JP2001067097A - Document preparation device and document preparing method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a document preparation device using a voice recognizing function, capable of efficiently using the voice recognizing function if sufficient texts can not be collected. SOLUTION: A voice recognizing part for analyzing inputted voice into a group of vocal-sound characteristic parameters and performing voice recognition based on the information of the characteristic parameters and on linguistic information, has a continuous voice recognizing part 3 for recognizing a continuously produced voice and a word voice recognizing part 4 for recognizing a voice spoken as a word. A user prepares a document while arbitrarily switching between the two voice recognizing parts by selecting it using a mouse or a keyboard or by vocally inputting a command, etc.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to a document creating apparatus using a plurality of voice recognition functions having different properties.

[0002]

2. Description of the Related Art Conventionally, a continuous speech recognition function has been widely used in document creation by dictation using a speech recognition function. This continuous speech recognition function operates using information about the quality of voice called an acoustic model and linguistic information about words. A Hidden Markov Model (HMM) is widely used as an acoustic model, a statistical model (bigram) regarding the order of occurrence of two sets of words, and a statistical model (trigram) regarding the order of occurrence of three sets of languages is widely used as a language model. (For example, Seiichi Nakagawa, "Speech Recognition by Probabilistic Model," IEICE, July 1988
Month).

In order to create the above language model, in the case of Japanese, texts similar to a desired text (text to be input) are collected, and divided into words.
It was used as a language model by determining the link statistics of the word.

[0004]

In the conventional dictation, when the amount of text that can be collected is small, the word chain of the text does not provide sufficient statistical information, or a word in which the statistical information does not function effectively (for example, a person's name). And the recognition performance for those words was not yet enough.

An object of the present invention is to solve the above-mentioned problem caused by using only the continuous speech recognition function in the process of creating the text by voice input, and to provide a document creation apparatus that uses the speech recognition function efficiently. Is to do.

[0006]

According to a first aspect of the present invention, there is provided a document creating apparatus which analyzes an input voice into a group of voice acoustic characteristic parameters, and performs recognition based on the characteristic parameter information and language information. A speech recognition device to perform has a continuous speech recognition unit for recognizing continuously uttered speech, and a word speech recognition unit for recognizing speech uttered as a word, and a user selects both recognition units with a mouse or a keyboard. Alternatively, a document is created while arbitrarily switching by input using a voice command or the like.

According to a second aspect of the present invention, in the document creating apparatus of the first aspect, the user selects one word dictionary from a plurality of word dictionaries in the word speech recognition. It is characterized by. Claim 3
According to a second aspect of the present invention, there is provided the document creating apparatus according to the second aspect, wherein all word dictionaries in which word items of all other word dictionaries are registered are registered as one of the word dictionaries.

According to a fourth aspect of the present invention, in the document creating apparatus of the first aspect, while the user is creating a document by using the continuous voice recognition unit, the user can generate a continuous voice by a voice command. It is characterized in that the recognition is interrupted and the word recognition unit is activated, and that the continuous speech recognition can be resumed after the completion of the word recognition.

[0009]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 shows a block diagram of the document creation apparatus of the present invention. The document creation device 1 includes a microphone 7-
1. Input means 7 including a keyboard 7-2, a mouse 7-3, etc., a main editor 2 having an acoustic model 5 and a speaker adaptation unit 6, a word n-gram (chain of n elements) dictionary The system includes a continuous speech recognition unit 3 provided with a word dictionary 3-1, a word speech recognition unit 4 provided with a plurality of word dictionaries 4-1 and a display device 8.

When a speech is input, the main editor 2 analyzes the input speech into a group of speech acoustic feature parameters using the acoustic model 5 and the speaker adaptation unit 6, and the continuously uttered speech is converted to the word n. The continuous speech recognition unit 3 uses the -gram dictionary 3-1 to recognize the speech, and the speech uttered as a word is converted to the word dictionary 3-
1, the main editor 2 creates a document and uses the display device 8
To be displayed. The main editor 2 has a function of switching between the continuous speech recognition unit 3 and the word speech recognition unit 4 by the input of the input unit 7, and a plurality of word dictionaries 4-4 provided in the word speech recognition unit 4.
It has a function to select 1.

