JP2000293195A - Voice inputting device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To simply execute a correction to a recognition result, that takes an especially long time for generation, by voice only and to efficiently conduct a voice inputting by conducting a correction corresponding to an objective to the contents being held when a correction word is included in the recognition result and outputting the result to a recognition result outputting means. SOLUTION: When a recognition object generating means 105 receives the recognition result of a previous uttering, that becomes a correction object, from a result holding means 103, a correction object dictionary is generated from a fundamental recognition object dictionary 106 and the recognition result. Then, a new recognition object dictionary is generated by combining the correction object dictionary, a correction word dictionary 107, that is beforehand specified, and the dictionary 106. The means 103 conducts discrimination to determine whether the recognition result includes a correction word or not. When a correction word is included in the recognition result, a correction corresponding to the objective of a correction word is performed to the contents being held and the corrected result is outputted to a recognition result outputting means 104 and a recognition object generating means 105.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a man-machine interface, and more particularly to a means for inputting data to a device, and more particularly to a speech recognition for converting speech into character information or the like.

[0002]

2. Description of the Related Art In a device utilizing speech recognition, it is inevitable that the device makes a recognition error. Therefore, in order to implement it as a device, it is essential to provide a means for correcting the recognition result. As a method of specifying the correction part,
Although there is a method of designating a correction part with a pointing device (mouse or touch panel), in the present invention, a situation is assumed in which only a voice is used or only a voice is usable.
In such a case, conventionally, for the recognition error, since the corrected portion cannot be specified, the same utterance content as that of the previous time is repeatedly repeated to perform the correction.

[0003]

When an attempt is made to correct a recognition error using only speech, it is troublesome to restate the entire utterance. In particular, it is extremely inconvenient and inefficient to restate the entire utterance of a long utterance such as a continuous digit having a large number of digits or an address because of a misrecognized part.

[0004] An object of the present invention is to provide a user who can efficiently correct a speech input device using speech recognition even if an unavoidable recognition error occurs with a relatively small amount of labor and using only speech. It is an object of the present invention to provide a voice input device having a machine interface or various information processing devices including the voice input device.

[0005]

According to the present invention, there is provided a recognition target generating means for generating a recognition target dictionary which also enables an input intended for correction, and a method for holding a previous recognition result and detecting a correction intention. The above problem is solved by providing a result holding unit for correcting and outputting the held recognition result.

In the voice input device of the present invention, voice input means for inputting voice, voice recognition means for recognizing the input voice within a specified dictionary to be recognized, and holding the recognition result of the voice recognition means And a recognition result generating means for generating a recognition target dictionary used by the voice recognition means based on a pre-designated basic recognition target dictionary and the like. The recognition target generation means generates a correction target dictionary from the basic recognition target dictionary and the recognition result when receiving a recognition result of a previous utterance that can be a correction target from a result holding means, and A new dictionary to be recognized is created by combining a corrected term dictionary specified in advance and the dictionary for basic recognition, and the result holding means includes the corrected term in the recognition result. If the recognition result includes a corrected term, a correction corresponding to the intention of the corrected term is made to the held content, and the corrected result is output to the recognition result output means and the recognition target. The recognition result is output to the recognition result output means and the recognition target generation means when the correction result does not include the corrected term.

According to another aspect of the present invention, there is provided a voice input device for inputting voice, a voice recognition device for recognizing the input voice within a specified dictionary to be recognized, and the voice recognition device. A result holding means for holding a recognition result of the means, a recognition result output means for outputting a recognition result, and a recognition target for generating a recognition target dictionary used by the voice recognition means based on a basic recognition target dictionary specified in advance. Generating means, wherein the recognition target generating means combines the basic recognition target dictionary with a pre-specified correction term dictionary,
When a new dictionary to be recognized is created, and the result holding unit includes a correction term in the recognition result of the speech recognition unit, a correction corresponding to the intention of the correction term is applied to the held content. Output the result.

