JPH0374426B2 - - Google Patents

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to an audio word processor, and in particular, to a plurality of types of displays corresponding to the certainty of the input when displaying text data input by voice, for example. The present invention relates to an audio word processor that allows an operator to predict the changes or effects that will occur due to the next operation by displaying them in different formats, thereby making it possible to efficiently create documents.

[Conventional technology and problems]

Recently, voice recognition technology has entered the stage of practical application.
Consideration is being given to inputting data into a Japanese word processor using voice input.

In such voice word processors, the part that performs voice recognition of the audio reading text data and the part that processes kana-kanji conversion are generally installed independently, and the part that processes kana-kanji conversion starts from the voice recognition part. Consonant (C) / Vowel (V)
Data regarding CV syllable candidate sequences consisting of combinations of are delivered. Then, by comparing ambiguous character strings made up of candidate syllable combinations with a word dictionary, invalid character strings are eliminated,
Appropriate character strings are sequentially displayed on the display. In particular, when extracting appropriate characters, kana-kanji conversion is often performed at the same time.

However, according to the conventional method, although the reading of the proper character sequence obtained from the ambiguous character string or the result of its kanji conversion is displayed sequentially in a predetermined priority order, from the operator's point of view, it is difficult to There was a problem in that it was impossible to predict what would be displayed, and the candidates had to be displayed in order, starting from the first candidate until the correct answer was reached. In particular, in the case of input that is unclearly uttered, it is not possible to obtain information on whether it is better to re-enter the input or make corrections to the conversion, so the best operation at that time is not available. I had a problem with not being able to choose.

[Means for solving problems]

The present invention aims to solve the above-mentioned problems, and displays information that makes it possible to predict the results of a plurality of instruction operations that the operator can select, together with the reading of the input character string or the result of its kanji conversion, to provide the optimal instruction. By selecting , you can quickly create Japanese documents.

That is, when inputting text data, the audio word processor of the present invention converts the reading voice of the data into a CV syllable candidate series based on voice recognition,
A voice word processor that creates a Japanese document by performing kana-kanji conversion processing using dictionary matching on the sequence includes a parameter extraction unit that extracts feature parameters of input speech, and feature parameters extracted by the parameter extraction unit. a speech recognition unit that selects candidate syllable sequences by comparing them with standard patterns previously stored in a dictionary based on a dictionary collation unit that searches the word dictionary for ambiguous character strings generated based on the above and extracts appropriate characters; and a dictionary matching unit that searches the word dictionary for ambiguous character strings generated based on the above and extracts appropriate characters; The present invention is characterized by comprising an instruction control unit that performs control, and a display selection unit that displays information that predicts the display content that will change in response to the input operation that will be selected next. Hereinafter, embodiments will be described with reference to the drawings.

〔Example〕

FIG. 1 is a block diagram of an embodiment of the present invention, FIG. 2 is an example of processing by the graphical display selection unit shown in FIG. 1, FIG. 3 is a diagram explaining the counting range of the number of syllable types,
FIG. 5 is a diagram explaining a display mode according to an embodiment of the present invention, FIG. 5 is a diagram explaining a display mode according to another embodiment of the present invention, FIG. FIG. 7 shows an explanatory diagram of a display that foretells a change in kanji.

In FIG. 1, reference numeral 1 denotes a microphone for inputting voice, 2 denotes a voice parameter extraction unit, 3 denotes a voice dictionary storing standard patterns related to voice characteristic parameters, and 4 denotes a voice recognition unit that performs continuous voice recognition or discrete voice recognition. 5 is a syllable latex storage unit in which the recognition result by the speech recognition unit 4 is stored; 6 is a syllable latex storage unit that generates ambiguous character strings based on combinations of candidate syllables stored in the syllable latex storage unit 5 and selects a conversion target. Conversion target selection section, 7 is an ambiguous character string buffer where ambiguous character strings to be converted into proper characters are stored, 8 is a word dictionary in which proper characters mixed with readings and kanji are registered in advance, 9 is a dictionary check 10 is an instruction control unit that controls instruction operations; 11 is a keyboard; 12 is a screen pointing device such as a mouse; 13 is a collation result storage unit that stores the collation results from the dictionary collation unit 9; A display selection section controls the display of the matching results in a display format that predicts changes in response to operations; 15 is a display; 16 is a reading display area on the display screen; 17 is a kanji display area in which the matching results are displayed.

When the voice that reads text data is input from the microphone 1, the voice parameter extraction unit 2 extracts a time series of voice characteristic parameters related to spectrum information and the like. The speech recognition section 4 converts the characteristic parameters of the unknown input speech extracted by the speech parameter extraction section 2 and the standard patterns stored in the speech dictionary 3, for example, in units of monosyllables, into a DP, for example.
This is the recognition result after matching using matching method etc.
The CV syllable candidate series is stored in the syllable latex storage unit 5 according to the candidate ranking. Note that distance information between each candidate syllable and the standard pattern is also stored in correspondence.

