JPH0863185A - Speech recognition device - Google Patents

Abstract

PURPOSE: To eliminate unnaturality in interaction by displaying words having meanings from results of recognition on a display in order of the higher similarity of speech recognition or order reverse therefrom and selecting any from the displayed words without generating speeches. CONSTITUTION: The 'PCM sound recorder' uttered by a speaker in this speech recognizing device is inputted to the speech recognizing device and the words PCM having the meanings are recognized. As a result, the four words 'PC98', 'PCM', 'TQC' and 'DM' are displayed. The data on the PCM is displayed from a data base on a screen by bringing a cursor 4 at the 'PCM' and clicking the cursor. The results of the recognition are lined up in order of the high reliability thereof according thereto and, therefore, a correct answer is easily selectable simply by slightly moving the cursor 4 for selecting the results of the recognition.

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 device for automatically listening to a spoken word in a normal dialogue and automatically providing a speaker with necessary information. It is related.

[0002]

2. Description of the Related Art The present applicant has previously proposed, as one of the fields of use of voice recognition, an apparatus for automatically providing information from the conversation of a talking speaker. In the case of this device, the voice recognition user does not have to issue a special command for the voice recognition device, and if the speakers are in a normal conversation, the information is automatically given to the speakers. It is epoch-making in terms of being displayed.

[0003]

However, when the voice recognition device is fully operated, there is no problem, but in reality, a considerable error occurs in the voice recognition device.
There was a drawback that the dialogue between the speakers became unnatural. In particular, one of the speakers does not use this device, and one speaker has an advantage over the other speaker, and the speaker who is inferior uses speech recognition. In such cases, it is almost impossible to re-speak the same word in order to correct an error in the recognizer. The above-mentioned information providing device proposed by the applicant of the present invention has such drawbacks.

For the purpose of compensating for such incomplete speech recognition, some candidates of the recognition result are shown, each candidate is numbered, and the user gives a correct answer (correct answer) with the ten-key pad. There is a method to
No. 154100). However, this method has a drawback in that the human has to associate the candidate with the number, and doing this during the dialog interferes with the dialog.

There is also a method of re-speaking when the voice recognition result is incorrect (Japanese Patent Laid-Open No. 61-248).
No. 198), uttering the same word twice or three times in a row during a dialogue has a drawback of causing unpleasant feelings to the other party.

The present invention has been made in view of the above-mentioned circumstances, and in a device for automatically providing information from the conversation of a talking speaker, the uncertainty of the voice recognition portion therein. The object of the present invention is to provide a method for compensating for the above, and making the device easy to use with no unnaturalness in the dialogue.

[0007]

