JPH0486800A - Voice recognition system by displaying plural candidates - Google Patents

Abstract

PURPOSE:To enable a user to easy find out a correct answer by providing a means for displaying the plural recognition candidates obtd. as the recognition result of voice recognition processing in predetermined priority. CONSTITUTION:The priority displaying means 101 which displays the plural recognition candidates 102 obtd. as the recognition result of the voice recognition processing in accordance with the predetermined priority is provided. Namely, the recognition candidates are displayed in the state of, for example, orderly arranged capital letters of alphabets as the recognition candidates are displayed in the predetermined priority. The user is able to easily find out the correct answer in this way when the user desires to assign the candidate to an input voice among the displayed recognition candidates.

Description

[Detailed Description of the Invention] [Summary] Regarding the display method of multiple recognition candidates in a speech recognition device such as a large word speech recognition system, it reduces the time required for the user to search for the correct answer, and improves the user's psychological In order to reduce the burden, the present invention is configured to include a priority order display means for displaying a plurality of recognition candidates obtained as recognition results of speech recognition processing in a predetermined priority order.

Further, as a result of the voice recognition process, a high-rank candidate storage means for storing a predetermined high-rank recognition candidate among a plurality of recognition candidates obtained in descending order of recognition degree, and an order in which the recognition candidates are stored in the high-rank candidate storage means are predetermined. The apparatus is configured to have a sorting means for sorting according to a predetermined priority order, and a display means for displaying recognition candidates on the top candidate storage means sorted by the sorting means in j order.

[Industrial Field of Application] The present invention relates to a method for displaying a plurality of recognition candidates in a speech recognition device such as a speech recognition system with a large number of words.

[Prior art] In speech recognition devices such as Daigotosei speech recognition system,
Due to performance limitations of speech recognition methods, the recognition candidate with the best recognition score is not necessarily the correct answer. For this reason, a man-machine method is used in which a plurality of top candidates with high scores are displayed on a display or the like, and the user is asked to select the correct answer from among the displayed candidates.

FIG. 5 shows a configuration diagram of a conventional example of a large word speech recognition system.

After the voice to be recognized is input from the microphone 501,
The signal is digitized by the A/D converter 502 and subjected to frequency analysis by the frequency analyzer 503, thereby converting it into a time-series pattern of parameters representing the frequency characteristics of the phonemes of the voice.

This time series pattern and the time series pattern of each word stored in the template memory 5゜5 are D T W (
Dynamic Time Warping)? Matching is performed at the soching section 5°4. As a result, the distance between each word stored in the template memory 505 and the input speech (for example, the Euclidean distance between parameters) is calculated.

These distances are sorted (rearranged) in order of distance by a distance order sorting unit 506, and information regarding character strings of candidate words up to a certain rank with short distances is stored in an upper candidate memory 5.
07. For example, if the top 30 candidate words are stored, the distance order sorting unit 506 executes the operation flowchart shown in FIG. That is, first, the memory of the upper candidate memory 507 is initialized (S
l).

Thereafter, the character string and distance value of the candidate word are read from the DTW matching unit 504 (SZ) and written into the address area of memory No. 31 of the upper candidate memory 507 (s4). The distance value written to this address is sequentially compared with the distance value of the candidate words already written in the 1st to 30th address areas of the upper candidate memory 507 in descending order, and the distance value of the 31st address area is compared. Sorting processing including values is executed (s5). The operations in steps 52 to S5 above are repeated until there is no more information regarding candidate words from the DTW matching unit 504, that is, until the determination in step s3 becomes NO.

When the distance order sorting is completed, a sort completion signal is sent from the distance order sorting section 506 to the display control section 508. When the display control unit 508 receives the sorting end signal, it displays on the display 509 the character strings of candidate units stored in the order of distance in the upper candidate memory 507 and their rank numbers in order.

For example, in the case of the above-mentioned example in which processing is performed for the top 30 positions, the top candidate memory 50 is
The character strings of candidate words stored in the 1st to 30th address areas of No. 7 are displayed. The display control unit 508
When the display ends, a display end signal is sent to the word selection control section 510.

When the word selection control unit 510 receives the display end signal,
The ranking number of the candidate word inputted by the user through the keyboard 511 is received, and the character string of the correct word corresponding to the inputted ranking number is output to another device or an application program (not particularly shown).

[Problem to be solved by the invention]

FIG. 7 shows a display 509 in the above-mentioned conventional example.
An example of candidate words displayed is shown below.

As shown in the figure, the user selects the word r0O5A corresponding to the input voice from among the displayed candidate words.
Even if I want to specify KA J, I cannot find the desired word in Nagana. As described above, the conventional example has a problem in that it takes a long time for the user to search for the correct answer, which places a large psychological burden on the user and makes him/her prone to stress.

The present invention aims to shorten the time required for the user to search for the correct answer and reduce the psychological burden on the user.

[Means to solve the problem]

FIG. 1(a) is a block diagram of a first aspect of the present invention.

The first aspect includes a priority order display means 101 that displays a plurality of recognition candidates 102 obtained as recognition results of speech recognition processing in accordance with a predetermined priority order.

