JP2017040856A - Voice recognition result editing device, voice recognition result editing method, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a voice recognition result editing device which enables a proof-reader to edit a voice recognition result as speedily as in real time.SOLUTION: A voice recognition result editing device includes: a voice collection section acquiring voice; a voice recognition section 12 which recognizes voice and acquires for each voice collection section and for each utterance a set of voice recognition results including a plurality of voice recognition results where reliability is imparted for every word; a recognition result collection section 13 which links and stores the set of voice recognition results, a microphone ID to be an identifier of the corresponding voice collection section, and an utterance start time of day of corresponding utterance to a prescribed storage domain and, when it receives a request signal, transmits the set of voice recognition results that correspond to a microphone ID designated by the request signal and that the utterance start time satisfies a prescribed condition; and a recognition result correction section which transmits the request signal, performs prescribed processing of words included in the respective voice recognition results in the set of voice recognition results whose reliability satisfies a prescribed condition when the set of voice recognition results are received, and displays the whole or a part of the voice recognition results included in the set of voice recognition results.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a speech recognition result editing apparatus, speech recognition result editing method, and program used by a proofreader for editing speech recognition results.

  There are many techniques for correcting misrecognition in the field of speech recognition. For example, for a video in which content to be uttered (such as speech) is recorded in advance in the manuscript database, the manuscript database is checked against the voice recognition result, and the text most similar to the voice recognition result is selected from the manuscript database. Thus, a technique is known in which subtitles are displayed so that subtitles can be added to video at a speed close to real time without requiring manual correction of recognition errors (Non-Patent Document 1). A speech correction interface is also known that presents recognition competition candidates for each word using a confusion network and allows the user to select (Non-Patent Document 2).

No author's description, “Subtitles for local programs”, NHK R & D, NHK Broadcasting Technology Laboratory, May 2014, No.145, p.70 Satoshi Ogata, Masataka Goto, “Speech Correction: A New Voice Input Interface that Corrects Recognition Errors by Selecting Operation”, Proceedings of 12th Interactive Software and Software Workshop (WISS2004), Japan Society for Software Science , December 2004, p.47-52

  The technology of Non-Patent Document 1 cannot achieve the above effect unless the utterance content can be recorded in the database in advance. In the technique of Non-Patent Document 2, the operation of the proofreader becomes complicated when correcting a plurality of words, and it is difficult to correct the speech recognition result at a speed close to real time.

  Therefore, an object of the present invention is to provide a speech recognition result editing apparatus that allows a proofreader to edit a speech recognition result at a speed close to real time.

  The speech recognition result editing apparatus according to the present invention includes a sound collection unit, a speech recognition unit, a recognition result aggregation unit, and a recognition result correction unit.

  The sound collection unit acquires an audio signal. The speech recognition unit recognizes a speech signal and acquires a speech recognition result set including a plurality of speech recognition results to which reliability is given for each word for each sound collection unit and each utterance. The recognition result aggregation unit associates the speech recognition result set, the microphone ID that is the identifier of the corresponding sound collection unit, and the utterance start time of the corresponding utterance, stores them in a predetermined storage area, and receives the request signal In this case, a speech recognition result set corresponding to the microphone ID specified by the request signal and having the utterance start time satisfying a predetermined condition is transmitted. When the recognition result correction unit transmits a request signal and receives the speech recognition result set, the recognition result correction unit determines a word satisfying a predetermined condition for the reliability of each word included in each speech recognition result in the speech recognition result set. Of the process of converting to a character string, the process of presetting a word whose reliability satisfies a predetermined condition to an editable state, and the process of assigning a shortening operation to enable editing of a word whose reliability satisfies a predetermined condition One or more processes are executed to display all or part of the speech recognition results included in the speech recognition result set.

  According to the speech recognition result editing apparatus of the present invention, the proofreader can edit the speech recognition result at a speed close to real time.