The conversion target selection unit 6 generates an ambiguous character string by combining candidate syllables based on the speech recognition results stored in the syllable latex storage unit 5, and stores the ambiguous character string in an ambiguous character string buffer 7. Combinations of candidate syllables are selected such that, for example, those within a predetermined distance or a predetermined number of combinations are selected in the order of the minimum sum of distances from the standard pattern of each syllable. Furthermore, if necessary, ambiguous character strings based on predetermined reading change rules may be incorporated, such as changing the pronunciation of "wa" to "ha" and the pronunciation of "e" to "he". The ambiguous character strings stored in the ambiguous character string buffer 7 are sequentially selected by the conversion target selection section 6 and delivered to the dictionary collation section 9.

The dictionary collation unit 9 divides the passed ambiguous character string into appropriate character strings as necessary, and searches the word dictionary 8 using the pronunciation as a key. In the word dictionary 8, proper characters or words including proper readings of characters and kanji characters corresponding to the readings are registered in advance. When the word dictionary 8 contains appropriate characters, etc., the matching results are stored in the matching result storage section 13.
Stored sequentially in .

The display selection section 14 displays each verification result stored in the verification result storage section 13 on the screen of the display 15. At this time, various display forms as described later are used to present information for determining what operation the operator should perform next to obtain an efficient and optimal result. The reading display area 16 of the display screen displays the reading of the selected ambiguous character string, and the kanji display area 17 displays the result of the kana-kanji conversion.
Note that the reading display area 16 and the kanji display area 17
does not necessarily have to be a fixed area on the display screen.

The instruction control unit 10 includes, for example, a keyboard 11,
Screen pointing device 12 such as a so-called mouse
The conversion target selection unit 6 receives, for example, an instruction to change the reading displayed in the reading display area 16, an instruction to repeat kanji conversion, an instruction to cancel the current input and re-input the voice, etc. It performs a process of notifying the dictionary collation unit 9, display selection unit 14, etc. That is, the operator can select, via the instruction control unit 10, how to handle text data input by voice.

Next, according to Figures 2 to 4, when displaying the input sentence data, regarding the displayed syllables,
An example will be described in which a plurality of types of display formats are displayed depending on the certainty of the input. In this example, the number of different syllables in the syllable sequence to be displayed is used as information regarding the certainty of the input. , the operator is made aware that it is better to make the correction directly from the keyboard 11 or the like, or to speak again, making it possible to speed up the correction speed.

Whether or not each syllable in the displayed syllable series is likely to change in the next operation or several operations is related to the number of times different syllables appear within a certain range. For example, ambiguous string buffer 7
It is assumed that the ambiguous character strings stored in . If we focus on the appropriate rankings from the highest to the lowest, there are two types of syllables for the first sound: "fu/ku". In the same range, the second note is "ji/gi/chi"
There are three types of syllables. Therefore, there is a greater possibility that the second note will change than the first note. Similarly, the third sound is one type of "tsu", so
At least four normal operations will result in no change.

In order to count the number of syllables, the display selection section 14 shown in FIG. 1 performs processing as shown in FIG. 2, for example. That is, first, in process P1, the first syllable is first pointed to as the processing target. Next, process P2
to set the count range. In the example shown in FIG. 3, the count range is set to "5", and for example, for the first display, the count range is from the first place to the second place. Then, in process P3, it is first determined whether the syllable of the highest rank is a new syllable type. Since the first syllable "fu" is always new, the number of syllable types is set to "1" in process P4. Then, according to the determination in process P5,
Process P3 and process P4 are repeated until the count range ends. The number of syllable types is counted up with the syllable of the digit "ku" and becomes "2".

When the counting for the first syllable is completed, the second syllable is pointed to in the next process P6, and processes P2 to P6 are similarly repeated based on the determination and branching of process P7.
Then, each syllable is processed in sequence, and when the final syllable is completed, a display color corresponding to the number of types of syllables is selected for each syllable and displayed on the display in process P8.

This display is performed, for example, as shown in FIG. 4A. That is, the first display is "Fujitsu", which has the highest phonetic suitability ranking, and the first syllable is displayed in yellow, for example, because the number of syllable types is two. Second
The syllable "ji" has three types of second syllables, so it is displayed in red, the third syllable "tsu" is displayed in blue, and the fourth syllable "u" is displayed in yellow. Looking at this, it can be seen that the second syllable is recognized most vaguely, and the certainty of the second syllable is small. Note that in this example, the display format corresponds to color, and the number of syllable types is represented by a change in color, but the display format type may be, for example, brightness, inverted display, blinking, or the like. The same applies to the embodiments described later.

When a display as shown in FIG. 4A is made and a normal rereading instruction operation is performed, "Fugitsu", which has the highest audio appropriateness ranking, is displayed. The count range in this second display is from the digit to the digit as shown in FIG. 3, and each syllable is displayed in the color shown in FIG. 4 (b).

