JP2007156107A - Speech recognition apparatus and method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To avoid repeated erroneous recognition in a speech recognition result.
SOLUTION: A voice input from a voice input unit 201 is subjected to voice recognition by an acoustic processing unit 202 to a search unit 204, and a plurality of recognition results are output together with a recognition score. The recognition result selection unit 205 selects a predetermined number of recognition results in descending order of recognition score, and when the recognition result output unit 209 outputs the selected predetermined number of recognition results, the recognition result history of the highest rank is displayed. It is held at 208. At this time, when the current highest-ranked recognition result matches the previous history held in the recognition-result history 208, the same erroneous recognition is prevented from being repeated by lowering the rank of the recognition result and outputting it. .
[Selection] Figure 2

Description

  The present invention relates to a speech recognition apparatus and method for performing speech recognition and outputting the recognition result.

  With recent improvements in voice recognition technology, devices that enable voice operation have been put into practical use. In the speech recognition technology, there is a problem of what to do when an erroneous recognition result is output.

  For this problem, for example, Patent Literature 1 and Patent Literature 2 disclose speech recognition apparatuses configured so as not to repeat the same recognition error.

  The speech recognition apparatus disclosed in Patent Literature 1 includes means for a user to input the correctness / incorrectness of the recognition result, and means for storing the number of times the user has determined that the recognition result is incorrect for each recognition candidate. And about the recognition candidate in which the frequency | count of misrecognition exceeded predetermined number, the output within the maximum output candidate number (henceforth, N) is abbreviate | omitted. However, if the score of the omitted recognition candidate is within a predetermined rank, it is output after the (N + 1) th.

  Further, the speech recognition apparatus disclosed in Patent Document 2 performs the following control. That is, if there is a re-input between the notification of the speech recognition result and the recognition result confirmation process, and the re-input speech belongs to the same predetermined category as the previous recognition result, the previous recognition result and the re-input speech If the recognition results are regarded as the same, the recognition result is determined by excluding the result. For example, if the recognition result is output as “Tokyo Ota Ward Jonanjima” for voice input “Tokyo Ota Ward Showajima”, the recognition result “Tokyo Metropolitan Ota Ward Jonanjima” will be output when re-input. As a result, the recognition dictionary is changed or the recognition result is post-processed. Thereby, it is avoided that the same erroneous recognition is repeated.

The speech recognition apparatus disclosed in Patent Document 2 is more convenient for the user than the speech recognition apparatus disclosed in Patent Document 1 in that the user does not need to make a correct / incorrect determination on the speech recognition result. Is expensive.
Japanese Patent No. 3505931 Japanese Patent No. 3580643

  The speech recognition device disclosed in Patent Document 1 requires a first operation for the user to make a correct / incorrect determination on the recognition result, and there is a large number of operations until the speech recognition result is confirmed. is there. In addition, recognition candidates for which the number of erroneous recognitions exceeds a predetermined number are output after the (N + 1) th even when the candidates are correctly recognized. Therefore, when the display area of the recognition result is limited to N or less, there is a second problem that the user needs to change the display such as screen switching or scrolling in order to search for the candidate. .

  In addition, since the speech recognition apparatus disclosed in Patent Document 2 does not require an operation for determining whether the recognition result is correct or not, the first problem related to Patent Document 1 is solved. However, since the same recognition candidate as the past recognition result is excluded at the time of re-input, there is a problem that a recognition candidate that has been erroneously recognized once is not output by re-input within a predetermined time (period until the confirmation process). For example, when the user erroneously performs a re-input operation even though the output recognition result is correct, a desired recognition result cannot be obtained for a predetermined time.

  Further, as another problem, it is necessary to deal with utterances that correct some but not all of the recognition results. In general, in the conversation between people, as shown below, when correcting a mistake, there are many cases where only the wrong part is corrected instead of repeating the same contents.

Mr. A: "Where is your address?"
Mr. B: “It is Shimomaruko, Ota-ku, Tokyo.”
Mr. A: “You ’re Shinmaruko, Ota-ku, Tokyo?”
Mr. B: “No, this is Shimomaruko”
In this way, in a conversation between people, there are many scenes where only the wrong part ("Shimomaruko" in the above example) is transmitted to the other party. Similarly, in the speech recognition apparatus, it is troublesome for the user to utter again including the correct part of the recognition result (in the above example, “Ota-ku, Tokyo”) when uttering for correction. . It is more convenient for the user that the recognition result can be corrected in the same procedure as the conversation between the previous persons.