In order to solve the above problems, the present invention provides (1) an environment in which a speaker can have a conversation without being aware of the use of a special device, and one of the conversations. Means for extracting the above signals and applying them to the voice recognition device,
According to the recognition result, in a voice recognition device which displays the result (and / or) outputs a signal for operating a specific one with the result, a meaningful word is selected from the obtained recognition results. Display on the display in descending order of similarity of voice recognition (smallest distance first) or in reverse order so that any of the displayed ones can be selected without uttering a voice, or (2 )
An environment in which a speaker can talk without being aware of the use of a special device, a means for extracting one or more signals from the conversation and applying them to a voice recognition device, and displaying the result according to the recognition result. (And / or) In a voice recognition device that outputs a signal that causes a specific one to operate as a result, correct a meaningful word from the obtained recognition result into a character string, and add a character close to the beginning. Are displayed on the display in a predetermined order or in the reverse order so that any one of the displayed can be selected without producing a sound, or (3) the speaker uses a special device. Environment in which a conversation can be made without being conscious of, and a means for extracting one or more signals from the conversation and applying them to a voice recognition device, and displaying the result (and / or) according to the recognition result. In a voice recognition device that outputs a signal that operates a specific one as a result, if there are numbers in meaningful words from the obtained recognition results, the numbers near the beginning are determined in the determined order. The recognition result is displayed on the display so that any of the displayed results can be selected without producing a voice, or (4) the speaker does not have to be aware of the use of a special device. The result is displayed according to the environment in which the conversation is possible, the means for extracting one or more signals from the conversation and applying them to the voice recognition device, and the recognition result (and / or
Or) In the voice recognition device which outputs a signal for operating a specific one as a result, when both a letter and a number are present in a meaningful word in the obtained recognition result, the above-mentioned ordering is performed. According to, display the recognition result on the display by prioritizing the one closest to the beginning,
One of the displayed conversations can be selected without uttering a voice, or (5) the environment in which the speaker can have a conversation without being aware of the use of a special device, and one of the conversations. A means for extracting a signal of a person or more and applying it to a voice recognition device, and a voice recognition for outputting a signal for displaying a result (and / or) operating a specific one according to the result of the recognition, depending on the recognition result. In the device, the meaningful words from the obtained recognition results are numbers,
Or, if it is only alphabetic characters, it finds a character string paired with the number from the separately stored information, displays it as the information of the recognition result on the display, and utters one of the displayed characters. Or (5) in any of the above (1) to (5), the cursor for result selection is made to coincide with the one having the highest similarity among the correct candidates. Displayed on the display so that you can select any of the displayed ones without uttering a sound, or (7) The same or similar words that the speaker spoke previously were continued. When a specific word whose utterance is not unnatural is recognized, the previous recognition result is discarded or corrected, or (8) a switch is provided, and the speaker switches (9) In (8), when the switch for controlling the voice input is turned on / off for a shorter time than a predetermined time, The previous recognition result is discarded or corrected, and further, (1
0) In any one of (1) to (9) above, the information already displayed is displayed again by a specific command, and further, in (10) above, it is determined as (11) a specific command. Using the voice recognition result by the specified word, or (12) using ON / OFF of the voice input switch as the specific command, and further, in any of the above (1) to (4), 13) Some of the recognition result candidates with high similarity are collected and displayed on the display so that any one of the displayed results can be selected without uttering a voice, or (14)
Matching the cursor for result selection to the one with the highest similarity among the correct answer candidates and displaying it on the display, it was possible to select any of the displayed ones without making a sound. Is (15) above (9)
In any one of (12) to (12), it is characterized in that no operation or calculation is performed when a specific word is recognized.

[0008]

By means of the voice recognition device, meaningful words are displayed on the display in the order of high similarity of voice recognition (smallest distance first) or vice versa from the obtained recognition results. You can select any of them without making a voice.

[0009]

【Example】

Example 1 (corresponding to claim 1) In this example, an environment in which a speaker can have a conversation without being aware of the use of a special device and a voice recognition device by extracting one or more signals from the conversation Of the recognition results obtained in the voice recognition device, which is configured to display the result (and / or) output a signal for operating a specific one according to the recognition result. Meaningful words are displayed on the display in the order of high similarity of voice recognition (smallest distance first) or vice versa, and you can select any of the displayed words without uttering a voice. It is a thing.

FIG. 1 is a diagram for explaining the present embodiment.
This example is a system in which speaker B provides information to speaker A over the telephone. The telephones of both talkers A and B are connected by a normal public telephone line C, but the transmission signal from the telephone of the talker B is applied to the voice recognition device 1. Originally, it often has a part that extracts only the voice section from the signal before the voice recognition, but by using the word spotting method, it is possible to operate without the section detection unit. . What needs a section detection unit is considered to be included in the voice recognition unit. The voice recognition device 1 converts the emitted voice into an electric signal and recognizes whether the necessary words are arranged in the necessary order in the electric signal by a method using word spotting or a dialogue template.

This recognition method is described, for example, in the Acoustic Society of Japan Spring Lecture Collection, March 1-4-1, and in the Information Processing Society of Japan ('93 .10) 47th National Convention Lecture Collection (2).
2-369. As a result, the obtained results include nouns and particles. However, since the particle does not have to be displayed as a recognition result, only the nouns are taken out, and they are taken out in the order of similarity and displayed on the display 2. Displaying the one having a high degree of similarity among the results of the voice recognition is disclosed in, for example, Japanese Patent Laid-Open No. 3-173.
According to this method, as described in Japanese Patent No. 248,
There is a drawback that the display is very difficult to see because the recognition result is displayed, not limited to the noun.
This is solved in the present embodiment.