FIG. 2 is an image diagram of a document creating apparatus according to the present invention. FIG. 2 shows a situation in which an input voice is converted into a character sequence by a voice recognition function to create a desired document. On the display device 8, the upper part shows the operation unit for operating the document creation command, and the lower part shows the recognition result, and shows that the document is being created, "I will contact you because the initial evaluation was performed." As one embodiment according to the present invention, S1: start and end continuous speech recognition by a continuous speech recognition start / end button (for example, a keyboard), and S2: start and end word speech recognition by a word speech start / end button. (Press the button at the start and press the button again at the end.) That is, with respect to the first aspect of the present invention, the present invention creates a document while combining continuous speech recognition and word speech recognition. Use continuous speech recognition to create documents such as commonly used phrases, and use word speech recognition to write proper nouns that are used only in special cases (for example, special names of people or places). , Create a document. This makes it possible to create a document more efficiently and without imposing a burden on the user as compared with the conventional document creation based on continuous speech recognition alone.

As another method of this embodiment, a method by selecting an operation menu is shown in FIG. For example, "insert"
The menu of S3 can be selected by selecting
An input operation is performed by selecting the "start continuous speech recognition" menu or the S4 "start word speech recognition" menu. During continuous voice input after menu selection, the “continuous voice recognition start” menu becomes a “continuous voice recognition end” menu, and the “word voice recognition start” menu becomes a “word voice recognition end” menu. By selecting them again, the voice input ends.

[0013] As another embodiment, an operation by voice,
That is, by uttering “start of continuous speech recognition” or “start of word speech recognition” (voice input), it is possible to use both recognitions efficiently. On the other hand, with regard to the word speech recognition according to the second aspect, before or after selection of the word recognition, a word dictionary to be recognized is selected, thereby narrowing down the vocabulary to be recognized and improving the recognition performance.

In FIG. 4, by pressing the ▼ button of the word dictionary, the type of the word dictionary appears, and one of them is selected. In FIG. 4, the word dictionary A is a person name dictionary, the word dictionary B is a place name dictionary, and the word dictionary C is a fish name dictionary. When the user utters “Ito”, “Ito” or “Ito” is selected if the personal name dictionary A is selected, “Ito” is selected if the place name dictionary B is selected, and the fish name dictionary C If "is selected," Itomi "will be output. FIG. 4 shows a state where the word dictionary A whose display is inverted is selected.

On the other hand, with respect to the all word dictionary of claim 3,
In the embodiment of FIG. 2, as shown in FIG. 5, a word dictionary A, a word dictionary B, a word dictionary C, and four
The user will select from one dictionary. Here, all the words in the word dictionaries A, B, and C are registered in the all word dictionary. This is used when the user does not know which dictionary entry contains the word to be output. Although the recognition accuracy and the processing time are lower than when the corresponding word dictionary is selected, it is possible to output the data reliably. For example,
In the above example, if you say "Ito", "Ito"
The candidates “Ito” and “Itomi” are output, and the user selects a candidate from the candidates.

On the other hand, with regard to the document creation method using the voice recognition function according to the fourth aspect, while the document is being created using the continuous voice recognition, for example, the user may use a voice command such as "word recognition". , The continuous speech recognition is interrupted to activate the word recognition unit, and when the word speech recognition is completed, the continuous speech recognition is resumed again, thereby efficiently creating a sentence. For example, "City name:
FIG. 6 shows a procedure when it is desired to create a sentence “City in Shizuoka Prefecture, Ito”.