1 is a block diagram showing a configuration of a speech recognition result editing apparatus according to Embodiment 1. FIG. 6 is a flowchart illustrating a transmission operation of the speech recognition result editing apparatus according to the first embodiment. 5 is a flowchart illustrating an output operation of the speech recognition result editing apparatus according to the first embodiment. FIG. 3 is a block diagram illustrating a configuration of a recognition result aggregation unit according to the first embodiment. 6 is a flowchart illustrating the operation of a recognition result aggregation unit according to the first embodiment. FIG. 3 is a block diagram illustrating a configuration of a recognition result correction unit according to the first embodiment. 5 is a flowchart illustrating the operation of a recognition result correction unit according to the first embodiment. The figure which illustrates the screen display which a proofreader refers. The figure which illustrates the shortcut key allocated to the speech recognition result displayed on an edit window.

  Hereinafter, embodiments of the present invention will be described in detail. In addition, the same number is attached | subjected to the structure part which has the same function, and duplication description is abbreviate | omitted.

  The configuration of the speech recognition result editing apparatus 1 according to the first embodiment will be described below with reference to FIG. FIG. 1 is a block diagram showing the configuration of the speech recognition result editing apparatus 1 according to the present embodiment. As shown in FIG. 1, the speech recognition result editing apparatus 1 according to the present embodiment has an arbitrary number of sound collecting units 11-1,..., 11-n,. 11-N, the speech recognition unit 12, the recognition result aggregation unit 13, the recognition result database 14, and an arbitrary number of recognition result correction units 15-1 (M is an integer equal to or greater than 1). , 15-m,..., 15-M. Assume that a default speaker is normally assigned to each of the sound collection units 11-1,..., 11-n,. It is assumed that speaker allocation to the sound collection unit is executed in advance using a microphone ID that is an identifier of the sound collection unit and a speaker ID that is an identifier of the speaker. The total number of speakers may be smaller than N or larger. When the total number of speakers is smaller than N, it is assumed that there is a sound collection unit that is not assigned to any speaker. When the total number of speakers is larger than N, it is assumed that there is a speaker that is not associated with any sound collection unit. On the other hand, it is assumed that one or more proofreaders (persons in charge of proofreading, editing, and correcting speech recognition results) are assigned to each recognition result correction unit. It is assumed that the proofreader can browse and operate the recognition result correction unit assigned to the proofreader. It is preferable if the proofreader can directly or indirectly listen to the sound collected by the sound collection units 11-1, ..., 11-n, ..., 11-N. For example, in the case of a live broadcast of a lecture in which multiple panelists participate, the proofreader performs proofreading work in the venue where the lecture is being held or behind the backstage of the lecture hall, and the lecture contents are directly edited in real time. It is good also as what can be heard. In addition to this, the proofreader is in a place other than the lecture hall, uses a computer to access a distribution site that delivers the lecture contents live, and browses this distribution site while listening to the lecture contents indirectly. The calibration operation may be performed.

  Hereinafter, an outline of the operation of the speech recognition result editing apparatus 1 according to the present embodiment will be described with reference to FIGS. FIG. 2 is a flowchart showing the transmission operation of the speech recognition result editing apparatus 1 of the present embodiment, and FIG. 3 is a flowchart showing the output operation of the speech recognition result editing apparatus 1 of the present embodiment.

  As shown in FIG. 2, the sound collection units 11-1, ..., 11-n, ..., 11-N acquire audio signals (S11-1 to N). The speech recognition unit 12 recognizes a speech signal and acquires a speech recognition result set including a plurality of speech recognition results to which reliability is given for each word for each sound collection unit and each utterance (S12). The voice recognition unit 12 performs voice recognition by a known technique such as Reference Non-Patent Document 1.

(Reference Non-Patent Document 1: Nippon Telegraph and Telephone Corporation, “Development of Voice Recognition Engine VoiceRex”, [online], Nippon Telegraph and Telephone Corporation, [Search August 6, 2015], Internet <URL: http: / /www.ntt.co.jp/svlab/activity/category_2/product2_12.html>)
The speech recognition unit 12 performs speech segment detection processing, noise suppression processing, and the like on the speech signal as necessary, converts the speech signal into an acoustic feature amount, and executes speech recognition processing. As the speech recognition result set, a re-scoring or the like by a known technique as described in Reference Non-Patent Document 2 may be used.