  The present invention has been made to solve the above-described problems, and an object of the present invention is to provide a speech recognition apparatus and method for efficiently correcting erroneous recognition.

  As a means for achieving the above object, a speech recognition apparatus of the present invention comprises the following arrangement.

  That is, receiving means for receiving voice information, voice recognition means for recognizing the voice information and acquiring a plurality of recognition results together with the recognition score, and a recognition result having the highest recognition score from the plurality of recognition results Selection means for selecting a plurality of recognition results including: output means for outputting the selected recognition results; history means for holding the recognition result having the highest recognition score among the selected recognition results as history And when the recognition result having the highest recognition score acquired by the voice recognition means matches at least a part of the history held in the history means, the ranking of the recognition results It is characterized by lowering.

  As a means for achieving the above object, the speech recognition method of the present invention comprises the following arrangement.

  That is, a reception step of receiving voice information, a voice recognition step of recognizing the voice information and acquiring a plurality of recognition results together with the recognition score, and a recognition result having the highest recognition score from at least the plurality of recognition results A selection step for selecting a plurality of recognition results including: an output step for outputting the selected recognition result; and a history step for holding the recognition result having the highest recognition score among the selected recognition results as a history. And when the recognition result with the highest recognition score acquired by the speech recognition step matches at least a part of the history held in the history means, the ranking of the recognition result It is characterized by lowering.

  According to the present invention having the above configuration, when the recognition result of the highest order matches the recognition result of the immediately preceding utterance, the recognition result is determined to be erroneous recognition, and the order is set to a lower rank than the highest order. By lowering, erroneous recognition can be corrected efficiently.

  Hereinafter, the present invention will be described in detail based on preferred embodiments with reference to the accompanying drawings. The configurations shown in the following embodiments are merely examples, and the present invention is not limited to the illustrated configurations.

<First Embodiment>
FIG. 1 is a block diagram showing a schematic configuration of a speech recognition apparatus according to an embodiment of the present invention. In FIG. 1, 101 is a central processing unit (CPU), 102 is a control memory (ROM), and 103 is a memory (RAM). Reference numeral 104 denotes operation keys such as a keyboard and buttons, 105 denotes a display device such as a liquid crystal, 106 denotes a sound input device such as a microphone, and 107 denotes a sound output device such as a speaker. A data bus 108 mediates transmission / reception of signals between the above components.

  A control program for realizing the speech recognition apparatus of the present embodiment and data used in the control program are recorded in the ROM 102. These control programs and data are appropriately fetched into the RAM 103 through the data bus 108 and executed by the CPU 101 under the control of the CPU 101. The execution result, that is, the result of speech recognition is displayed on the display device 105 or output from the speaker 107 using speech synthesis.

  FIG. 2 is a block diagram illustrating a functional configuration of the speech recognition apparatus according to the present embodiment. In FIG. 2, 201 is a voice input unit, 202 is an acoustic processing unit, 203 is a likelihood calculation unit, 204 is a search unit, and 205 is a recognition result output unit. Reference numeral 206 denotes an acoustic model, 207 denotes a language model, and 208 denotes a recognition result history.

  The input voice signal input to this apparatus is captured by the voice input unit 201, and a voice section is detected. The acoustic processing unit 202 extracts feature amounts used for speech recognition, such as LPC cepstrum and mel cepstrum coefficients, from the detected speech signal of the section. The likelihood calculation unit 203 calculates the likelihood of an acoustic model (HMM) necessary for the search process executed by the search unit 204 with reference to the extracted feature quantity and the acoustic model 206.

  The search unit 204 configures an HMM sequence necessary for speech recognition from a recognition dictionary that lists recognition target words, a language model 207 that records acceptable grammars, and an acoustic model 206. Then, referring to the likelihood of each HMM calculated by the likelihood calculation unit 203, a cumulative value (recognition score) of likelihood is obtained for each HMM sequence using the Viterbi algorithm or the like. That is, the higher the recognition score, the higher the certainty of recognition. Although an example using the HMM has been described here, other existing methods such as using DP may be used. In this case, the recognition score is assumed to increase when the certainty of recognition is high.

