JP7481863B2 - Speech recognition error correction support device, program, and method - Google Patents

Description

The present invention relates to a speech recognition error correction assistance device, program, and method for assisting in the correction of speech recognition errors, and a speech recognition device.

When audio content (including video and audio content) recorded during program reporting, etc. is to be used as text, the task of transcribing the audio is essential. Normally, this task is performed by an operator who listens to the audio of the recorded content and enters text using a terminal keyboard, etc. In this case, the operator must frequently repeat the playback and pausing of the audio, or listen to the same part of the audio several times, and it is said that this task takes about four times the time of the content recording, even for an experienced operator.
As a technology to assist with conventional speech transcription work, there has been disclosed a technology for a speech recognition error correction assistance device that enables speech to be played back on a word-by-word basis and can quickly correct speech recognition errors by reducing waiting time at multiple correction terminals, even for content for which speech recognition has already been completed or is currently being recognized (see Patent Documents 1 and 2).

JP 2018-180519 A JP 2019-197210 A

The conventional techniques disclosed in Patent Documents 1 and 2 allow a user to play back speech synchronized with a word in the recognition results by simply clicking the word with a mouse or the like on the editing terminal. Therefore, the conventional techniques allow the user to quickly play back the speech of a specified word and efficiently correct speech recognition errors.
On the other hand, this conventional technology does not allow the operator of the editing terminal to recognize sections without speech or sections where speech recognition is difficult and no recognition results are output (unrecognized sections). Therefore, the conventional technology does not allow the operator to specify the playback position of the speech in the unrecognized section. For example, even if the operator wants to check the speech in the latter half of the unrecognized section, the conventional technology must play back the speech from the word immediately before the unrecognized section and wait until the desired confirmation position.
In response to such conventional techniques, there is a demand for a system that allows for efficient correction work by playing back audio at a specified position even in an unrecognized section.

Therefore, an object of the present invention is to provide a speech recognition error correction assistance device, program, and method that can visualize unrecognized sections of speech and play back speech at a specified position even in an unrecognized section so that correction work can be performed.

In order to solve the above problems, the speech recognition error correction assistance device according to the present invention is a speech recognition error correction assistance device that corrects speech recognition errors, and is equipped with an editing means, which is configured to include an editing screen control means, a voice output means, and an error correction means.

In such a configuration, the speech recognition error correction assistance device corrects the speech recognition error by the editing means.
Here, the editing means displays an editing screen including words constituting the speech recognition result and predetermined symbols indicating unrecognized sections of speech for which no recognition result was generated in the speech recognition by the editing screen control means. The editing screen control means switches between displaying and not displaying the symbols of the unrecognized sections by selection on a menu screen or by pressing a predetermined shortcut key.
The editing means also outputs, by the audio output means, audio corresponding to time information of a word or symbol specified on the editing screen. This allows the worker to check the recognized text and audio for each word. The worker can also skip the audio playback to a specified position in an unrecognized section and check the audio.
Then, the speech recognition error correction assistance device updates the word or symbol with the correction content corrected on the editing screen by the error correction means.

In order to solve the above problem, the speech recognition error correction assistance device according to the present invention is a speech recognition error correction assistance device that corrects speech recognition errors at multiple correction terminals, and is equipped with a storage means and multiple editing means, and the editing means is configured to include an editing screen control means, a voice output means, and an error correction means.

In this configuration, the speech recognition error correction assistance device stores in a storage means predetermined symbols indicating words in the speech recognition result and unrecognized sections of speech for which no recognition result was generated in the speech recognition, in association with the respective time information.

Then, the speech recognition error correction assistance device corrects the speech recognition error by the editing means.
Here, the editing means displays an editing screen including words constituting the speech recognition result and predetermined symbols indicating unrecognized sections on the corresponding editing terminal by the editing screen control means. The editing screen control means switches between displaying and not displaying the symbols of unrecognized sections by selection on a menu screen or by pressing a predetermined shortcut key.
The editing means also outputs, by the voice output means, a voice corresponding to the time information of the word or symbol designated on the editing screen to the corresponding editing terminal.
Then, the speech recognition error correction assistance device updates the words or symbols stored in the storage means with the correction contents corrected on the editing screen by the error correction means, with postscript taking priority.
The speech recognition error correction assistance device can be operated by a speech recognition error correction assistance program for causing a computer to function as each of the above-mentioned means.
In order to achieve the above object, the speech recognition error correction assistance method according to the present invention can be operated in a procedure including the operations of the above-mentioned means as steps.