(Reference Non-Patent Document 2: Akio Kobayashi, 5 others, “Speech recognition by discriminative rescoring based on word error minimization”, [online], January 2012, NHK STRL, [August 2015 6-day search], Internet <URL: http: //www.nhk.or.jp/strl/publica/rd/rd131/PDF/P28-39.pdf>)
The recognition result aggregating unit 13 associates the speech recognition result set, the microphone ID that is the identifier of the corresponding sound collection unit, and the utterance start time of the corresponding utterance, and stores them in a predetermined storage area (recognition result database 14). (S13A). The information stored in the recognition result database 14 may be sorted in the order of the utterance start time.

  The recognition result correction units 15-1, ..., 15-m, ..., 15-M transmit request signals (S15A). The request signal includes information specifying a proofreader ID and a microphone ID (a set of microphone IDs when a plurality of microphone IDs are specified). There are various timings at which the request signal is transmitted. For example, the recognition result correction units 15-1,..., 15-m,..., 15-M may transmit the request signal periodically according to the operation and setting of the proofreader. It is assumed that the periodic transmission of the request signal can be set by the proofreader to be executed / not executed. As will be described later, the recognition result correction units 15-1, ..., 15-m, ..., 15-M may transmit the request signal at the same timing as the transmission of the corrected speech recognition result.

  When receiving the request signal, the recognition result aggregating unit 13 corresponds to the microphone ID (microphone ID set) specified by the request signal, and the utterance start time satisfies a predetermined condition (for example, the time is the oldest). A speech recognition result set is transmitted (S13B). When the recognition result correction unit 15-1,..., 15-m,..., 15-M receives the speech recognition result set, the reliability of each word included in each speech recognition result in the speech recognition result set is determined. Processing for converting a word satisfying a predetermined condition into a predetermined character string, processing for presetting a word whose reliability satisfies a predetermined condition to be editable, and editing a word whose reliability satisfies a predetermined condition At least one of the processes for assigning a shortening operation (shortcut key) is executed to display all or part of the speech recognition results included in the speech recognition result set (S15B). The displayed speech recognition result is viewed, selected and corrected by the proofreader.

  As shown in FIG. 3, the recognition result correction units 15-1,..., 15-m,..., 15-M select the corrected speech recognition results for each utterance by the proofreader. Is sent to the recognition result aggregating unit 13 (S15C). The recognition result aggregating unit 13 outputs the received corrected speech recognition result (S13C). The recognition result aggregating unit 13 stores and manages the received corrected speech recognition result in the recognition result database 14 in association with the corresponding speech recognition result set. The recognition result aggregating unit 13 can select whether to output or discard the received corrected speech recognition result. In step S <b> 13 </ b> C, the recognition result aggregating unit 13 outputs the managed corrected speech recognition results managed in order of utterance start times as needed.

  Hereinafter, the configuration and operation of the recognition result aggregation unit 13 including exception processing and the like will be described in detail with reference to FIGS. 4 and 5. FIG. 4 is a block diagram illustrating the configuration of the recognition result aggregating unit 13 according to the present embodiment. FIG. 5 is a flowchart showing the operation of the recognition result aggregating unit 13 of this embodiment.

  As illustrated in FIG. 4, the recognition result aggregating unit 13 according to the present exemplary embodiment includes a registration unit 131, a recognition result transmission unit 132, a management unit 133, and an output unit 134.

  As illustrated in FIG. 5, the registration unit 131 registers the speech recognition result set, the utterance start time, and the microphone ID in the recognition result database 14 in association with each other (S131). Next, when the recognition result transmission unit 132 receives a request signal from a certain recognition result correction unit (in this case, the recognition result correction unit 15-m), the recognition result transmission unit 132 transmits the request signal to the recognition result correction unit 15-m that has transmitted the request signal. The speech recognition result set corresponding to the microphone ID (microphone ID set) specified by the recognition result correction unit 15-m in the request signal and having the oldest utterance start time, the microphone ID corresponding to the set, and the microphone ID The corresponding speaker ID is transmitted, and the proofreader ID corresponding to the recognition result correction unit 15-m is registered in the recognition result database 14 in association with the transmitted speech recognition result set (S132). The above steps S131 and S132 are performed as needed. For example, step S131 is executed every time voice input is started to any of the sound collection units. Step S132 is executed every time a request signal is received from any recognition result correction unit.