First, continuous speech recognition is activated, and S7: "Toshinomae Koron" is uttered. At this point, "City name:" is displayed. Here, with the place name word dictionary selected, S9: “Word recognition mode” is uttered, and S1
0: By uttering “Ito”, the word recognition unit is automatically activated, and the uttered voice “Ito” is recognized,
S11: "City name: Ito" is output. At this point, the process has already resumed continuous speech recognition, and S1
2: "Speaking that you are in the space" is displayed, and "City name: Ito city in Shizuoka prefecture" is displayed, and the desired sentence can be created by ending the continuous speech recognition. By calling the word speech recognition during the continuous speech recognition as described above, it is possible to efficiently create a sentence.

As described above, by providing a sentence creation apparatus having a continuous speech recognition function and a word speech recognition function, a user can efficiently create a document by switching and using both functions. it can.

[0019]

As described above in detail, according to the present invention, in a document creation device (dictation device) using a speech recognition function, a document is created while efficiently switching between a continuous speech recognition function and a word speech recognition function. By creating the document, there is an advantage that a more excellent usability (convenience) can be provided by a conventional document creating apparatus using only the continuous speech recognition function.

[Brief description of the drawings]

FIG. 1 is a block diagram of a document creation device of the present invention.

FIG. 2 is an image diagram showing an example of switching between continuous speech recognition and word speech recognition by buttons of a recognition unit and document creation.

FIG. 3 is an image diagram showing an example of switching between continuous voice recognition and word voice recognition by menus of a recognition unit and a document creation example.

FIG. 4 is a diagram showing a selection menu for a plurality of word dictionaries.

FIG. 5 is a diagram showing a menu of word dictionaries including all word dictionaries in word speech recognition.

FIG. 6 is a diagram showing a procedure for creating a document by partially performing word speech recognition while performing continuous speech recognition.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Document creation apparatus 2 Main editor 3 Continuous speech recognition part 3-1 Word n-gram dictionary 4 Word speech recognition part 4-1 Word dictionary 5 Acoustic model 6 Speaker adaptation part 7 Input means 8 Display device

──────────────────────────────────────────────────続 き Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI Theme coat ゛ (Reference) G10L 15/10 G10L 3/00 531M 15/00 551B 15/22 571V (72) Inventor Katsutoshi Ohtsuki Tokyo 2-3-1 Otemachi, Chiyoda-ku Nippon Telegraph and Telephone Corporation F-term (reference) 5B009 KB01 MB06 ME12 5D015 AA01 BB01 HH06 HH12 KK01 KK03 LL08 LL10

Claims (4)

[Claims] An analyzing unit for analyzing the input speech into a group of speech acoustic feature parameters; a speech recognizing unit having a speech recognition function for performing recognition based on information on the feature parameters and linguistic information; A document creation device comprising a document creation unit for creating a document by using the speech recognition unit, wherein the speech recognition unit recognizes a continuously uttered speech, and a word speech recognition that recognizes a speech uttered as a word. A document creation apparatus comprising: an input unit that allows a user to arbitrarily switch between the two recognition units; and creates a document while the user switches between the two recognition units arbitrarily. 2. The apparatus according to claim 1, wherein said word speech recognition section includes a plurality of word dictionaries, and word dictionary selecting means for allowing a user to arbitrarily select one of the plurality of word dictionaries. The document creation device according to claim 1. 3. The document creation apparatus according to claim 2, wherein one of the plurality of word dictionaries includes an all word dictionary in which word items of all other word dictionaries are registered. 4. A continuously uttered speech having an analysis unit for analyzing an input speech into a group of speech acoustic feature parameters and a speech recognition function for performing recognition based on information of the feature parameters and linguistic information. A continuous speech recognition unit for recognizing, and a word speech recognition unit for recognizing a voice uttered as a word, input means capable of arbitrarily switching the two recognition units by a user, and both recognition units being optional by the user And a document creation device provided with a document creation unit for creating a document using the recognition result. Procedure 1 for creating a document using the continuous speech recognition unit; The procedure 2 interrupts the continuous speech recognition of the continuous speech recognition unit and activates the word speech recognition unit to perform the word speech recognition. The procedure 3 restarts the continuous speech recognition after the completion of the word speech recognition. Document creation method characterized by comprising.