The present invention provides the following excellent effects.
According to the present invention, as a result of speech recognition of content, time periods during which speech recognition was not performed can be visualized with predetermined symbols.
As a result, the present invention makes it possible to specify the playback position of the audio of the content in a time section where audio recognition was not performed, and also makes it possible to perform text editing in the same manner as correcting the results of audio recognition.

1 is a block diagram showing a configuration of a speech recognition error correction system according to an embodiment of the present invention; 1 is a block diagram showing a configuration of a speech recognition error correction assistance device according to an embodiment of the present invention; FIG. 2 is an explanatory diagram for explaining contents stored in a content information storage means; FIG. 2 is an explanatory diagram for explaining the contents of voice recognition information. FIG. 11 is an explanatory diagram for explaining the contents of segment information. 11A and 11B are diagrams showing examples of screens for selecting content, where (a) is an example of a selection screen, (b) is an example of a list box, and (c) is an example of a selection screen when a current time setting button is pressed. FIG. 11 is an explanatory diagram for explaining a start time and an end time when selecting content. FIG. 13 is a screen configuration diagram showing an example of a segment list screen showing a list of segments into which a voice recognition result of a content is divided. 13 is a screen configuration diagram showing an example in which a speech recognition result is displayed in an editing area on the segment list screen. FIG. 13A and 13B are explanatory diagrams showing editing screens on different editing terminals and explaining a state in which the same object is being edited. 11 is an explanatory diagram for explaining an example of an editing operation on a speech recognition result in an editing area. FIG. 11 is an explanatory diagram for explaining an example of a voice reproduction process for an unrecognized section of voice in an editing area; FIG. 11 is an explanatory diagram for explaining an example of a character string replacement process for an unrecognized section of speech in an editing area; FIG. 11 is an explanatory diagram illustrating an example of a comment adding process for an unrecognized section of speech in an editing area; FIG. FIG. 11 is an explanatory diagram for explaining an example of a correction result generated by a correction result output means; 5 is a flowchart showing a segment information generating operation for generating a speech recognition result in units of segments in the speech recognition error correction assistance device according to the embodiment of the present invention. 10 is a flowchart showing a segment information presenting operation of presenting a speech recognition result of the speech recognition error correction assistance device according to the embodiment of the present invention to a correction terminal on a segment basis. 4 is a flowchart showing a segment correction operation of correcting a recognition result while playing back a voice in the speech recognition error correction assistance device according to the embodiment of the present invention; FIG. 13 is a block diagram showing the configuration of a speech recognition error correction assistance device according to a modified embodiment of the present invention. FIG. 13 is a screen configuration diagram showing an example of a segment list screen on which a context menu is displayed. 1 is a block diagram showing a configuration of a voice recognition device according to an embodiment of the present invention;

Hereinafter, an embodiment of the present invention will be described with reference to the drawings.
[Outline of the speech recognition error correction system]
First, with reference to FIG. 1, an overview of a speech recognition error correction system S including a speech recognition error correction assistance device 1 according to an embodiment of the present invention will be described.

The speech recognition error correction system S corrects speech recognition errors in content that includes at least speech. In this embodiment, the content is described as content consisting of video and audio, for example, broadcast material content, but is not limited to broadcast material content as long as it includes audio.
The speech recognition error correction system S includes a speech recognition error correction assistance device 1 and a plurality of correction terminals 2 (2, . . . , 2).

The speech recognition error correction assistance device 1 corrects errors in speech recognition of content.
This speech recognition error correction assistance device 1 sequentially recognizes the speech of the content, associates the recognition result with the speech for each word, and enables the speech to be played back from any word position designated by the worker at the editing terminal 2. Furthermore, the speech recognition error correction assistance device 1 sequentially reflects corrections made by the multiple editing terminals 2, and when different corrections are made at the multiple editing terminals 2, validates the later correction without performing exclusive control.

The correction terminal 2 corrects errors included in the recognition result, which is text data obtained by the speech recognition error correction assistance device 1. The correction terminal 2 includes an input device such as a keyboard, a display device for displaying an editing screen, a speaker for displaying voice, etc. The display device may be configured to include a touch panel.
The speech recognition error correction assistance device 1 and the correction terminal 2 are connected via the Internet, an intranet or the like.
In this manner, the speech recognition error correction assistance device 1 corrects speech recognition errors using a plurality of correction terminals 2 without performing exclusive control for the same content.