  The management unit 133 monitors whether or not any correction result is returned within a predetermined first period (S133A). This first period is a period provided for each recognition result correction unit based on the time when the recognition result aggregation unit 13 transmits the speech recognition result set to each recognition result correction unit. For example, when the first period is set to 5 minutes, the time when the recognition result aggregation unit 13 transmits the speech recognition result set to a certain recognition result correction unit 15-m is used as a reference for 5 minutes from the reference time (first The management unit 133 monitors whether or not the correction result is returned from the recognition result correction unit 15-m within the period. Therefore, the management unit 133 determines whether or not there is a response of a correction result from each recognition result correction unit within a first period from each reference time to a plurality of recognition result correction units having different reference times. Monitor. When no correction result is returned within the first period from each reference time (S133AN), the output unit 134 associates the correction time excess information with the corresponding speech recognition result set as a time-out process, and recognizes the result database. 14, the corresponding uncorrected speech recognition result set is discarded, or any speech recognition result in the set (for example, the speech recognition result having the first recognition result rank) is output (S134B). When any correction result is returned within the first period from each reference time (S133AY), the management unit 133 checks whether there is an uncorrected speech recognition result (S133B). If there is no uncorrected speech recognition result set, that is, if all corrected speech recognition results are returned for all speech recognition result sets transmitted in response to the request (S133BN), a step to be described later S134A is executed. On the other hand, when only a part of the corrected speech recognition result is returned (S133BY), the management unit 133 monitors whether all the correction results are returned within the second period (S133C). The second period is a period that is longer than the first period, and is provided for each recognition result correction unit based on the time when the speech recognition result set is transmitted as in the first period. In step S133B, if the uncorrected speech recognition result set includes the oldest utterance start time set, the second period may be changed to a short period in order to avoid impairing real-time performance. Good.

  When there is no reply of all the correction results within the second period from each reference time (S133CN), the output unit 134 outputs the corrected speech recognition result and the corrected speaker ID, and the uncorrected speech recognition result set Is discarded, or any speech recognition result in the set (for example, the speech recognition result of the first recognition result rank) is output (S134C). On the other hand, when all the correction results are returned within the second period from each reference time (S133CY) and in the case of step S133BN described above, the output unit 134 displays all the corrected speech recognition results and corrected. The speaker ID is output (S134A).

  Hereinafter, with reference to FIGS. 6 and 7, the configuration and operation of the recognition result correction units 15-1,..., 15 -m,..., 15 -M (hereinafter represented by the recognition result correction unit 15 -m). Will be described in detail including exception handling. FIG. 6 is a block diagram illustrating a configuration of the recognition result correction unit 15-m according to the present embodiment. FIG. 7 is a flowchart showing the operation of the recognition result correction unit 15-m of this embodiment.

  As illustrated in FIG. 6, the recognition result correction unit 15-m according to the present embodiment includes a request transmission unit 151, an output control unit 152, and a correction result transmission unit 153.

  As illustrated in FIG. 7, the request transmission unit 151 transmits a heartbeat and a request signal to the recognition result aggregation unit 13 (S151). The request transmission unit 151 periodically transmits only the heartbeat while the speech recognition result is being corrected. A heartbeat is a signal transmitted to inform the outside that a computer or network device is operating normally on a network.

  Note that the recognition result aggregating unit 13 described above receives the request signal from the request transmission unit 151 and returns the speech recognition result set and the like, and then the request transmission unit When the heartbeat does not reach from 151, any of the following processes A to C may be executed after a predetermined time has elapsed since the last heartbeat was received. A) The recognition result aggregation unit 13 discards the corresponding speech recognition result set. B) The recognition result aggregating unit 13 outputs one of the speech recognition results in the corresponding speech recognition result set (for example, the speech recognition result of the first recognition result rank). C) The recognition result aggregating unit 13 changes the status “correcting” of the corresponding speech recognition result to “uncorrected”, and corresponds to the recognition result correcting unit 15-m that transmitted the request signal when the next request signal is received. Returns the voice recognition result of.