According to another aspect of the present invention, there is provided a voice input device comprising: a voice input unit; a recognition unit for recognizing the input voice; an output unit for outputting a recognition result of the input voice;
The recognition means, when a part of the recognition result is a misrecognized word, corrects the misrecognized word, a correct word, and a phrase indicating which of those words is correct or incorrect, and The correction unit recognizes the correction and corrects the erroneously recognized word of a part of the recognition result into a correct word and outputs the result as a final result.

A voice input device having another configuration according to the present invention includes a means for inputting a numeric voice, a recognition means for recognizing the input voice, and an output means for outputting a recognition result of the input voice. The recognition means, when the number of the recognition result is a misrecognized word, a phrase indicating a mathematical operation difference between the misrecognized number and a correct number by a combination of an operator and a number,
A correction is input from the input means, the correction is recognized, and a number obtained by performing an operation indicated by the correction on the erroneously recognized numeral is output as a final result.

[0010]

DESCRIPTION OF THE PREFERRED EMBODIMENTS A first embodiment of the present invention will be described.

FIG. 1 shows a block diagram of a voice input device.
The uttered voice is converted into a digital signal by voice input means (corresponding to microphone to A / D conversion), and the voice recognition means provides a range of a recognition target dictionary which is given in advance or specified externally each time. Then, the digitalized speech is recognized, and the recognition result is output. The recognition result is held by the result holding means, and is presented to the user through a recognition result output means by a screen or voice as appropriate. The recognition target generation means generates the recognition target dictionary from the basic recognition target dictionary, and reconstructs the recognition target dictionary so that the user's utterance including the intention of the correction can be recognized simultaneously using the correction term dictionary. In addition, when the recognition target generating unit receives the recognition result from the result holding unit, the recognition target generation unit also uses the recognition result at the same time to generate a recognition target dictionary so that the user's utterance including the intention of correction can be recognized. Reconfigure.

A case will be described as an example where a continuous number of unlimited digits is to be recognized. For the sake of simplicity, the dictionary for basic recognition is made up of only arbitrary continuous numbers, but if an actual application system is assumed, various control commands may be added, and this is not the case.

Assuming that the utterance of the user is U and the output of the device is S, the input of the user and the output of the device are, for example, U “010” (user utters) S “080” (device is “1”) U is replaced with “8”) U “8 is 1” (user utters with intention of correction) S “010” (the device replaces “8” of previous recognition result with “1”)
Corrected).

In order to recognize the input "8 to 1" of the second user, the recognition target generation means reconfigures the recognition target dictionary. The dictionary to be recognized is, in addition to the “basic dictionary to be recognized”,
The combination of “correction target dictionary” + “correction term” + “basic recognition target dictionary corresponding to the correction target dictionary” becomes a dictionary in parallel. In this case, the correction term is "", but is not limited thereto, and may be "kara" or "yo". The correction target dictionary is a combination of one or more digits for the recognition result. In this example, “0”, “8”, “08”, “8”
0 "and" 080 ". However, since it is not possible to specify the first digit “0” or the third digit “0” only with “0”,
It is better not to set "0" as a correction target dictionary.

The "basic recognition target dictionary corresponding to the correction target dictionary" can be corrected more reliably by removing the correction target from arbitrary continuous numbers. For example, “8 to 1”, “8 to 88”, “8 to 11”, etc. may be included in the recognition target, but “8 to 8” may not be recognized.

The previous example is an example of a replacement error (a certain number is erroneously recognized as another number) when a continuous number is to be recognized. However, a source error (a number that is not actually uttered) is included in the recognition result. The same correction is possible in the case of an included error or an omission error (an error in which an actually uttered number is missing from the recognition result). For example, as an example of a source error, U “4100” S “42100” (the device sourced “2” error) U “421” 41 (speech including the intention of correction) S “4100” (device is U “0123” S “023” (device loses “1”) U “012” (speech with intention of correction) S “0123” (The device corrects the previous recognition result.)