In FIG. 1, it is assumed that the speaker A says "I would like to know about PCM recorder" on the telephone and B answers "Is it a PCM recorder?" In the voice recognition device 1, "PCM recorder," which is uttered by the speaker B is input, and a meaningful word PCM is recognized. As a result, the words that gave the most probable result are, for example,
As shown in FIG. 2, the correct answer is "PC98".
"M" is in second place. In this case, the particle "de" is also recognized.
As shown in, the recognition results are “PC98” and “PC
Four words "M", "TQC", and "DM" are displayed. This is not limited to 4 words. Rather, it may be better to set a threshold for the degree of similarity, that is, the certainty of the result, rather than the number of words. It is displayed in a good shape like a pull-down menu, which can be selected with the mouse 3. As a result, match the cursor 4 with "PCM",
By clicking, data related to PCM from the database 3 is displayed on the screen.

According to this embodiment, since the recognition results are arranged in descending order of reliability, the correct answer can be easily selected by slightly moving the cursor 4 for selecting the recognition result. However, the correct answer is not necessarily in the upper rank of the candidates, and in the present embodiment, such as a symbol represented by some letters of the alphabet, the nerve is used to read which is the correct answer. In this case, the natural purpose of dialogue, which is the original purpose, is lost.

Embodiment 2 (corresponding to claim 2) In this embodiment, an environment in which a speaker can have a conversation without being aware of the use of a special device, and one or more signals are extracted from the conversation, Recognition obtained by means for applying to a voice recognition device and a voice recognition device adapted to output a signal for displaying the result (and / or) operating a specific one according to the result of recognition The meaningful words in the result are converted into a character string, the characters near the beginning are displayed on the display in the determined order or vice versa, and any of the displayed words can be selected without uttering a sound. It was done like this.

FIG. 3 shows an example in alphabetical order (ABC order). 1 is included in a portion having the same operation as in FIG.
Although the same reference numerals as those in the case of No. 1 are given and the description thereof is omitted, it is considered that the words of the recognized alphabet of the noun near the beginning are arranged in the order of ABC. Therefore, it is not so difficult to find a desired correct answer from these candidates. FIG. 4 shows a display example in which the characters are arranged in the order of ABC, but not only the alphabet but also the kanji can be arranged in the order of the Japanese syllabary. Although such characters and symbols are good, in the case of numbers, the following example 3 is used.

Embodiment 3 (corresponding to claim 3) In this embodiment, an environment in which a speaker can have a conversation without being aware of the use of a special device, and one or more signals are extracted from the conversation, Recognition obtained by means for applying to a voice recognition device and a voice recognition device adapted to output a signal for displaying the result (and / or) operating a specific one according to the result of recognition If there is a number in a meaningful word from the result, the recognition result is displayed on the display so that the number closest to the beginning is in the determined order, and select one from the displayed without uttering a voice The display example is shown in FIG. In the illustrated case, the numbers on the left side are arranged in ascending order, but this makes it easier to select, as in the previous embodiment.

Embodiment 4 (corresponding to claim 4) In this embodiment, an environment in which a speaker can have a conversation without being aware of the use of a special device, and one or more signals are extracted from the conversation, Recognition obtained by means for applying to a voice recognition device and a voice recognition device adapted to output a signal for displaying the result (and / or) operating a specific one according to the result of recognition If there are both letters and numbers in the meaningful words in the result, according to the above ordering, the recognition result is displayed on the display with the one closest to the beginning preferentially ordered, and one of the displayed ones is displayed. It is possible to select without making a sound.

FIG. 6 shows an example of the display. When alphabets and numbers are combined, the one closer to the left side is preferentially ordered. In this example, first, the alphabetic characters are AB
Priority is given to arranging in C order, and then the numbers are arranged. In this way, the sound to be uttered first has priority, so that it is easy to find. As a compromise between these, a voice recognition device as shown in the following fifth embodiment can be considered.

Embodiment 5 (corresponding to claim 5) In this embodiment, in the speech recognition apparatus according to any one of Embodiments 1 to 4, some of the recognition result candidates having a high degree of similarity are collected and displayed. Is displayed, and any of the displayed items can be selected without producing a voice. FIG. 7 is a diagram showing an embodiment thereof. As shown, the results of voice recognition by the voice recognition device 1 are arranged in the memory 1 after being arranged in the order of similarity, and then in the memory 2 after being arranged in the order of Japanese syllabary. Is displayed in addition to the top row of the Japanese syllabary with a high degree of similarity.

