign words used in audio word processors. This invention relates to a monosyllabic voice input method that can easily correct the difference in reading).

(2) Prior Art and Problems Conventionally, in Japanese spoken word processors, the sounds of foreign words cannot be directly displayed in Japanese Katakana, but are sometimes displayed as a combination of two characters.
For example, if you enter the foreign word ``Fuair'' vocally and recognize it as a monosyllable, if a recognition device that does not recognize ``Hua'' recognizes ``Hua'' as ``Ha'', then ``Hail'' is displayed as a character. is inappropriate, and it is necessary to correct the recognized "ha" to "hua". Of course, the voice input is "F", "A", "I",
If you input ``ru'', the characters will be displayed appropriately, but the voice input will be unnatural, and you will be using the voice word processor without its features. Similarly, “Huo”
In the case of , even if the monosyllable is recognized as "ho", the character may be displayed as "huo". In this way, even if a single katakana character is recognized in monosyllabic recognition, it may not be appropriate as Japanese if it is not expressed as two katakana characters in character display. However, since the input specification for foreign words is in Katakana, there is no one that has the function of converting two characters for one character.

In addition, when inputting a foreign word by voice, there are vocal fluctuations (prolongation of the vowel), and simply recognizing this as is will result in a monosyllabic character string that is not in the foreign word dictionary. can't get it. For example, when the ``ta'' in words such as ``computer'' and ``motor'' is input by voice, in monosyllabic recognition, long sound processing is performed after the monosyllabic candidate ``ta'', and ``ta'' is
is recognized. Therefore, in this case, it is necessary to correct "ta" to "ta".

That is, in monosyllabic recognition, a long sound is added to one kana character and it is recognized, and in character display, it may not be possible to match it with a dictionary unless it is expressed using that one kana character. The same applies to "sir".

As mentioned above, when the recognition result of monosyllable recognition differs from the written expression, it is necessary to correct it, but in the conventional method, this correction is not necessary to obtain the intended foreign word. No measures were taken.

(3) Purpose of the Invention The purpose of the present invention is to correct the recognition result when monosyllable recognition is performed on the input speech of foreign words used in an audio word processor, which differs from the normal character display. It is an object of the present invention to provide a monosyllabic voice input method that can be easily performed.

(4) Structure of the invention In order to achieve the above object, the monosyllabic speech input method of the present invention inputs unknown input speech into a monosyllabic recognizer,
It is compared with the monosyllabic standard pattern registered in the dictionary in advance, and the monosyllabic candidate group resulting from the recognition is entered into the character string creation section, and based on the foreign word input specification, the foreign word correction table is further referred to to create a monosyllable suitable for character display. This method is characterized by adding candidates and comparing them with a vocabulary dictionary of foreign words to obtain the character representation of the intended foreign word.

(5) Embodiments of the invention Generally, when a foreign word is to be recognized, the pronunciation of the foreign word has many unstable elements, and the content of the pronunciation and the content written in Katakana may not match. There are many.

Therefore, even if the recognition result is correct, it often happens that the foreign word cannot be input because it does not match the contents of the foreign word dictionary.

First, statistical data shows that loanwords such as
We have a matrix of 101 monosyllables that shows which monosyllables will be used, and after registering the monosyllables, we have created a matrix for the person who inputs the speech by having the individual pronounce the foreign word. However, by referring to this table during recognition, it is possible to add appropriate monosyllable candidates to the character display (reading) of monosyllables that are not subject to recognition. As a result, the recognition result of "c" can be corrected to "fa".

Furthermore, by creating rules based on statistically generated pronunciation fluctuations of foreign words, storing them in the same table as above, and referring to the table during recognition, it is possible to add monosyllable candidates in the same way. can.
As a result, the recognition result of "ta" can be corrected to "ta".

The figure is a configuration explanatory diagram of an embodiment of the present invention.

In the figure, an input speech is inputted into a monosyllable recognition unit 1, and a monosyllable standard pattern 2 registered in advance in a dictionary is used.
(pattern learning), and the recognition result candidate group of each monosyllable is sent to the character string creation unit 3 together with information such as the similarity of the recognition result and the likelihood feature.
The character string creation unit 3 refers to the foreign word correction table 4 (monosyllabic error learning) based on the sent information and the foreign word input designation (katakana designation).
This foreign word correction table includes the matrix of the voice inputter, as well as the characters that are not recognized as mentioned above,
Single syllable candidates corresponding to vocal fluctuations are stored. In either case, a character string is created by adding appropriate monosyllable candidates to the character display and sent to the kanji/kana mixed sentence conversion section 5. This kanji/kana mixed sentence conversion unit 5 compares the sent character string with a foreign word dictionary 6 (frequency learning), performs kanji/kana mixed sentence conversion, displays candidates, and selects the intended one. While inputting Japanese audio.

(6) Effects of the Invention As explained above, according to the present invention, when monosyllable recognition is performed on input speech using foreign words used in an audio word processor, the recognition result is a character display in Katakana. If it is different from the original word, in order to correct it, the system refers to the foreign word correction table based on the foreign word input specification, adds appropriate monosyllable candidates to the character display, and obtains the intended foreign word. As a result, as mentioned above, even if the monosyllable candidate in the input speech recognition result is recognized as "ha" or "ho", the monosyllable candidate added from the loanword correction table is "hua" or "huo". Similarly, even if a single syllable candidate in the recognition result of the input speech is recognized as "ta" or "sah" by long sound processing, it will not be added from the loan word correction table. By making the single syllable candidate ``ta'' or ``sa'' possible, the intended foreign word can be obtained.

[Brief explanation of the drawing]

The figure is an explanatory diagram of the configuration of an embodiment of the present invention, in which 1 is a monosyllable recognition unit, 2 is a monosyllable standard pattern, and 3 is a monosyllable recognition unit.
4 is a character string creation unit, 4 is a foreign word correction table, 5 is a kanji-kana mixed sentence conversion unit, and 6 is a vocabulary dictionary (loan words).

Claims (1)

[Scope of Claims] 1. Entering unknown input speech into a monosyllable recognition unit, comparing it with a monosyllabic standard pattern registered in a dictionary in advance, and inputting a group of monosyllabic candidates resulting from the recognition into a character string creation unit,
Based on the foreign word input specification, the system also refers to the foreign word correction table and adds appropriate monosyllabic candidates to the character display.
A monosyllabic speech input method characterized by obtaining the character representation of an intended foreign word by comparing it with a dictionary of foreign words. 2. Even if the monosyllable candidate in the recognition result of the input speech is recognized as "ha" or "ho", the monosyllable candidate added from the loan word correction table is "hua" or "huo". A monosyllabic voice input method according to claim 1. 3 Even if the monosyllable candidate in the recognition result of the input speech is recognized as "ta" or "sa" through long sound processing, the monosyllable candidate added from the loanword correction table is "ta" or "sa". A monosyllabic voice input method according to claim 1, characterized in that: