JPH09138695A - Voice recognition device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To make possible studying all similar words by oneself from a formed word dictionary of large vocabularies. SOLUTION: A similar word detection means 16 gains a certain word among the total number of words in the word dictionary 14 formed using a word forming means 17 after obtaining a recognition result, and decides automatically the similar word by comparing the gained word with all remaining words excepting the word in the checks of (a) check on one character difference, (b) check on one character omission, (c) choked sound check, (d) check on devocalized word and (e) check on vowel coincidence. Then, the similar word detection means 16 outputs the detection result of the obtained similar word to an output device 15, and stimulates a user to notice (imparts a warning to a user).

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice recognition device, and more particularly to a voice recognition device for recognizing a word uttered by a user.

[0002]

2. Description of the Related Art Conventionally, after converting a word voice uttered by a user into a digital signal, a feature is extracted by an analysis unit, and the feature is matched using a word dictionary to obtain a recognition result. 2. Description of the Related Art There is known a voice recognition device configured to output a spoken word recognition result to an output device.

In such a voice recognition device, in order to check a similar word in a word dictionary, a user generally performs a recognition process, and searches a similar word in the word dictionary from a word that is erroneously recognized among the recognition results.・ Judging or investigating all the words entered by the user to judge similar words.

[0004]

In the above speech recognition apparatus, if there are many similar words, erroneous recognition is likely to occur and the recognition rate is lowered, so it is necessary to look up similar words in the word dictionary. However, the number of words recognized by a speech recognition device has been about tens to hundreds of words in the past, but now it is possible to recognize a large vocabulary of more than 1,000 words. It has become difficult to look up all the similar words in the dictionary.

[0005] The present invention has been made in view of the above points,
An object of the present invention is to provide a voice recognition device that can automatically find similar words from a created large vocabulary word dictionary.

[0006]

In order to achieve the above object, the present invention converts a word voice uttered by a user into an electric signal, extracts a feature, and matches the feature with a word dictionary. In a voice recognition device configured to output a recognition result by the user, the word recognition result uttered by the user to the output device, a word dictionary creating means for creating a word dictionary by previously inputting a plurality of words in the word dictionary, To detect similar words by sequentially comparing one word obtained from all the words in the word dictionary and all the remaining words according to a predetermined rule,
It is configured to have similar word detecting means for acquiring and repeating each of all the words in the word dictionary and outputting the obtained detection result to the output device to notify the user.

Here, the similar word detecting means in the present invention compares one acquired word and all the remaining words in kana notation among all the words in the word dictionary, and the kana notation difference between the compared words is different. Detecting a similar word by one character difference, one missing character, presence of consonants, existence of syllables that are easily devoiced, or a combination of any two or more rules of matching vowels of each syllable. Is characterized by.

According to the present invention, the similar word detection means detects one of all the words in the word dictionary and all the remaining words sequentially according to a predetermined rule to detect the similar word. It is possible to automatically detect whether or not there is a similar word relationship between all the words in the word dictionary by acquiring and repeating each of all the words. Further, in the present invention, since the above-mentioned detection result of the similar word is output to the output device to notify the user, the user can be warned of the existence of the similar word.

[0009]

BEST MODE FOR CARRYING OUT THE INVENTION Next, an embodiment of the present invention will be described with reference to the drawings. FIG. 1 shows a block diagram of an embodiment of a voice recognition device according to the present invention. This voice recognition device includes a microphone (microphone) 10, an A / D converter 11, an analysis unit 12, a recognition processing unit 13, a word dictionary 14, an output device 15, a similar word detection unit 16 and a word dictionary creation unit 17. The feature is that the similar word detecting means 17 is provided.

To explain the outline of the operation of this embodiment, an electric signal (voice signal) obtained by acoustic-electrically converting the voice uttered by the user by the microphone 10 is A /
The signal is converted into a digital signal by the D converter 11 and is input to the analysis unit 12, where the feature extraction of voice is performed by a known means.

On the other hand, the word dictionary 14 stores words created by the word dictionary creating means 17 in advance. The contents are created in Kana notation as shown in FIG. The recognition processing unit 13 matches the features extracted by the analysis unit 12 with the words in the word dictionary 14, and obtains the recognition result of the voice uttered by the user. The recognition result is output to the output device 15 and notified to the user.

Although the above speech recognition operation is the same as the conventional one, in this embodiment, before the above recognition result is obtained, the similar word detecting means 16 detects the similar words of the word group stored in the word dictionary 14. It is characterized in that the detection is automatically performed, and the result is output to the output device 15 to call the user's attention (give a warning).

Next, the operation of the similar word detecting means 16 which is the main part of this embodiment will be described in more detail with reference to the flowcharts of FIGS. The similar word detecting means 16 obtains the total number N of words in the word dictionary 14 created by the word creating means 17 (step 3 in FIG. 3).
1). At this time, the variable M in step 32 described later is set to the initial value "1". Subsequently, the Mth (M = 1 to 1
N) words are acquired (step 32 in FIG. 3), and the acquired M-th word and all other (N-1) remaining words are (a) checked for one character difference, (b) ) A similar word is determined by comparison by checking for missing one character, (c) consonant check, (d) unvoiced word check, and (e) vowel match check (step 33 in FIG. 3).

The similar character check rule (a) in the step 33, which is one character difference check, is a rule in which two words are regarded as similar words when only one character in kana notation is different in reading. Further, it is also possible to consider a means of judging similar words based on the ratio of the number of characters that matches the total number of characters instead of the difference of one character.

The check for missing one character in (b) of the rule for checking similar words in the above step 33 is 3 in FIG.
2 like the word “Kanagawa” and the fourth word “Kagawa”
It is a rule that if one kana notation of one word is missing, it is regarded as a similar word. This rule is described in (c) and (d) below.
The rule is also included, and when narrowing down the reason for similar words, it can be realized by removing from the check method.