FIG. 8 is a diagram showing a display on the display 2. First, in a manner similar to that of the second embodiment, the candidates are arranged in the order of the Japanese syllabary (key system and below), and the candidate with the highest degree of similarity is at the top. This is an example in which two (copy, coffee) are arranged.
Since a candidate having a high degree of similarity has a high probability of being a correct answer, the correct answer can be selected from these two in many cases. However, even when there is no correct answer here, as shown in the second embodiment, It becomes easier to choose the correct answer. Of course, this is not only applicable to the second embodiment, but also applicable to other methods.
In this case, a word with a high degree of similarity is also included in the Japanese syllabary order, but this can be removed. Furthermore, not only two pieces having a high degree of similarity are displayed, but one piece, three pieces, or four pieces may be displayed if necessary. As a similar method, there is a method as shown in the following sixth embodiment.

Embodiment 6 (corresponding to claim 6) The present invention provides a speech recognition apparatus according to any one of Embodiments 1 to 4.
It is the one with the highest similarity among the correct answer candidates, which is displayed on the display with the cursor for result selection aligned with it so that you can select any of the displayed candidates without uttering a voice. . In this embodiment, as in the case of the third embodiment (see FIG. 5), the numbers are arranged in descending order of the numbers, but as shown in FIG. 9, the cursor 4 is positioned on the candidate (copy) having the largest similarity. are doing. Therefore, in many cases, it is not necessary to press the selection key as it is, and even in the case of an error, it is easy to select the correct answer.

The configuration is the same as in FIG. 7, and one noun with the highest degree of similarity is selected and stored in the memory 1. on the other hand,
From the nouns in memory 2 arranged in the order of the Japanese syllabary, select the same words as those in memory 1 and invert the lightness and darkness at the time of display. An example is shown in FIG. Of course, the present invention is not limited to inversion, and it is sufficient to make it possible to know that the cursor 4 matches the word. Since the degree of similarity is the maximum, the possibility of a correct answer is the highest, and at that time, the return key may be pressed without doing anything. If the word with the maximum similarity is not the correct answer, select the correct one with the mouse and click.

Embodiment 7 (corresponding to claim 7) In this embodiment, an environment in which a speaker can have a conversation without being aware of the use of a special device and one or more signals from the conversation are extracted. Recognition obtained by means for applying to a voice recognition device and a voice recognition device adapted to display (and / or) output a signal for operating a specific object depending on the recognition result If the result is only numbers or letters, it finds a character string that pairs with the number from the separately stored information, displays it as the information of the recognition result on the display, and selects one from the displayed It is possible to select without outputting a voice.

This embodiment is for a case where a plurality of words cannot be extracted from the uttered voice by spotting. F
Consider a case where a combination of letters and numbers such as X250 is meaningful. As shown in FIG. 10, this is because if there is the word FX250, which relates numbers and letters,
If the memory has a relation table 6 indicating that X and 250 are paired, and the recognition result is, for example, only FX or 250 (step 1), from this table 6, Find the combination, make up for the shortage (step 2), and display it on the display 2. As a result, not only information that is missing due to a recognition error can be added and displayed, but also results can be obtained when, for example, the speaker makes an expression such as "a machine called FX somehow". By such a method, the user can select one from the correct answer candidates, but there are cases where the correct answer is not included in the candidates. For that reason, the following Example 8 is considered.

Embodiment 8 (corresponding to claim 8) In this embodiment, one of the displayed recognition results is selected in the speech recognition apparatus of Embodiments 1 to 4 or 7, and a selected character or numeral is selected. A part of is corrected by an input means other than voice. In the same way as described above, a correct answer candidate is given, and if there is no correct answer, a word that seems to be close to the correct answer is selected, and the character string is corrected correctly from the keyboard to obtain the result. A similar invention is disclosed in Japanese Patent Laid-Open No. 59-214899. This is to re-speak by designating only a suspicious digit in the speech recognition result, and in the situation described in the present invention, re-speaks. It had the drawback of making people feel uncomfortable. In view of this, in the present embodiment, with the configuration shown in FIG. 11, one of the displayed items is selected once with the keyboard 7, and its spelling is corrected with the keyboard 8,
The information is transferred to the information retrieval section.