As a result, the speech recognition error correction system S can check the speech and recognition results on a word-by-word basis even for content that is currently being recognized, and since it does not perform exclusive control, it can quickly correct speech recognition errors in real time using multiple correction terminals 2.
The configuration and operation of the speech recognition error correction assistance device 1 will be described below.

[Configuration of the speech recognition error correction support device]
First, the configuration of the speech recognition error correction assistance device 1 will be described with reference to FIG.
As shown in FIG. 2, the speech recognition error correction assistance device 1 comprises a content input means 10, a speech recognition means 11, a segment division means 12, an unrecognized section division means 13, an intra-segment information extraction means 14, a content information storage means 15, an editing means 16 (16, ..., 16), and an information deletion means 17.

The content input means 10 is used to input content.
The content input means 10 may input the content from an external storage medium, for example, or may input the content via a communication line.
Of the input content, the content input means 10 outputs the audio to the audio recognition means 11. The content input means 10 also writes and stores the input content (video and audio) in the content information storage means 15 for use in editing work by the editing means 16 described below.

The voice recognition means 11 recognizes the voice of the content input by the content input means 10, and generates a recognition result that is text data and time information for each word that constitutes the recognition result.
The speech recognition means 11 performs speech recognition using a language model, an acoustic model, and a pronunciation dictionary (not shown), and generates recognized words and time information indicating the start time of the speech of the words (word start time) and their duration. The speech recognition means 11 notifies the segment division means 12 of the generated recognition result words and time information, and writes and stores them in the content information storage means 15.
The voice recognition method used by the voice recognition means 11 may be, for example, a method disclosed in Japanese Patent Application Laid-Open No. 2010-175765, which recognizes a word string from a voice and outputs the result of the recognition.

The segment division means 12 divides, according to a predetermined criterion, the word string recognized by the speech recognition means 11. Hereinafter, each group (word string) generated as a result of the division by the segment division means 12 will be called a segment.
The division criterion used by the segment division means 12 can be any predetermined criterion. For example, a silent section of speech can be used as the division criterion. In this case, the segment division means 12 uses time information of the speech recognition result to determine that a silent section exists when the time interval between words is equal to or longer than a predetermined time, and divides the speech into before and after the silent section.

Also, for example, a cut point in the video can be used as a division criterion. In this case, the segment division means 12 detects frames as cut points from the video stored in the content information storage means 15, where the image features of adjacent frames differ more significantly than a predetermined criterion, and divides the recognition result into a time before and after the cut point.

Also, for example, meta information previously added to the content may be used as a division criterion. Meta information may be location information (geotag) from the Global Positioning System (GPS). In this case, the segment division means 12 divides the recognition result at the point where the location where the content was photographed or audio was collected differs depending on the location information.

The segment division means 12 outputs the divided segments to the unrecognized section division means 13 and the intra-segment information extraction means 14. In addition, the segment division means 12 writes and stores the words and their time information for each segment in the content information storage means 15.

The unrecognized section dividing means 13 divides the unrecognized sections of unrecognized speech into predetermined time intervals for each segment divided by the segment dividing means 12, and associates a predetermined symbol indicating unrecognized speech with time information corresponding to the symbol for each time interval.
Here, an unrecognized section is a section that is not recognized as speech, such as silence, speech with extremely low volume, speech with excessive background noise such as music superimposed, speech in a foreign language that is not the target of recognition, or speech produced by multiple people speaking simultaneously.
The unrecognized section dividing means 13 sequentially divides a section in which the elapsed time from a recognized word to the next word is equal to or longer than a predetermined time as an unrecognized section, based on the words recognized by the speech recognition means 11 and the time information. The unrecognized section dividing means 13 also sequentially divides a section from the beginning of a segment to a recognized word, or from the end time of a recognized word to the end of the segment, as an unrecognized section.

The unrecognized section division means 13 writes and stores, in the content information storage means 15, a symbol indicating the divided unrecognized section (unrecognized section symbol) and its start time and length as time information for the unrecognized section at predetermined time intervals (e.g., 2 seconds), in the same manner as recognized words.
The unrecognized section symbol may be any predetermined character other than the characters used as the voice recognition result, for example, a symbol such as ">".