FIG. 1(b) is a block diagram of a second aspect of the present invention, which is a further embodiment of the first aspect.

The second aspect includes an upper candidate storage means 103 that stores a predetermined higher recognition candidate 106 among a plurality of recognition candidates obtained in descending order of recognition degree as a result of the speech recognition process.

Next, it has a sorting means 104 that rearranges the storage order of the recognition candidates 106 in the upper candidate storage means 103 according to a predetermined priority order. The means performs sorting, for example, in dictionary order (character string code order) or in the order of the initial letter of each recognition candidate 106.

Furthermore, it has a display means 105 for sequentially displaying the recognition candidates 106 on the top candidate storage means 103 sorted by the sorting means 104. Here, the means stores the top candidate storage means 10 sorted by the sorting means 104.
When displaying the recognition candidates 106 above in order, the recognition candidates with a high possibility of being correct and the recognition candidates with a low possibility of being correct are displayed by changing, for example, the brightness, display color, or font, or by inverting the recognition candidates. It is also possible to configure the display to be distinguished by, etc. Further, the same means includes sorting means 10
When displaying in order the recognition candidates 106 on the top candidate storage means 103 sorted by 4, the recognition candidate with the highest possibility of being correct may be displayed.

[For production]

In the present invention, the recognition candidates are displayed in a predetermined priority order, so that, for example, the initial letters of Alphahet are displayed neatly aligned, and the user can select the input voice from among the displayed recognition candidates. When you want to specify a candidate corresponding to a question, you can easily find the correct answer.

In addition, in the display state described above, it is possible to display recognition candidates that are likely to be correct and those that are not, or to place a cursor or the like in advance at the position of a recognition candidate that is likely to be correct. By configuring it to be moved, the user can work out the correct answer even more quickly.

〔Example〕

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 2 is a configuration diagram of an embodiment of the present invention. In the figure, parts with the same numbers as in the conventional example of FIG. 5 have the same functions.

The difference between this embodiment and the conventional example is that the distance order sorting unit 506
The sorting end signal (hereinafter referred to as distance sorting end signal in this embodiment) from is inputted to the display control section 508 (dictionary ordering sorting section 201. In other words, the dictionary ordering sorting section 201 Upon receiving the distance order sorting end signal, the character strings of candidate words stored in the distance order in the upper candidate memory 507 are sorted again in dictionary order. Then, the dictionary order sorting unit 201 ends the dictionary order sorting process. Then, the display control unit 508 sends a dictionary order end signal to the display control unit 508. When the display control unit 508 receives the sort end signal, the display control unit 508 stores the character strings of the candidate units stored in the dictionary order in the upper candidate memory 507 and their rank numbers. They are displayed on the display 509 in order.

The subsequent operations are similar to those in the conventional example shown in FIG.

In the above-described embodiment, the operation of this embodiment when the top 30 candidate words are obtained from the distance sorting unit 506 in order of distance and are sorted and displayed in dictionary order, that is, in ascending order of character code. A flowchart is shown in FIG.

First, steps 51 to 51 are executed by the distance order sorting unit 506.
The operation at S5 is the same as in the conventional example shown in FIG. As a result, character strings of candidate words and distance values are obtained in the 1st to 30th address areas of the upper candidate memory 507 in order of distance.

Next, from the distance order sorting unit 506 to the dictionary order sorting unit 201
When the distance order sorting end signal is input to
Sort the candidate words stored in , in ascending order of character code.

First, pointers I, J, and M for pointing to the 1st to 30th address areas in the upper candidate memory 507 are prepared, and the starting address value 1 of the upper candidate memory 507 is placed in the pointer I (S7). Pointer I to pointer M
The contents of pointer J and pointer M plus 1 are respectively set (S,). That is, pointer M is set to the first address value 1, and pointer J is set to the next address value 2.

From this state, while the value of pointer J is incremented by +1 (S+□), the code value of the character string stored at the address indicated by pointer M and the character string stored at the address indicated by pointer J (S,) is compared with the coat value of (S,).

If the code value of pointer J is smaller than the code value of pointer M, pointer M moves to the address indicated by pointer J (S, -S, O).

The above operation is repeated until the value of pointer J reaches 30, that is, until the last address of upper candidate memory 507 is reached (S1+).

When the last address is reached, the code value of the character string at the address indicated by pointer M is the smallest code value among the code values of each character string of the 30 candidate words. Therefore, by swapping the character string at the address indicated by pointer M and the character string at the first address indicated by pointer I, the candidate word that is the character string with the smallest code value is placed at the first address. is stored.

Subsequently, the value of pointer I is incremented by 1 (S1s), and M
=I, J=M+l (S8), S again.

The processing operations of ~S, 2 are repeated. Then, when the judgment of Sl+ becomes YES, the code value of the character string at the address indicated by pointer M is equal to the code value of each character string of the 29 candidate words excluding the character string with the minimum code value. This is the smallest code value. Therefore, pointer M
By swapping the character string of the address indicated by and the character string of the second address indicated by pointer I, the candidate word whose code value is the second smallest character string is placed at the second address. is stored.