Embodiment 9 (corresponding to claim 9) In this embodiment, in the speech recognition apparatus according to any one of Embodiments 1 to 4 or 7, the same or similar words as previously spoken by the speaker are continuously used. When a specific word that is not unnatural to utter is recognized, the previous recognition result is discarded or corrected. As already mentioned in the prior art, when the recognition is not successful, uttering the same word again has a drawback that the dialogue becomes unnatural. Therefore, here, when I say a word that does not make the dialogue unnatural, I try to deny the previous answer. As a word that does not make this dialogue unnatural, "I would like to say again" or "I will repeat" is good,
When such a word is recognized by the recognition device, the previous result is replaced with the next recognized word.

It is necessary to re-voice the voice if the voice recognition is not successful and no recognition result is obtained, or if the result is a ridiculous character string that cannot be corrected. It may come out. In such a case, if the same thing is repeated twice, the other person will feel uncomfortable as well as unnatural. For that reason, the words "I will say again" or "I will repeat" will be used as a keyword, and if this is recognized, the previous recognition result will be denied and the mode will be re-entered. .

Embodiment 10 (corresponding to claim 10) In this embodiment, a switch is provided in any one of Embodiments 1 to 4 or 7, and a speaker controls the voice input to the apparatus by the following. This is to eliminate such drawbacks. For example, consider the case where an operator is using this device,
For example, if you say "Yes, this is Ricoh for office automation equipment,""This is Motoo of the General Affairs Department,"
It is expected that this recognition device will operate like connecting a phone to Mr. Motofuji. However, if Mr. Oe is among the employees, it is fully conceivable that Mr. Oe will be obtained as a recognition result because the voice of "OA" of the OA device and the part of "OEE" of Oe are close. In order to prevent such a mistake, a switch is provided on the side of the speaker A, and the switch activates the voice recognition device when the operator repeats the name of the person.

FIG. 12 is a block diagram showing the essential part of the embodiment. When the recognition device 1 is activated by the switch 9,
By using this switch 9, it is possible to correct the recognition result without giving unnaturalness to the other party of the dialogue described above. For that purpose, it is preferable to do as in Example 11 described below.

Embodiment 11 (corresponding to claim 11) In this embodiment, when the switch for voice input control is turned on / off for a shorter time than the predetermined time in the tenth embodiment, the recognition result before that is given. Is to be abandoned or modified. The same idea is also described in, for example, Japanese Patent Laid-Open No. 3-278297. This means that after the recognition result is recognized and the result is displayed, the recognition result is treated as correct when there is no input for a predetermined time or more in the next recognition waiting state. However, in the case of a correct result, it suffices that the result is output after waiting for a fixed time, but this method cannot be used for correction in the case of an incorrect result. The present embodiment is made against such a background, and since one meaningful voice has a minimum length of about 0.5 seconds, when the switch is turned on / off below that, The recognition result of is corrected or canceled.

FIG. 13 is a diagram for explaining the eleventh embodiment. As shown in the figure, a voice section detector 10 is provided after the switch 9, and here, a voice input microphone is used rather than a voice signal section. The section is detected and whether the section length is shorter than t (about 0.5 seconds) is determined depending on whether the electric signal from the section flows or is cut off. When it is short, it recognizes it first, erases the display of the currently displayed word, and when it is long, it recognizes the voice as usual. After the recognition, the previous information is erased and the current information is displayed as the information required next to the currently displayed information.

Next, in the case of providing information by telephone,
A series of incoming calls may be requesting the same information. In such a case, it is not necessary to repeat the recognition operation having the possibility of erroneous recognition and select the result, and the procedure of the twelfth embodiment may be performed.