  Next, the output control unit 152 receives the speech recognition result set from the recognition result aggregating unit 13, and converts a word whose reliability is less than the threshold value α into a predetermined character string (for example, a full-width space or a □ symbol). The set is output, and the corresponding speaker ID is output in an editable state (S152A). As described above, the speaker ID is linked to the microphone ID, but the speaker (sound collecting unit) used by the speaker is exchanged or the speaker who is not assigned the microphone (sound collecting unit) speaks. When the microphone (sound collecting unit) moves, the relationship between the microphone ID and the speaker ID associated in advance may be changed. When the relationship between the microphone ID and the speaker ID is changed, such as when the proofreader can directly or indirectly listen to the speaker's remarks, this may be noticed and the speaker ID may be corrected. Is possible.

  The output control unit 152 monitors whether or not the proofreader selects any speech recognition result (S152B). The selection of the speech recognition result by the proofreader is executed using a mouse, a keyboard, or other input means, but is not limited to this. When the proofreader has not selected any speech recognition result (default state, S152BN), the output control unit 152 outputs the speech recognition result having the first recognition result rank in the edit window in an editable state. A shortcut key (also referred to as an abbreviated operation in this specification) is assigned to a word whose reliability is greater than or equal to the threshold value α and less than the threshold value β (S152D). On the other hand, when the proofreader selects any speech recognition result (S152BY), the output control unit 152 outputs the speech recognition result selected by the proofreader in an editable state in the edit window, and the reliability is the threshold value α. Shortcut keys are assigned to words that are equal to or smaller than the threshold β (S152C). It is assumed that the value α is smaller than the value β. Therefore, in step S152A, for words whose reliability is less than the threshold value α (words belonging to the category with the lowest reliability), character string conversion is performed, and in step S152C, the reliability is equal to or higher than the threshold value α. Shortcut keys are assigned to words less than β (words belonging to a category having a medium reliability). Since the word of the category with the lowest reliability is often an odd word that is far from the correct answer, displaying it as it is may interfere with the proofreading work. Since corrections are mostly made, it is decided to convert them into predetermined character strings so that the proofreader can easily see them. Also, for words with medium reliability, this word may be correct as it is, so it is meaningful to display the recognition result as it is, but it is highly likely that it will be corrected by the proofreader. Therefore, a shortcut key is assigned so that the proofreader can quickly correct the corresponding part. It is assumed that the proofreader can move the recognition result in units of words by pressing an arrow key on the keyboard, for example, and can delete the character string for each word by pressing the delete key.

  The proofreader can select a result output for the speech recognition result corrected as described above. On the other hand, the proofreader can select discarding of a speech recognition result set before correction, a speech recognition result being corrected, and the like. The reason why the proofreader discards the speech recognition result set before correction or the speech recognition result in the middle of correction is because, for example, the correction work is prolonged and subsequent utterances occur one after another, and the real-time property of the calibration work is impaired. Such as when it is determined that By selecting discard, the proofreader can skip the received speech recognition result set and receive the next speech recognition result set, so that his / her proofreading work does not greatly retreat from the current utterance content.

  The correction result transmission unit 153 monitors whether the above-mentioned result output or discard is selected by the proofreader (S153A). When discard is selected by the proofreader (S153A discard), the modification result transmission unit 153 transmits a notification (discard notification) that the speech recognition result set has been discarded to the recognition result aggregation unit 13 (S153C). On the other hand, when the result output is selected by the proofreader (S153A result output), the correction result transmission unit 153 transmits the corrected speaker ID and the corrected speech recognition result to the recognition result aggregation unit 13 (S153B).