The intra-segment information extraction means 14 extracts characteristic words contained in each segment divided by the segment division means 12 .
The characteristic words are characteristic words contained in the segment. For example, the intra-segment information extraction means 14 extracts words that characterize the segment using the TF-IDF method (TF: Term Frequency, word occurrence frequency; IDF: Inverse Document Frequency). TF-IDF is a type of weighting for words in a document (segment in this embodiment), and is mainly used in fields such as information retrieval and text summarization.
The intra-segment information extraction means 14 excludes unrecognized interval symbols contained in a segment from the calculation for extracting characteristic words.
Specifically, the intra-segment information extraction means 14 calculates the occurrence frequency tf(w, s) of a word w in a segment s using the following formula (1).

In this formula (1), n _w,s indicates the number of occurrences of a word w (excluding unrecognized interval symbols; the same applies below) in a segment s, and Σ _{t ∈} s n _t,s indicates the sum of the number of occurrences of all words in the segment s.
Furthermore, the intra-segment information extraction means 14 calculates the inverse document frequency idf(w) of a certain word w by the following formula (2).

In this formula (2), N is the total number of segments in the content, and df(w) is the number of segments in the content in which a certain word w appears (total number of segments [total number of documents]).
Then, for each word in the segment, the segment information extraction means 14 determines the word with the largest product of the tf value in equation (1) and the idf value in equation (2) as shown in the following equation (3), or a word with a product greater than a predetermined reference value, as the characteristic word of the segment.

The intra-segment information extraction means 14 writes and stores the extracted characteristic words in the content information storage means 15 in association with the segments.
The intra-segment information extraction means 14 may perform morphological analysis on the segment without using the TF-IDF method, and extract nouns and proper nouns as characteristic words.

In addition, when the content includes video, the intra-segment information extraction means 14 may extract thumbnail images from the video in the time period corresponding to the segment in addition to the characteristic words. For example, the intra-segment information extraction means 14 extracts the first frame of the video in the time period corresponding to the segment as a thumbnail image. The intra-segment information extraction means 14 writes and stores the extracted thumbnail image in the content information storage means 15 in association with the segment.

The content information storage means (storage means) 15 stores the content for which voice recognition errors are to be corrected, and various pieces of information obtained by dividing the content into segments. This content information storage means 15 can be configured with a general storage medium such as a hard disk or semiconductor memory.

Here, the content information stored in the content information storage unit 15 will be specifically described with reference to FIGS. 3 to 5 (and also with reference to FIG. 2 as appropriate).
3, the content information storage means 15 stores the content (video and audio) Con for which a speech recognition error is to be corrected. The content (video and audio) Con is stored by the content input means 10.

As shown in FIG. 3, the content information storage means 15 stores voice recognition information Rec and segment information Seg in association with each other for each content Con. The content Con is provided with identification information (such as an ID) and time information (not shown). The time information is a predetermined reference time, such as the time input to the voice recognition error correction assistance device 1 (input start time (year, month, day, hour, minute, etc.)).

The speech recognition information Rec includes words and time information that are recognized and associated by the speech recognition means 11. Note that the speech recognition information Rec also includes time information associated with symbols of unrecognized sections (unrecognized section symbols) divided by the unrecognized section division means 13, in the same manner as with words.
4, the voice recognition information Rec associates time information t with a correction flag f for each word w in association with identification information for identifying the content (content identification information). Here, the correction flag f is information indicating whether or not a correction has been made by the editing terminal 2 (for example, "1" for correction, "0" for no correction).
In FIG. 4, the symbol ">" in the word w column indicates a symbol (unrecognized section symbol) associated with a section obtained by dividing the unrecognized section by a predetermined time (here, 2 seconds).

The segment information Seg is information indicating the contents of the segments into which the voice recognition information Rec is divided by the segment division means 12 .
For example, as shown in Figure 5, segment information Seg corresponds to content identification information, and for each identification information (such as a number) that identifies an individual segment, a start time ts, an end time te, a characteristic word k, and a thumbnail image g.
Here, the start time ts and the end time te indicate the time interval of the segment. The characteristic word k is a word extracted by the intra-segment information extraction means 14. The thumbnail image g is a frame image extracted from the video of the content by the intra-segment information extraction means 14, which corresponds to the time information of the start of the segment.