  The recognition result selection unit 205 selects predetermined N recognition results in descending order of recognition score. At the time of this selection, the order of recognition results to be selected is determined with reference to the recognition result history 208 that stores the previous recognition result. Details of this will be described later. The recognition result output unit 209 displays the recognition result at a predetermined position according to the order determined by the recognition result selection unit 205.

  The recognition result selection process in the speech recognition apparatus configured as described above will be described with reference to the flowchart of FIG.

  The recognition result selection unit 205 compares the recognition result having the first score with the previous recognition result stored in the recognition result history 208. If the recognition result matches all or a part of the stored recognition results (S101), the rank of the recognition results is lowered to N (S102). At this time, the ranks of the recognition results from the 2nd place to the N-1th place are raised one by one. If the previous recognition result is not stored in step S101 or if the recognition result does not match the previous recognition result at all, the rank is not changed.

  Then, according to the determined order, the recognition result output unit 209 outputs N recognition results (S103), stores the first recognition result in the recognition result history unit 208, and updates the recognition result history (S104).

  Hereinafter, the operation of the speech recognition apparatus of the present invention will be described using a specific utterance example. Here, a case where the voice recognition device of the present invention functions as an address input device is taken as an example.

  First, the case where the user utters “Ota Ward Showajima” in the first utterance and misrecognizes it, and utters the same content again will be described with reference to FIG. In this example, the number N of recognition results to be output is described as N = 3.

  FIG. 4 is a table showing the user's utterance contents and the recognition results. In FIG. 4, column 301 is the number of utterances, column 302 is the recognition result in the order of recognition scores obtained by search unit 204, column 303 is the first recognition result of the last utterance stored in recognition result history 208, and column 304. Is a recognition result output by the recognition result output unit 205. A line 305 indicates the first utterance, and a line 306 indicates the second utterance. First, in the case of the first utterance shown in the row 305, the top three recognition scores shown in the column 302 are obtained as follows for the user's utterance content “Ota Ward Showajima” shown in the column 301.

1st place: Ota-ku Jonanjima 2nd place: Ota-ku Showajima 3rd place: Ota-ku Keihinjima In this case, the recognition result selection unit 205 does not change the rank because there is no previous recognition result history shown in column 303 ( S101). As shown in the column 304, the recognition result output unit 209 outputs the recognition results in the order of the recognition scores (S103), and stores the output first-ranked recognition result “Ota Ward Jonanjima” in the recognition result history 208 (S104). .

  Then, since the first place of the output recognition result is different from the actual utterance content, the user utters “Showajima” for the second time as shown in line 306 in order to correct the recognition result. For the second utterance, the top three recognition scores shown in column 302 are obtained as follows.

1st place: Jonanjima 2nd place: Showajima 3rd place: Keihinjima At this time, the first recognition result "Jonanjima" is a part of the previous recognition result "Ota-ku Jonanjima" shown in row 303 "Jonanjima" (S101). Accordingly, the recognition result selection unit 205 lowers the rank of “Jonanjima” to the third place, and raises the second place “Showajima” to the first place and the third place “Keihinjima” to the second place (S102). Then, as shown in the column 304, the recognition result output unit 209 outputs the recognition results according to the order changed by the recognition result selection unit 205 (S103). The recognition result history is “Showajima” output as the first recognition result. 208 is stored (S104).

  As described above, when the recognition result matches a part of the recognition result of the immediately preceding utterance, the same recognition error (in this example, “Jonanjima”) can be avoided by lowering the rank. .

  Hereinafter, the display method of the recognition result in the voice recognition device (address input device) of the present embodiment will be described with reference to FIG. FIG. 5 is a diagram showing a display example of the recognition result. The recognition result output unit 209 displays “kube” and “town name” in the first recognition result in areas 401 and 402, respectively. At the same time, the first to third recognition results are displayed in areas 403, 404, and 405, respectively. Further, the areas 403 to 405 can be selected by the operation keys 104 such as a mouse and a keyboard, and the selected recognition result is reflected in the areas 401 and 402.

  FIG. 6 shows a display transition for the voice input example shown in FIG. 4 by such a display method.

  In FIG. 6, 500 is a display example before the first utterance is performed, and indicates that all display areas are ineffective.