  The handling of the discard notification in the recognition result aggregation unit 13 will be described below. For example, when the recognition result aggregating unit 13 receives discard notifications from all the recognition result correcting units, the recognition result aggregating unit 13 may execute step S134B. Similarly, when the recognition result aggregating unit 13 receives a discard notification from some of the recognition result correcting units, the recognition result aggregating unit 13 may execute step S134C.

  On the other hand, the discard notification may be interpreted as being within the category of the corrected speech recognition result. In this case, the recognition result aggregating unit 13 executes step S134A even if a part or all of the speech recognition result is constituted by a discard notification. In this case, the recognition result aggregating unit 13 may discard the speech recognition result set corresponding to the discard notification and display nothing as the corresponding speech recognition result, or the corresponding speech recognition result as the corresponding speech recognition result. Any speech recognition result in the result set (for example, the speech recognition result of the first recognition result ranking) may be output.

  Hereinafter, an operation example of the output control unit 152 will be described with reference to FIGS. 8 and 9. FIG. 8 is a diagram illustrating a screen display referred to by the proofreader. FIG. 9 is a diagram illustrating shortcut keys assigned to the speech recognition results displayed in the editing window.

  In the example of FIG. 8, the speech recognition results “Today is good weather”, “Today is good weather”, “Today is good weather”, “Today is good weather” for the utterance “Today is good weather” It is assumed that a speech recognition result set including “...” is generated. In addition, in this example, the word “sphere” is given a reliability C that is less than the threshold α, and the word “dana” is given a reliability that is greater than or equal to the threshold α and less than β. And Further, it is assumed that the speaker linked by default to the microphone ID corresponding to this utterance is Mr. Yamayama. The speaker ID may be composed of numbers, English letters, etc., but in the example shown in the figure, the name of the speaker “Yoyama △ male” is used as it is as the speaker ID.

  In this example, when the proofreader has not selected one of the speech recognition results (default state), as shown in the figure, the speech recognition result of the first recognition result ranking is “Today is good weather. "Ne" is output to the editing window in an editable state. In addition to this, on the screen viewed by the proofreader, the speaker ID before correction, a list of voice recognition results used when the proofreader selects any voice recognition result, and the proofreader corrects the speaker ID. It is possible to display a list of speaker ID lists used in the above.

  As described above, the speaker ID before correction is output in a state where it can be edited by the proofreader. Regarding the list of speech recognition results, the word “sphere” to which the reliability C less than the threshold value α is assigned can be converted into a predetermined character string, for example, □, and output.

  For example, if the proofreader selects the speech recognition result “Sphere is good weather”, the proofreader's selected speech recognition result “Sphere is good weather” can be edited in the editing window as shown in FIG. Outputs in the correct state. The word “sphere” having a reliability less than the threshold value α is converted into a predetermined character string “□”, and is output in a selected state and editable state by default. On the other hand, the shortcut key [Alt key + 1] is assigned to the word “Dana” whose reliability is equal to or higher than the threshold α and lower than the threshold β. When the proofreader presses the shortcut key [Alt key + 1], the word “Dana” changes to a selected state and an editable state. When a plurality of words having a threshold value α and less than the threshold value β exist in one speech recognition result, different shortcut keys are assigned to these words.

  The speech recognition result editing apparatus 1 according to the present embodiment assigns a proofreader for each utterance every time a recognition result assignment request (request signal) is received from a proofreader (recognition result correction unit), and is a fixed time after proofreader assignment. Since the unrecognized speech recognition result set after the lapse is discarded, or the speech recognition result is selected and output from the set, the entire processing is delayed even if there is a slow proofreader. It is possible to correct the recognition result. The speech recognition result editing apparatus 1 according to the present embodiment specifies a microphone ID that the proofreader is responsible for in the request signal, so that when there are a plurality of proofreaders, there are a plurality of sound collection units (microphones). Even in this case, calibration processing can be performed in parallel, and correction results can be output in chronological order and at a speed close to real time even when a plurality of speakers are simultaneously speaking. In addition, the speech recognition result editing apparatus 1 according to the present embodiment allows the proofreader to efficiently perform editing work by replacing a low-reliability word with a specific character string or assigning a shortcut key or the like. Therefore, the correction time for the proofreader can be reduced.