Embodiment 12 (corresponding to claim 12) In this embodiment, in the voice recognition device of Embodiments 1 to 7 or 7 to 11, the information which has already been displayed is displayed again by a specific command. It was done. For example, by typing "1" or "return", the previous information is displayed as it is on the screen. As shown in FIG. 14, after the voice is recognized and displayed by the conventional method, the displayed data is stored in the memory 1. At this time, the contents in the memory 3 are deleted and the memory 2
Contents in memory 3 and contents in memory 1 in memory 2
Then, the latest data is written in the memory 1. Keyboard 1 before entering the recognition state or that state
When a number is struck from 1, the contents of the memory designated by the number are displayed on the display screen 2, and after speaker B uses the information, the same as the information obtained by using voice recognition. The contents of the memory are shifted, and the displayed information is written in the memory 1. By doing this, it is possible to display information without re-speaking the same words.

Embodiment 13 (corresponding to claim 13) In this embodiment, in the voice recognition apparatus of the embodiment 12, as a specific command, the voice recognition result by a predetermined word that does not make the dialogue unnatural is used. It is the one.
As a specific command, words such as "I understand" are good, and when this is recognized, the information previously displayed on the display is displayed again. This makes it possible to directly obtain information without selecting the correct answer from among several candidates. Of course, it is conceivable that two or three information items are displayed depending on the words, but if it is too complicated, it will not only be difficult to use, but also a large memory for temporarily storing information. turn into. Furthermore, it is easy for the speaker B to press the numeric buttons during the conversation with A, which is confusing and easy to make a mistake. So it would be even more convenient if this could be voiced. Therefore, as a voice command, "I understand"
Choose something like and decide. When this word is recognized, the previous information is displayed on the display. The memories shown in FIG. 14 are integrated into one, and when a word of the kind "I understand" is recognized, the information in the word is output.

Embodiment 14 (corresponding to claim 14) In this embodiment, the voice recognition apparatus of embodiment 12 is adapted to utilize ON / OFF of a voice input switch as a specific command. It is also a combination with. In the embodiment shown in FIG. 15, the switch 9 is shortly turned on /
Although the operation when it is turned off is written, there is another method such as when a short operation is recognized several times in a row.

Embodiment 15 (corresponding to claim 15) In this embodiment, the speech recognition apparatus according to any one of Embodiments 11 to 14 is configured so that no operation or calculation is performed when a specific word is recognized. is there. In a device such as the thirteenth embodiment, the switch is turned on for the purpose of correcting the previous recognition result, but when it is found that the correction is not necessary,
There was a drawback that the correction command could not be canceled. So, if you decide a word that is close to an auction that doesn't have meaning, for example, something like "Yes, that's right", and if this is recognized, just exit the recognition state and do not do anything. To

[0037]