  510 shows an example in which the first utterance “Ota Ward Showajima” is performed and the recognition result shown in the row 305 is displayed. That is, for the utterance of “Ota Ward Showajima”, “Ota Ward” of “Ota Ward Jonanjima”, which is the first recognition result selected by the recognition result selection unit 205, is in the region 511, and “Johnanjima” is It is displayed in area 512. In addition, the first to third recognition results obtained at this time are displayed in regions 513 to 515.

  Reference numeral 520 shows a display example when the second utterance “Showajima” is recognized. In this case, as described above, the recognition result of “Jonanjima” is obtained in the first place with respect to the utterance of “Showajima”. However, since a part of the previous recognition result “Ota Ward Jonanjima” coincides (S101), the recognition result selection unit 205 lowers the rank of “Jonanjima” to the third place, The ranking of a certain “Showajima” and “Keihinjima” is advanced (S102). Accordingly, the recognition result output unit 209 displays “Showa Island”, which has been ranked first, in the area 522, and displays the recognition results of the first to third positions according to the order determined by the recognition result selection unit 205. Output to areas 523 to 525.

  In the present embodiment, a result desired by the user can be selected from a plurality of displayed recognition results. For example, as shown in the area 525 of the display example 520, “Jonanjima” is displayed in descending order by the recognition result selection unit 205 and displayed as the third recognition result, but this can be selected with the operation key 104 such as a mouse. Can do. The display after this selection becomes 530. That is, “Seongnamjima” selected by the user is displayed in area 532. In this case, the recognition result history 208 stores the recognition result “Seongnamjima” selected by the user.

  As described above, in this embodiment, even when the recognition result matches a part or all of the previous recognition result, that is, even when it is determined that the recognition result is erroneous recognition, the recognition result is Nth (this embodiment In the example, N = 3) is output and displayed. Therefore, if there are N or more recognition result display areas, the user can select the recognition result without performing operations such as screen switching.

  In addition, although this embodiment demonstrated the case where a user corrects only a town name, this invention is not restricted to this example of course, and when correcting only a ward name or both a ward name and a town name. Is also applicable. That is, when the recognition result “Keihamajima” in “Setagaya-ku” is obtained in the first place against the utterance “Ota-ku Showajima”, “Ota-ku” is used to correct only the ward name. Also, the same procedure as in the present embodiment is applied. The same is true if you say “Ota Ward Showajima” to correct both the ward name and the village name.

  In addition, in step S102 in FIG. 3, when the obtained first-order recognition result matches all or part of the previous recognition result, the recognition result rank is ranked N (N is recognized by the speech recognition apparatus). The example of setting the number of results to be output) has been described. However, the present invention is not limited to this example, and the same effect can be obtained even if the recognition result that was first place is changed to an arbitrary order from the second place to the Nth place.

  As described above, according to the present embodiment, the same erroneous recognition is prevented from being repeatedly output as the first recognition result, and the recognition results are output in a rank other than the first. Therefore, if the recognition result is correct, that is, if the determination that it is erroneous recognition is incorrect, there remains room for the user to select the recognition result. Further, when the user corrects the recognition result, only the part to be corrected needs to be uttered, so that the convenience for the user is improved.

<Other embodiments>
Although the embodiment has been described in detail above, the present invention can take an embodiment as a system, apparatus, method, program, storage medium (recording medium), or the like. Specifically, the present invention may be applied to a system composed of a plurality of devices, or may be applied to an apparatus composed of a single device.

  In the present invention, a software program for realizing the functions of the above-described embodiments is supplied directly or remotely to a system or apparatus, and the computer of the system or apparatus reads and executes the supplied program code. Is also achieved. The program in this case is a program corresponding to the flowchart shown in the drawing in the embodiment.

  Accordingly, since the functions of the present invention are implemented by computer, the program code installed in the computer also implements the present invention. In other words, the present invention includes a computer program itself for realizing the functional processing of the present invention.

  In that case, as long as it has the function of a program, it may be in the form of object code, a program executed by an interpreter, script data supplied to the OS, or the like.

  Recording media for supplying the program include the following media. For example, floppy disk, hard disk, optical disk, magneto-optical disk, MO, CD-ROM, CD-R, CD-RW, magnetic tape, nonvolatile memory card, ROM, DVD (DVD-ROM, DVD- R).