By repeating the above operation while incrementing the value of pointer I by +1 (S + 5),
When the value of pointer I becomes 29 (when the judgment of S and a becomes No), characters with small code values are selected from the 1st address to the 30th address of the upper candidate memory 507. Candidate words are stored in order from column to column.

When the above sorting process in the dictionary order sorting unit 201 is completed, a dictionary order sorting completion signal is sent from the dictionary order sorting unit 201 to the display control unit 508.

As a result, the display control unit 508 performs step 5 in FIG.
16, the character strings of candidate words stored in the 1st to 30th address areas of the upper candidate memory 507 are displayed on the display 509.

Candidate word display 509 based on the above processing operations
An example of the display is shown in FIG. As can be seen from the figure, when the initial letters of alpha heads are displayed neatly aligned and the user wants to specify, for example, the word r 0O5AKA J corresponding to the input voice from among the displayed candidate words. You can easily find the correct answer.

In the embodiment described above, the dictionary order sorting unit 201 performs the sorting in the dictionary order, that is, in the ascending order of the code of the character string, but when handling kanji etc., the arrangement may be such that the sorting is performed in the order of the first letter. You can also do it. In this case, instead of the dictionary order sorting unit 201, a sorting unit that sorts by Kanji code may be provided.

Further, in the display state as shown in FIG. 4, words that are likely to be correct words, that is, words that have a small distance value, and words that are not likely to be correct words may be displayed separately. In this case, it is sufficient to leave in the top candidate memory 507 the results first sorted in distance order by the distance order sorting unit 506, and then send the ranking to the display control unit 508 for control. In this case, the display can be differentiated by, for example, changing the brightness, display color, and font depending on the distance ranking.
Possible methods include displaying the image in reverse video.

In addition, when the word selection control unit 510 accepts the ranking number of the candidate word inputted by the user from the keyboard 511, the word selection control unit 510 places the ranking number of the candidate word input by the user from the keyboard 511 in the position of the word that is likely to be the correct word.
It is also possible to control the movement of the cursor in advance. Note that if a pointing device such as a mouse is used instead of the keyboard 511, the operability will be further improved.

[Effects of the Invention] According to the present invention, recognition candidates are displayed in a predetermined priority order, so that, for example, the initial letters of Alpha and G are neatly aligned, and the user can When you want to specify a candidate that corresponds to the input voice from among the recognized recognition candidates, you can easily find the correct answer.

In addition, in the display state described above, it is possible to display recognition candidates that are likely to be correct and those that are not, or to place a cursor or the like in advance at the position of a recognition candidate that is likely to be correct. By configuring it to be moved, the user can more quickly figure out the correct answer.

[Brief explanation of the drawing]

FIGS. 1(a) and 1(b) are the Pronk diagram of the present invention, and the second
3 is an operational flowchart of this embodiment when candidate words are displayed in dictionary order. FIG. 4 is a diagram showing a display example of this embodiment. Figure 5 is a configuration diagram of the conventional example, Figure 6 is an operation flowchart of the conventional example, and Figure 7 is a diagram showing display examples of the conventional example. . ...Priority order display means, ...Multiple recognition candidates. ...Top candidate storage means, ...Sort means, ...Display means, ...Recognition candidates.

Claims (1)

[Claims] 1) In a display method of a plurality of recognition candidates in a speech recognition device, a priority order in which a plurality of recognition candidates (102) obtained as recognition results of speech recognition processing are displayed in accordance with a predetermined priority order. A multi-candidate display speech recognition system characterized by comprising a display means (101). 2) In a method for displaying a plurality of recognition candidates in a speech recognition device, a high-ranking candidate storage unit stores a predetermined high-ranking recognition candidate (106) among a plurality of recognition candidates obtained in descending order of recognition degree as a result of speech recognition processing. (103), a sorting means (104) for rearranging the storage order of the recognition candidates (106) in the upper candidate storage means according to a predetermined priority order, and the upper candidate storage sorted by the sorting means. A multi-candidate display speech recognition system, comprising: a display means (105) for sequentially displaying recognition candidates (106) on the means (103); 3) The multi-candidate display speech recognition system according to claim 2, wherein the sorting means rearranges the storage order of the recognition candidates in the top candidate storage means in dictionary order. 4) The multi-candidate display speech recognition system according to claim 2, wherein the sorting means rearranges the storage order of the recognition candidates in the top candidate storage means in the order of initial letters of each recognition candidate. 5) When displaying in order the recognition candidates on the top candidate storage means sorted by the sorting means, the display means displays recognition candidates with a high probability of being correct and recognition candidates with a low probability of being correct. 3. The multi-candidate display speech recognition system according to claim 2, wherein the multiple-candidate display speech recognition system displays the candidates separately. 6) The display means, when sequentially displaying the recognition candidates on the top candidate storage means sorted by the sorting means, instructs to display a recognition candidate with the highest possibility of being correct. The multiple candidate display speech recognition method according to claim 2.