As is apparent from the above description, according to the present invention, in the device for automatically providing information from the conversation of the talking speaker, the uncertainty of the voice recognition portion therein is uncertain. It is possible to provide an easy-to-use device that compensates for gender and eliminates unnaturalness in dialogue. Furthermore, in each claim, the following effects can be obtained. Effect corresponding to claim 1: Meaningful words are selected from the recognition result in the order of high similarity of speech recognition (smallest distance),
Or, since it was displayed on the display in reverse order so that you can select any of the displayed ones without uttering a voice, that is, the recognition results are arranged in order of high reliability. It is easy to select the correct answer by slightly moving the cursor for selection. Effect corresponding to claim 2: Meaningful words from the obtained recognition result are converted into a character string, and characters near the beginning are displayed on the display in a determined order or in the reverse order. When the letters are composed of alphabets, it is considered that the words near the beginning among the words of the recognized noun alphabet are arranged in the order of ABC,
It is not so difficult to find a desired correct answer from these candidates, and any one of the displayed candidates can be selected without uttering a voice. Effect corresponding to claim 3: When a number is present in a meaningful word in the recognition result, the recognition result is displayed on the display so that the number near the beginning is in a determined order. It is possible to select any of the above without making a sound. Effect corresponding to claim 4: When both a letter and a number are present in a meaningful word in the obtained recognition result, for example, when an alphabet and a number are combined, the one closer to the left side is given priority. For example, if alphabetical letters are arranged in order of ABC, priority is given to the alphabetical letters, and if numbers are successively arranged, the sound to be uttered first has priority, which makes it easier to find and utters one of the displayed sounds. Can be selected without. Effect corresponding to claim 5: Since some of the recognition result candidates having a high degree of similarity are collected and displayed on the display and the correct answer is selected from the candidates having a high degree of similarity, the probability of a correct answer is high. Therefore, it is possible to select any of the displayed ones without producing a sound. Effect corresponding to claim 6: Since the cursor having the highest degree of similarity among the correct answer candidates is displayed on the display by matching the cursor for result selection, any one of the displayed candidates can be selected. You can select without uttering a voice. Effect corresponding to claim 7: When the obtained recognition result is only numbers or letters, a character string paired with the number is found from the separately stored information, and that is also displayed as the recognition result information. Since it was displayed above,
It is possible to select any of the displayed ones without producing a sound. For example, when the recognition result is, for example, only FX or 250, the combination is found from this table, the shortage is compensated, and the result is displayed on the display. FX by adding information (eg FX or 250)
It can be displayed as 250. Effect corresponding to claim 8: When a correct answer is not included in correct answer candidates, a word that is likely to be close to the correct answer can be selected, and the character string can be corrected correctly from the keyboard to obtain the result. . Effect corresponding to claim 9: A specific word for which it is not unnatural for the speaker to continuously say the same or similar words as previously said, for example, "I would like to say again." I will repeat it "is recognized, so the previous recognition result is discarded or corrected, so that the dialogue does not become unnatural and does not cause discomfort to the other party. Effect corresponding to claim 10: Since a switch is provided on the input side of the voice recognition device so that the speaker can control voice input to the voice recognition device, for example, "OA" of OA equipment and Oe You can make sure you are not mistaken for Mr. Effect corresponding to claim 11: When the switch of claim 10 is used, when the switch is turned on / off for less than a predetermined time, the recognition result before that is discarded or corrected. Therefore, the recognition result can be corrected without giving unnaturalness to the other party of the dialogue. Effect corresponding to claim 12: Since the information that has already been displayed is displayed again by a specific command,
Information can be displayed without re-speaking the same word. Effect corresponding to claim 13: As a specific command, for example, a word such as "I understand" is used, and when this is recognized, the information previously displayed on the display is displayed again. It is possible to directly obtain information without selecting the correct answer from those candidates. Effect corresponding to claim 14: By using ON / OFF of the voice input switch as the specific command, the same effect as in the case of claim 10 can be obtained. Effect corresponding to claim 15: Since no operation or calculation is performed when a specific word is recognized, for example,
I decided a word close to an auction that doesn't make sense, for example, something like "yes, that's right", and if this is recognized, I just exit the recognition state and do nothing. Therefore, unnatural dialogue does not occur.

[Brief description of drawings]

FIG. 1 is a circuit configuration diagram of a main part for explaining a first embodiment of the present invention.

FIG. 2 is a diagram showing a display example on a display for explaining the first embodiment of the present invention.

FIG. 3 is a circuit configuration diagram of a main part for explaining a second embodiment of the present invention.

FIG. 4 is a diagram showing a display example on a display for explaining a second embodiment of the present invention.

FIG. 5 is a diagram showing a display example on a display for explaining a third embodiment of the present invention.

FIG. 6 is a diagram showing a display example on a display for explaining a fourth embodiment of the present invention.

FIG. 7 is a circuit configuration diagram of a main part for explaining a fifth embodiment of the present invention.

FIG. 8 is a diagram showing a display example on a display for explaining a fifth embodiment of the present invention.

FIG. 9 is a diagram showing a display example on a display for explaining a sixth embodiment of the present invention.

FIG. 10 is a flowchart for explaining a seventh embodiment of the present invention.

FIG. 11 is a diagram for explaining eighth and ninth embodiments of the present invention.

FIG. 12 is a main part configuration diagram for explaining a tenth embodiment of the present invention.

FIG. 13 is a main part configuration diagram for explaining an eleventh embodiment of the present invention.

FIG. 14 is a main part configuration diagram for explaining a twelfth and a thirteenth embodiments of the present invention.

FIG. 15 is a main part configuration diagram for explaining fourteenth and fifteenth embodiments of the present invention.

[Explanation of symbols]

1 ... Voice recognition device, 2 ... Display, 3 ... Mouse, 4
… Cursor, 5… Database, 6… Relationship table,
7, 8 ... Keyboard, 9 ... Switch, 10 ... Voice section detector, 11 ... Keyboard.