  The speech recognition result editing apparatus 1 according to the present embodiment may be configured as a speech recognition result editing system including a plurality of hardware. In this case, for example, the N sound collecting units 11-1,..., 11-n,..., 11-N are N sound collecting devices, respectively, and the speech recognition unit 12, the recognition result aggregating unit 13, and the recognition result database 14 are used. , 15-m, M-recognition result correction units 15-1,..., 15-M are used as M recognition result correction devices. A result editing system may be used.

<Supplementary note>
The apparatus of the present invention includes, for example, a single hardware entity as an input unit to which a keyboard or the like can be connected, an output unit to which a liquid crystal display or the like can be connected, and a communication device (for example, a communication cable) capable of communicating outside the hardware entity. Can be connected to a communication unit, a CPU (Central Processing Unit, may include a cache memory or a register), a RAM or ROM that is a memory, an external storage device that is a hard disk, and an input unit, an output unit, or a communication unit thereof , A CPU, a RAM, a ROM, and a bus connected so that data can be exchanged between the external storage devices. If necessary, the hardware entity may be provided with a device (drive) that can read and write a recording medium such as a CD-ROM. A physical entity having such hardware resources includes a general-purpose computer.

  The external storage device of the hardware entity stores a program necessary for realizing the above functions and data necessary for processing the program (not limited to the external storage device, for example, reading a program) It may be stored in a ROM that is a dedicated storage device). Data obtained by the processing of these programs is appropriately stored in a RAM or an external storage device.

  In the hardware entity, each program stored in an external storage device (or ROM or the like) and data necessary for processing each program are read into a memory as necessary, and are interpreted and executed by a CPU as appropriate. . As a result, the CPU realizes a predetermined function (respective component requirements expressed as the above-described unit, unit, etc.).

  The present invention is not limited to the above-described embodiment, and can be appropriately changed without departing from the spirit of the present invention. In addition, the processing described in the above embodiment may be executed not only in time series according to the order of description but also in parallel or individually as required by the processing capability of the apparatus that executes the processing. .

  As described above, when the processing functions in the hardware entity (the apparatus of the present invention) described in the above embodiments are realized by a computer, the processing contents of the functions that the hardware entity should have are described by a program. Then, by executing this program on a computer, the processing functions in the hardware entity are realized on the computer.

  The program describing the processing contents can be recorded on a computer-readable recording medium. As the computer-readable recording medium, for example, any recording medium such as a magnetic recording device, an optical disk, a magneto-optical recording medium, and a semiconductor memory may be used. Specifically, for example, as a magnetic recording device, a hard disk device, a flexible disk, a magnetic tape or the like, and as an optical disk, a DVD (Digital Versatile Disc), a DVD-RAM (Random Access Memory), a CD-ROM (Compact Disc Read Only). Memory), CD-R (Recordable) / RW (ReWritable), etc., magneto-optical recording medium, MO (Magneto-Optical disc), etc., semiconductor memory, EEP-ROM (Electronically Erasable and Programmable-Read Only Memory), etc. Can be used.

  The program is distributed by selling, transferring, or lending a portable recording medium such as a DVD or CD-ROM in which the program is recorded. Furthermore, the program may be distributed by storing the program in a storage device of the server computer and transferring the program from the server computer to another computer via a network.

  A computer that executes such a program first stores, for example, a program recorded on a portable recording medium or a program transferred from a server computer in its own storage device. When executing the process, the computer reads a program stored in its own recording medium and executes a process according to the read program. As another execution form of the program, the computer may directly read the program from a portable recording medium and execute processing according to the program, and the program is transferred from the server computer to the computer. Each time, the processing according to the received program may be executed sequentially. Also, the program is not transferred from the server computer to the computer, and the above-described processing is executed by a so-called ASP (Application Service Provider) type service that realizes the processing function only by the execution instruction and result acquisition. It is good. Note that the program in this embodiment includes information that is used for processing by an electronic computer and that conforms to the program (data that is not a direct command to the computer but has a property that defines the processing of the computer).