  As a program supply method, the following method is also possible. That is, the browser of the client computer is connected to a homepage on the Internet, and the computer program itself (or a compressed file including an automatic installation function) of the present invention is downloaded to a recording medium such as a hard disk. It can also be realized by dividing the program code constituting the program of the present invention into a plurality of files and downloading each file from a different homepage. That is, a WWW server that allows a plurality of users to download a program file for realizing the functional processing of the present invention on a computer is also included in the present invention.

  In addition, the program of the present invention is encrypted, stored in a storage medium such as a CD-ROM, distributed to users, and key information for decryption is downloaded from a homepage via the Internet to users who have cleared predetermined conditions. It is also possible to make it. That is, the user can execute the encrypted program by using the key information and install it on the computer.

  Further, the functions of the above-described embodiments are realized by the computer executing the read program. Furthermore, based on the instructions of the program, an OS or the like running on the computer performs part or all of the actual processing, and the functions of the above-described embodiments can also be realized by the processing.

  Further, the program read from the recording medium is written in a memory provided in a function expansion board inserted into the computer or a function expansion unit connected to the computer, and then executed, so that the program of the above-described embodiment can be realized. Function is realized. That is, based on the instructions of the program, the CPU provided in the function expansion board or function expansion unit can perform part or all of the actual processing.

1 is a block diagram illustrating a schematic configuration of a speech recognition apparatus according to an embodiment of the present invention. It is a block diagram which shows the function structure in a speech recognition apparatus. It is a flowchart which shows a speech recognition process. It is a table | surface explaining utterance content and its recognition result. It is a figure which shows an example of a speech recognition result display. It is a figure which shows an example of a speech recognition result display.

Claims (12)

Receiving means for receiving audio information;
Voice recognition means for recognizing the voice information and acquiring a plurality of recognition results together with the recognition score;
Selecting means for selecting a plurality of recognition results including at least the recognition result having the highest recognition score from the plurality of recognition results;
Output means for outputting the selected recognition result;
A history unit that holds the recognition result having the highest recognition score among the selected recognition results as a history, and
The selection means lowers the rank of the recognition result when the recognition result having the highest recognition score acquired by the voice recognition means matches at least a part of the history held in the history means. Voice recognition device.   The selection means lowers the rank of the recognition result when the recognition result having the highest recognition score acquired by the voice recognition means matches a part of the latest history held in the history means. The speech recognition apparatus according to claim 1, wherein:   The speech recognition apparatus according to claim 1, wherein when the rank of the recognition result having the highest recognition score is lowered, the selection means lowers the rank to a rank within the number of recognition results selected by the selection means.   The speech recognition apparatus according to claim 1, wherein the output unit displays the selected recognition result. Furthermore, it has an operator selection means for selecting a recognition result according to an instruction from the operator among the recognition results displayed by the output means,
5. The speech recognition apparatus according to claim 4, wherein the output means outputs the recognition result selected by the operator selection means as the highest-order recognition result.   6. The voice recognition apparatus according to claim 5, wherein the operator's instruction in the operator selection means is performed by a pointing device.   7. The speech recognition apparatus according to claim 5, wherein the history unit holds a recognition result selected by the operator selection unit as a history. A receiving step for receiving audio information;
A speech recognition step of recognizing the speech information and obtaining a plurality of recognition results together with the recognition score;
A selection step of selecting a plurality of recognition results including at least the recognition result having the highest recognition score from the plurality of recognition results;
Outputting the selected recognition result; and
A history step of holding the recognition result having the highest recognition score among the selected recognition results as a history, and
The selection step lowers the rank of the recognition result when the recognition result having the highest recognition score acquired by the speech recognition step matches at least a part of the history held in the history means. Voice recognition method.   9. The speech recognition method according to claim 8, wherein the selection step lowers the rank of the recognition result having the highest recognition score to a rank within the number of recognition results selected by the selection step. The output step displays the selected recognition result,
Of the recognition results displayed by the output means, further comprising an operator selection means for selecting a recognition result according to an instruction from the operator,
9. The speech recognition method according to claim 8, wherein the output means outputs the recognition result selected in the operator selection step as the highest-order recognition result.   The speech recognition method according to claim 10, wherein the history step holds the recognition result selected in the operator selection step as a history.   A program for realizing the speech recognition method according to any one of claims 8 to 11 by being executed on an information processing apparatus.

Images