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification number Office reference number FI technical display location G10L 3/00 571 G (72) Inventor Masako Hirose 1-3-6 Nakamagome, Ota-ku, Tokyo Stock company Ricoh

Claims (15)

[Claims] 1. An environment in which a speaker can have a conversation without being aware of the use of a special device, a means for extracting one or more signals from the conversation and applying them to a voice recognition device, and a result of recognition. In a voice recognition device that displays the result (and / or) outputs a signal that causes a specific one to operate according to the result, a meaningful word is selected from the obtained recognition results to be similar to the voice recognition. A voice recognition device characterized in that it is displayed on the display in the order of high performance (in order of increasing distance) or in the reverse order so that one of the displayed can be selected without uttering a voice. 2. An environment in which a speaker can have a conversation without being aware of the use of a special device, a means for extracting one or more signals from the conversation and applying them to a voice recognition device, and a recognition result. In the voice recognition device, which displays the result (and / or) outputs a signal for operating a specific one according to the result, converts a meaningful word from the obtained recognition result into a character string. A voice recognition device characterized in that characters near the beginning are displayed on the display in a predetermined order or in the reverse order, and any one of the displayed characters can be selected without uttering a voice. 3. An environment in which a speaker can have a conversation without being aware of the use of a special device, a means for extracting one or more signals from the conversation and applying them to a voice recognition device, and a recognition result. , In the voice recognition device which displays the result (and / or) outputs a signal for operating a specific one with the result, there is a number in a meaningful word from the obtained recognition results In this case, the speech recognition device is characterized in that the recognition results are displayed on the display so that the numbers near the beginning are in a determined order, and any of the displayed results can be selected without uttering a voice. 4. An environment in which a speaker can have a conversation without being aware of the use of a special device, a means for extracting one or more signals from the conversation and applying them to a voice recognition device, and a result of recognition. In the voice recognition device, which displays the result (and / or) outputs a signal for operating a specific one according to the result, a meaningful word in the obtained recognition result is composed of letters and numbers. If both are present, according to the above ordering, the one closest to the beginning is preferentially ordered and the recognition result is displayed on the display, and one of the displayed ones can be selected without making a sound. And a voice recognition device. 5. The method according to any one of claims 1 to 4, wherein some of the recognition result candidates having a high degree of similarity are collected and displayed on a display, and one of the displayed results is not uttered. A voice recognition device characterized by being selectable. 6. The method according to claim 1, wherein a cursor for selecting a result is displayed on the display with a cursor having a highest similarity among the correct answer candidates.
A voice recognition device characterized in that it is possible to select any of the displayed ones without producing a voice. 7. An environment in which a speaker can have a conversation without being aware of the use of a special device, a means for extracting one or more signals from the conversation and applying them to a voice recognition device, and a recognition result. In the voice recognition device, which displays the result (and / or) outputs a signal for operating a specific one according to the result, a meaningful word from the obtained recognition result is a number, or If it is only English letters, it finds a character string that pairs with the number from the separately stored information, displays it as the information of the recognition result on the display, and does not utter any of the displayed characters. A voice recognition device characterized by being selectable. 8. The method according to claim 1, wherein one of the displayed recognition results is selected and a selected character,
Alternatively, a voice recognition device characterized by correcting a part of a numeral by an input means other than voice. 9. The specific word according to any one of claims 1 to 4 or 7, wherein a specific word is recognized so that it is not unnatural for the speaker to continuously say the same or similar words as previously said. The previous recognition result is discarded, or
A voice recognition device characterized by being modified. 10. The voice recognition device according to claim 1, wherein a switch is provided, and a speaker controls voice input to the device by the switch. 11. The voice recognition device according to claim 10, wherein when the switch for voice input control is turned on / off for a shorter time than a predetermined time, the recognition result before that is discarded or corrected. And a voice recognition device. 12. The voice recognition device according to claim 1, wherein the information already displayed is displayed again by a specific command. 13. The voice recognition device according to claim 12, wherein a voice recognition result by a predetermined word that does not make the dialogue unnatural is used as the specific command. 14. The voice recognition device according to claim 12, wherein on / off of a voice input switch is used as the specific command. 15. A voice recognition apparatus according to claim 11, wherein no operation or calculation is performed when a specific word is recognized.