By the way, the example here is 3 to 3 for simplicity.
It is clear that the number of digits is four consecutive digits and the number of digits is small. However, when the number of digits increases, it is clear that the burden on the user is smaller than restatement of all. Further, the correction can be repeatedly performed by the same method.

Further, the present invention can also be used when the recognition target is a place name such as an address. For example, U “Higashi Koigakubo, Kokubunji, Tokyo” S “Nishi Koigabo, Kokubunji, Tokyo”
Is replaced with "Nishi-Koigabo") U "Higashi-Koigabo in Nishi-Koigabo" (Speech with intention of correction) S "Higashi-Koigabo in Kokubunji, Tokyo" (The device corrects the previous recognition result) . In this case, the dictionary for basic recognition is, for example, an address in the whole of Japan, and the corrected term is "". The correction target dictionary is obtained from the basic recognition target dictionary and the recognition result. For example, “Tokyo”, “Kokubunji”
What should be three words of "Nishi Koigabo". In addition, the unit that may be misrecognized may be calculated in consideration of the restriction as an actual address (Kokubunji city is only in Tokyo). "Kokubunji City Nishi Koigabo""NishiKoigabo""west".

Next, a second embodiment of the present invention will be described.
FIG. 2 shows a block diagram of the voice input device. The difference from the block diagram of FIG. 1 is that the recognition target generation means does not use the recognition result.

As the correction term, "OK" or the like is used.

For example, U "010" S "080" (device replaces 1 with 8) U "OK, 1" (1st digit is not corrected, 2nd digit is 1) S "010" (device Corrects the previous recognition result). The correction part is specified by the number of "OK".

The same applies to the case where the recognition target is a place name. For example, U “Higashi Koigakubo, Kokubunji, Tokyo” S “Nishi Koigabo, Kokubunji, Tokyo”
Is replaced with "Nishi-Koigabo") U "OK, OK, Nishi-Koigabo" (speech including intention of correction) S "Higashi-Koigabo, Kokubunji-shi, Tokyo" (The device corrects the previous recognition result).

Next, a third embodiment of the present invention will be described. Although the third embodiment is an example in which the recognition target is limited to numerals, the block diagram of the voice input device is the same as FIG. As the correction term, “add”, “subtract”, “plus”, “minus” and the like are used.

For example, if "add" is used, U "67890" S "67190" (the device replaces 8 with 1) U "add 700 (Nanahyaku)" (speech including intention of correction) S " 67890 "(the device corrects the previous recognition result) If" pull "is used, U" 010 "S" 080 "(the device replaces 1 with 8) U" pull 70 (Nanaji-oo) "(correction) S "010" (the device corrects the previous recognition result). What is important here is that the utterance including the intention of the user to correct the utterance includes utterances such as “Hyaku” and “Ju”, and the correction position can be specified by the information of the order.

From the above embodiments, the present invention is characterized in that a user who inputs a correction does not re-input only a phrase to be corrected, but always adds an auxiliary word (correction term). That is, a correction method is indicated by a correction term. Either utter the correct content and the correct content at the same time, and use a word indicating the direction, such as "o", "kara", or "yo", which indicates which is the correct content or the correct content. And the correction term is used as a signal of digit shift. Of course, the correction term is not limited to “OK”. In the case of numeral recognition, recognition and calculation can be combined. An operator is used as a correction term, and a number is calculated on the misrecognized word according to the operator attached to the number to be input to the misrecognized number to be corrected, and the result is regarded as a correct recognition result.

Although the present invention does not particularly limit the speech recognition algorithm, it is possible to use, for example, an HMM (Hidden Marko Model) in carrying out the invention. The details of the speech recognition by the HMM are described in "Seiichi Nakagawa: Speech Recognition by Stochastic Model, IEICE, 1988" and the like, and are omitted here.