In the example of FIG. 3, the voice recognition for content Con3 has not been completed and the segment information Seg has not been finalized.
Also, here, the content Con, the voice recognition information Rec, and the segment information Seg are stored in the same storage means, but they may be stored in separate storage means.
Returning to FIG. 2, the configuration of the speech recognition error correction assistance device 1 will be further described.

The editing means 16 allows an operator to use an externally connected editing terminal 2 to edit the voice recognition results stored in the content information storage means 15. A plurality of editing means 16 exist for each of the editing terminals 2. The editing means 16 operates as one process for the editing terminal 2, and the process may be duplicated every time an editing terminal 2 is connected.
As shown in FIG. 2, the editing means 16 includes a correction target selection means 160, an editing screen control means 161, an error correction means 162, an image/audio reproduction means 163, and a correction result output means 164.

The correction target selection means 160 selects the target content for correcting the speech recognition error. The correction target selection means 160 displays a screen for selecting the target content for correction on the editing terminal 2 and accepts the selection from the editing terminal 2.
An example of a screen for selecting content will now be described with reference to Fig. 6. For example, the correction target selection means 160 displays a selection screen 30 as shown in Fig. 6(a).

The selection screen 30 is a screen for specifying the content to be modified by specifying a time period. The selection screen 30 includes a time setting area 301, a current time setting button 302, and a modification start button 303.

The time setting area 301 is an area for setting the start time and end time. For example, the time setting area 301 is a list box for setting the date, hour, and minute. For example, as shown in FIG. 6(b), this time setting area 301 displays a list of candidates with a scroll bar 301b when button 301a is pressed. The same applies to the hours and minutes. The content to be modified is identified by the start time and end time.

The current time setting button 302 is a setting button for setting the content currently being input (or the content thereafter) from the current time as the object of correction.
For example, as shown in FIG. 6(c), by pressing the current time setting button 302, the start time is set to the current time, the end time is not specified, and the content currently being input from the current time (or thereafter) is made the subject of correction.
The start modification button 303 is a button for starting modification of the content to be modified after the time is set in the time setting area 301 or the current time setting button 302 .

Now, with reference to FIG. 7, a description will be given of the time period of the content specified by the start time and end time selected by the correction target selection means 160.
FIG. 7 shows a time series from n days (a predetermined number of days) before today to today.

Time section T1 is an example in which the start time is a time two days before today, and the end time is a time one day before today. Note that the start time of time section T1 is a time that is a predetermined time bt (for example, several tens of seconds to several minutes) earlier than the set time. This allows the operator of editing terminal 2 to check the audio from before the time of actual editing.

Time interval T2 is an example of the oldest day of content stored in the content information storage means 15, and the start time and end time are set. If there is no content that is a predetermined time bt earlier than the set time, the disclosure time for time interval T2 is set to the time corresponding to the oldest content.

The time interval T3 is an example in which a certain time today is set as the start time and an end time is not set. The time interval T3 also has a start time that is a predetermined time bt earlier than the set time.
Returning to FIG. 2, the configuration of the speech recognition error correction assistance device 1 will be further described.
The correction object selection means 160 outputs identification information such as the file name of the set content to the edit screen control means 161 .

The editing screen control means 161 displays an editing screen on the editing terminal 2 that includes the words that constitute the speech recognition results of the content and symbols for each predetermined time interval of the unrecognized section. Here, the editing screen control means 161 displays, for each segment, a selection button that specifies whether or not to display the characteristic words and the word strings (words, unrecognized section symbols) included in the segment, and controls whether or not to display the word strings of the segment by pressing the selection button.

Here, with reference to FIG. 8 and FIG. 9 (and also with reference to FIG. 2 as appropriate), examples of screens displayed by the edit screen control means 161 will be described together with the contents of control thereof.
As shown in FIG. 8, the editing screen control means 161 displays a segment list screen 31 on the screen of the editing terminal 2 corresponding to the editing screen control means 161 .

The segment list screen 31 is composed of a selection button 311, a characteristic word display field 312, a thumbnail image display area 313, a timetable display field 314, a scroll bar display field 315, and a correction result acquisition button 316.