The grammatical sound check (c) of the similar word check rule in the above step 33 is a rule to regard a similar word when the difference in the kana notation of each word is a grammatical sound. Further, considering that the consonant is likely to be omitted when uttered, the consonant of the 15th word "Sapporo" in FIG. 2 can be removed and the word can be checked again as the 16th word "Sapollo".

The check of the unvoiced word in (d) of the rule for checking similar words in step 33 is performed as shown in 7 of FIG.
The word "Seijo" and the eighth word "Ijou", 9
The word "Suido" and the tenth word "Ido", 11
"Sa line" or "Ha line" is the word with the smaller number of characters, such as the word "Synth" and the word 12 "Fu Shinsetsu".
It is a rule to consider similar words when there is a syllable that is easily devoiced.

Further, the vowel matching check (e) of the rule for checking similar words in step 33 is performed as shown in FIG.
This is a rule to consider similar words when the number of characters of both words is the same and the vowels of each syllable are the same, such as “Aichi” of the 13th word and “Haiki” of the 14th word.

The above five rules (a) to (e) can be freely combined. For example, except for (b) and (e), (a), (c) and (d) are excluded. It is also possible to check similar words with).

When the similar word detecting means 16 acquires "Wakayama" as the first word (M = 1) in the word dictionary 14 in step 33, for example, as shown in FIG.
Check according to the word "Okayama" and the rules (a) to (e) above. Here, the result of checking that "Okayama" and "Okayama" are similar words according to the rule of (a) is obtained.

Then, the first word "Wakayama" currently acquired is checked in accordance with the third word "Kanagawa" shown in FIG. 2 and the above rules (a) to (e). In this case, a check result that "Kanagawa" is not similar to "Wakayama" is obtained even if any of the rules (a) to (e) is applied.

Subsequently, the currently acquired first word "Wakayama" is checked in accordance with the fourth word "Kagawa" shown in FIG. 2 and the above rules (a) to (e). In this case, a check result that "Kagawa" is not similar to "Wakayama" is obtained even if any of the rules (a) to (e) is applied. Thereafter, in the same manner as above, "Wakayama"
Iterate until all the remaining N-1 words except for are checked.

When the check result of the first similar word is obtained, the check result is output to the output device 15 (step 34 in FIG. 3). It should be noted that regardless of whether the check results are similar or not, the check result may be output to the output device 15 each time one check result is obtained, or may be output only when the check results are similar.

Next, the similar word detecting means 16 uses the word dictionary 1
It is determined whether the check of the total number N of words in 4 has been completed (step 35 in FIG. 3). Since only the check result of the first similar word is obtained here, the value of the variable M is incremented by "1" and the value of M is "2".
(Step 36 in FIG. 3) and then step 32 in FIG.
Return to and get the second word "Okayama".

Next, the process of step 33 is repeated until the check of all the remaining N-1 words except "Okayama" is completed, and the check result is output to the output device 15 by step 34. Thereafter, in the same manner as described above, when all the words are acquired and the processing for checking the similar words and the output to the output device 15 are performed, the detection processing ends. Note that the check result may not be sequentially output to the output device 15, but the result may be held and output at the end of all word checks.

As described above, in this embodiment, the similar word detecting means 16 automatically investigates the similar words of all the words in the word dictionary 14, and the result is output to the output device 15. The user can be alerted. Further, the user can delete the similar word or correct the replacement based on the output result to the output device 15 by using the word dictionary creating means 17. As a result, a plurality of words used for recognition have no similar words, and the recognition rate can be improved as compared with the conventional case.

[0027]

As described above, according to the present invention,
Since it is possible to automatically detect whether or not there is a similar word relationship between all the words in the word dictionary, the similar word can be accurately detected even if the word dictionary is a large vocabulary word dictionary of 1,000 words or more. Further, according to the present invention, since the detection result of the similar word is output to the output device to notify the user and the user is warned about the existence of the similar word, the user does not perform the recognition, In addition, the user can easily delete or correct (replace) similar words without investigating all the words created by himself, which eliminates the similar words from the multiple words used for recognition. The recognition rate can be improved.

[Brief description of the drawings]

FIG. 1 is a block diagram of an embodiment of the present invention.

FIG. 2 is a diagram showing words in a word dictionary for explaining the operation of the similar word detecting means in FIG.

FIG. 3 is a flowchart for explaining the operation of the similar word detecting means in FIG.

[Explanation of symbols]

 10 Microphone (microphone) 11 A / D converter 12 Analyzer 13 Recognition processing unit 14 Word dictionary 15 Output device 16 Similar word detecting means 17 Word dictionary creating means

Claims (2)

[Claims] 1. A result of word recognition uttered by a user after converting a word voice uttered by a user into an electric signal, extracting a feature, and matching the feature with a word dictionary to obtain a recognition result. In a voice recognition device configured to output to an output device, a word dictionary creating means for creating a word dictionary by previously inputting a plurality of words in the word dictionary, and one word obtained from all the words in the word dictionary. Detecting similar words by sequentially comparing all the remaining words according to a predetermined rule is repeated for each of all the words in the word dictionary, and the obtained detection result is output to the output device for use. And a similar word detecting means for notifying the person. 2. The similar word detecting means compares one acquired word and all the remaining words in kana notation among all the words in the word dictionary, and the difference in kana notation between the compared words is one character. Difference, 1 character omission, presence of consonant, presence of syllable that tends to be unvoiced, matching of vowels of each syllable or any combination of two or more rules is detected as a similar word The voice recognition device according to claim 1.