[0027]

According to the present invention, especially the recognition result of a long utterance can be easily corrected by using only the voice, and the voice can be input efficiently.

Further, since the correction is made a partial input or an input method different from the previous input is used by using the correction term, various corrections can be realized and a voice input device which is easy for the user to input can be provided. . In addition, it is possible to provide a voice input device that not only simplifies the input but also improves the final recognition result.

[Brief description of the drawings]

FIG. 1 is a block diagram showing one embodiment of a voice input device according to the present invention.

FIG. 2 is a block diagram showing another embodiment of the voice input device according to the present invention.

[Explanation of symbols]

101: voice input means, 102: voice recognition means, 103: result output means, 104: recognition result output means, 105: recognition target generation means, 106: basic recognition target dictionary, 107: correction term dictionary

──────────────────────────────────────────────────続 き Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI Theme coat ゛ (Reference) G10L 3/00 551P 561E

Claims (6)

[Claims] A voice input means for inputting voice, a voice recognition means for recognizing the input voice within a specified dictionary to be recognized, a result holding means for holding a recognition result of the voice recognition means, A recognition result output means for outputting a recognition result;
A recognition target generating means for generating a recognition target dictionary used by the voice recognition means based on a basic recognition target dictionary specified in advance; When the utterance recognition result is received, a correction target dictionary is generated from the basic recognition target dictionary and the recognition result, and the correction target dictionary, a correction term dictionary specified in advance, and the basic recognition target dictionary are stored. A new recognition target dictionary is created in combination with the result dictionary, and the result holding unit determines whether the recognition result includes a corrected term. If the recognition result includes the corrected term, the intention of the corrected term is determined. Is performed on the held contents, and the corrected result is output to the recognition result output means and the recognition target generation means, and the recognition result includes the corrected term. When no outputs as results for the recognition result output means and the recognition object generation means, voice input device, characterized in that. 2. A voice input means for inputting voice, a voice recognition means for recognizing the input voice within a specified dictionary to be recognized, a result holding means for holding a recognition result of the voice recognition means, A recognition result output means for outputting a recognition result;
A recognition target generating means for generating a recognition target dictionary used by the voice recognition means based on a pre-specified basic recognition target dictionary and the like; A new dictionary to be recognized is created by combining the corrected term dictionaries, and the result holding unit, when the recognition result of the speech recognition unit includes the corrected term, corresponds to the intention of the corrected term. A speech input device for outputting a result obtained by applying a correction to a held content. 3. The voice input device according to claim 2, wherein the basic recognition target dictionary is a dictionary for recognizing a stick reading not including a digit position and a dictionary for recognizing a digit reading including a digit position. When the result of the speech recognition includes a corrected term and includes a digit reading of a number, the result holding unit specifies a correction position based on the digit information of the digit reading. Apply a correction corresponding to the intention of the correction term to the held content, and output the corrected result;
A voice input device characterized by the above. 4. A telephone, a car navigation system, a personal information terminal, a game machine, or various personal computers comprising the voice input device according to claim 1. 5. An apparatus for inputting a voice, a recognizing means for recognizing the input voice, and an output means for outputting a recognition result of the input voice, wherein the recognizing means detects a part of the recognition result as an error In the case of a recognition word, the erroneous recognition word, the correct word, and a phrase indicating which of these words is correct or incorrect are input from the input means as a correction, and the correction is recognized and the recognition result is obtained. A voice input device for outputting a result obtained by correcting a part of misrecognized words to a correct word as a final result. 6. A means for inputting a numerical voice, a recognizing means for recognizing an input voice, and an output means for outputting a recognition result of the input voice, wherein the recognizing means is configured such that the number of the recognition result is In the case of a misrecognized word, a word indicating a mathematical operation difference between the misrecognized number and the correct number by a combination of an operator and a number is input as a correction from the input means, and the correction is recognized. A voice input device for outputting a number obtained by performing an operation instructed by the correction on the misrecognized number as a final result.