In this embodiment, a hardware entity is configured by executing a predetermined program on a computer. However, at least a part of these processing contents may be realized by hardware.

Claims (8)

A sound collection unit for obtaining an audio signal;
A speech recognition unit that recognizes the speech signal and obtains a speech recognition result set including a plurality of speech recognition results to which reliability is given for each word for each sound collection unit and each utterance;
The speech recognition result set, the microphone ID that is the identifier of the corresponding sound pickup unit, and the utterance start time of the corresponding utterance are linked and stored in a predetermined storage area, and when the request signal is received, the request A recognition result aggregating unit that transmits a speech recognition result set corresponding to the microphone ID specified by the signal and whose utterance start time satisfies a predetermined condition;
When the request signal is transmitted and the speech recognition result set is received, a word satisfying a predetermined condition for the reliability of each word included in each speech recognition result in the speech recognition result set is converted into a predetermined character string. One of processing for assigning an abbreviated operation for enabling editing of a word whose reliability satisfies a predetermined condition, and processing for presetting a word satisfying a predetermined condition for reliability. A recognition result correction unit that performs the above processing and displays all or part of the speech recognition results included in the speech recognition result set;
A speech recognition result editing apparatus including: The speech recognition result editing apparatus according to claim 1,
The recognition result correction unit
The proofreader sends a speech recognition result selected and corrected one by one for each utterance and the next request signal,
The recognition result aggregating unit
A speech recognition result editing apparatus for outputting a received corrected speech recognition result. The speech recognition result editing apparatus according to claim 2,
The microphone ID is associated with a speaker ID that identifies the speaker in advance.
The recognition result aggregating unit
A corresponding speaker ID is transmitted together with the speech recognition result set;
The recognition result correction unit
The speaker ID is displayed in an editable state, and the corrected speaker ID is transmitted together with the corrected voice recognition result.
The recognition result aggregating unit
A speech recognition result editing apparatus that outputs a received corrected speech recognition result and a received corrected speaker ID. The speech recognition result editing apparatus according to claim 2 or 3,
The recognition result aggregating unit
One of a process of discarding a corresponding speech recognition result set and a process of outputting any speech recognition result in the corresponding speech recognition result set when a corrected speech recognition result is not received within a predetermined period. A speech recognition result editing apparatus that executes two or more processes. The speech recognition result editing apparatus according to any one of claims 1 to 4,
The value α is assumed to be smaller than the value β,
The recognition result correction unit
A shortening operation for converting a word whose reliability is less than a predetermined threshold α into a predetermined character string and enabling editing of a word whose reliability is equal to or higher than the predetermined threshold α and lower than the predetermined threshold β. A speech recognition result editing apparatus for displaying all or part of the speech recognition results included in the speech recognition result set. A speech recognition result editing method executed by a speech recognition result editing apparatus including a sound collection unit for acquiring a speech signal,
Recognizing the speech signal, obtaining a speech recognition result set including a plurality of speech recognition results with reliability given to each word for each sound collecting unit and each utterance;
Storing the speech recognition result set, the microphone ID that is the identifier of the corresponding sound collection unit, and the utterance start time of the corresponding utterance in a predetermined storage area;
Sending a request signal;
When receiving the request signal, transmitting a speech recognition result set corresponding to the microphone ID specified by the request signal and having an utterance start time satisfying a predetermined condition;
When the speech recognition result set is received, a process for converting a word satisfying a predetermined condition for each word included in each speech recognition result in the speech recognition result set to a predetermined character string, the reliability is Executing at least one of a process of presetting a word satisfying a predetermined condition into an editable state, a process of assigning a shortening operation for enabling the word whose reliability satisfies a predetermined condition, and Displaying all or part of the speech recognition results included in the speech recognition result set;
Speech recognition result editing method including The speech recognition result editing method according to claim 6,
The speech recognition result editing device
Transmitting a speech recognition result selected and corrected one by one per utterance by the proofreader, and a next request signal;
A speech recognition result editing method for executing a step of outputting a received corrected speech recognition result.   A program for causing a computer to function as the speech recognition result editing apparatus according to any one of claims 1 to 5.

Images