FIG. 5 shows an example in which the certainty in syllable recognition is determined by the degree of similarity based on the distance to the standard pattern calculated at the time of syllable recognition. That is, an example is shown in which different display forms such as colors are used depending on the degree of similarity of each syllable.

For example, say “fujitsuu” and say “fujitsuu”.
When I was trying to input the reading, the fifth
Assume that "Ujitsu" is displayed as a recognition result as shown in Figure A. Looking at the information on the degree of similarity based on color, etc., the degree of similarity of the first sound "u" is the smallest. Therefore, it can be determined that if the operation of specifying the next ambiguous character string is performed, there is a high possibility that the first sound will change and a correct result will be obtained.

On the other hand, suppose that, for example, the same input is displayed as shown in FIG. 5B. Looking at this similarity, the similarity of the incorrect first sound "u" is large,
The similarity of the correct fourth sound "u" is small. If the similarity of the erroneous syllable is large, there is a small possibility that it will be corrected with the next ambiguous character string. Therefore, it is not appropriate to sequentially replace ambiguous character strings, and it is better to directly instruct the correction of the first character, for example, by re-entering the voice or by inputting from the keyboard 11 or screen pointing device 12. I know it's good.

FIG. 6 shows an example of display in yet another embodiment. For example, suppose that ambiguous character strings are stored in the ambiguous character string buffer 7 in the order shown in FIG. The first display in the reading display area 16 of the display screen is "Fujitsu", which has the highest syllable suitability ranking. When performing this display, the illustrated display selection unit 14 in FIG. 1 reads in advance the next candidate, the ambiguous character string "Fujitsu", and extracts the portion that differs from the currently displayed "Fujitsu". Then, a change notice display 20, such as an underline, is attached to the position where the change occurs. Similarly, when displaying the second-ranked "fugitsu", a change preview display 20 is attached to the first and second sounds in comparison with the highest-ranked "kujitsu". The same applies to the third and subsequent displays.

While FIG. 6 shows an example of changing readings, FIG. 7 shows an example of displaying when changing kanji. It is assumed that the contents of the verification result storage section 13 are as shown in FIG. The first display is "Fujitsu." On the other hand, if the second kana-kanji conversion result is "Fujitsu", the kanji replacement instruction will change "Fuji" to "Fuji", so "Tomi" and "Shi" will be changed. A change notice display 21 is attached to each character. The same applies to the following displays. This allows the operator to predict changes caused by the next operation. Note that by displaying the number of remaining operations/number of remaining candidates, etc., the operator can more appropriately judge whether it is better to continue replacing readings or kanji, or whether it is better to re-enter.

〔Effect of the invention〕

As explained above, according to the present invention, the operator can predict changes that will occur due to the next operation. Therefore, the operator can select an appropriate operation means and can quickly display the desired Japanese text.

[Brief explanation of drawings]

FIG. 1 is a block diagram of an embodiment of the present invention, FIG. 2 is an example of processing by the graphical display selection unit shown in FIG. 1, FIG. 3 is a diagram explaining the counting range of the number of syllable types,
FIG. 5 is a diagram explaining a display mode according to an embodiment of the present invention, FIG. 5 is a diagram explaining a display mode according to another embodiment of the present invention, FIG. FIG. 7 shows an explanatory diagram of a display that foretells a change in kanji. In the figure, 2 is a speech parameter extraction unit, 3 is a speech dictionary, 4 is a speech recognition unit, 5 is a syllable latitude storage unit,
6 is a conversion target selection part, 7 is an ambiguous character string buffer,
8 is a word dictionary, 9 is a dictionary collation unit, 10 is an instruction control unit, 11 is a keyboard, 12 is a screen pointing device, 13 is a collation result storage unit, 14 is a display selection unit, 15 is a display, 20 and 21 are change notices Represents a display.

Claims (1)

[Claims] 1. Regarding the input of text data, the reading voice of the data is converted into a CV syllable candidate series based on voice recognition, and the series is converted into a Japanese document by a kana-kanji conversion process using dictionary matching. In the speech word processor that creates the input speech, there is a parameter extraction section that extracts feature parameters of the input speech, and candidates are created by comparing the feature parameters extracted by the parameter extraction section with standard patterns stored in a dictionary in advance. A speech recognition unit that selects a syllable sequence, a word dictionary in which information regarding each proper character is registered, and the word dictionary is searched for ambiguous character strings generated based on the recognition results by the speech recognition unit to extract proper characters. an instruction control unit that detects an input operation for an instruction including at least a voice re-input instruction, a rereading instruction, and a kanji replacement instruction and controls the instruction content; A voice word processor comprising: a display selection section that displays information for predicting changing display content. 2 The information that predicts the changing display contents mentioned above is
The speech word according to claim 1, characterized in that the speech word has a plurality of display formats selected based on the number of different syllables in the syllable series to be displayed or the similarity of each syllable to a standard pattern. processor. 3 The information that predicts the changing display contents mentioned above is
2. The audio word processor according to claim 1, wherein the display information is a preview of characters to be replaced in response to the reading replacement instruction and/or the kanji replacement instruction.