The selection button 311 is a button for selecting whether or not to display a word string for each segment.
The characteristic word display column 312 is an area for displaying characteristic words extracted from a segment. The editing screen control means 161 reads out a characteristic word (characteristic word k in FIG. 5 ) corresponding to the segment from the content information storage means 15 and displays it in the characteristic word display column 312. Note that when segment information is not yet finalized because it is currently being input, such as in the case of live material, the editing screen control means 161 leaves the characteristic word display column 312 blank.

The thumbnail image display area 313 is an area for displaying thumbnail images extracted from within a segment. The editing screen control means 161 reads out a thumbnail image (thumbnail image g in FIG. 5) corresponding to the segment from the content information storage means 15 and displays it in the thumbnail image display area 313. Note that if the segment information corresponding to the content has not been determined, the editing screen control means 161 displays the first frame of the corresponding content as a thumbnail image.

The timetable display field 314 is a field for displaying a timetable showing the segment positions on the time axis of the content. The editing screen control means 161 generates and displays a timetable by referring to the segment time information (start time ts and end time te in FIG. 5) in the content information storage means 15. If the segment information corresponding to the content has not been determined, the editing screen control means 161 refers to the time information t (see FIG. 4) of the first and last words of the corresponding content.
The scroll bar display field 315 is a field for displaying a scroll bar that indicates which part of the segment is being displayed when the segment list does not fit on the screen. The editing screen control unit 161 updates the segment list on the screen by moving the scroll bar up and down.

The correction result acquisition button 316 is a button for requesting the correction result for the content to be corrected selected by the correction target selection means 160. When the correction result acquisition button 316 is pressed, the editing screen control means 161 outputs identification information (content identification information) of the content to be corrected selected by the correction target selection means 160 to the correction result output means 164.
In this way, by displaying the segment list screen 31, the worker can check the segments and can easily select the segment for which he or she wants to check the voice recognition results, compared to when the voice recognition results are displayed all at once.

When the worker clicks the mouse on the editing terminal 2 or touches the touch panel to press the selection button ("open" in Figure 8) 311 on this segment list screen 31, the editing screen control means 161 displays an editing area 317 (see Figure 9) on the segment list screen 31 where the word string of the segment can be edited.

FIG. 9 is an example of a screen showing a segment list screen 31B on which an editing area 317 is displayed.
In contrast to the segment list screen 31 described with reference to FIG. 8, this segment list screen 31B displays a moving image display area 313B and an editing area 317 for the selected segment.

The video display area 313B is an area in which the content corresponding to the segment is played back. When the segment is selected, the editing screen control means 161 refers to the time information of the segment (start time ts and end time te in FIG. 5 ) in the content information storage means 15, and displays the first frame of the video section of the corresponding content in the video display area 313B. If the segment information corresponding to the content has not been finalized, the editing screen control means 161 refers to the time information t of the first word of the corresponding content (see FIG. 4 ), and displays the first frame of the corresponding content in the video display area 313B.
When the image area of this video display area 313B is clicked with a mouse or the like, or when the playback start button st is pressed, the editing screen control means 161 instructs the video/audio playback means 163 to play back the content.

The editing area 317 is an area for displaying a word string corresponding to a segment and for performing editing such as correction. The editing screen control means 161 expands in the editing area 317 the word string corresponding to the segment stored in the content information storage means 15.
At this time, the editing screen control means 161 displays an unrecognized section symbol (">") SC as a word corresponding to the segment in addition to the recognized words. In the example of FIG. 9, there are 11 consecutive unrecognized section symbols (">"), which indicates that the unrecognized section is divided into 11 sections, and therefore if the predetermined time interval is 2 seconds, it can be seen that the time during which speech recognition was not successful lasted for at least 22 seconds. In this way, the speech recognition error correction assistance device 1 can make the operator aware of the presence of an unrecognized section in the speech and how long the unrecognized section continues, and can also make the unrecognized section a target for editing.

When the editing area 317 is displayed, the editing screen control means 161 adds a character string to be displayed each time a word is added to the voice recognition information Rec (see FIG. 3). Accordingly, the editing screen control means 161 expands the editing area 317. This allows the voice recognition error correction assistance device 1 to correct errors in the voice recognition of the content in real time.
Editing of the editing area 317 is performed by the error correction means 162 described later. When an arbitrary word in the editing area 317 is clicked with a mouse or the like, the editing screen control means 161 instructs the video/audio playback means 163 to play back the content corresponding to the time of the clicked word.
Furthermore, when the selection button ('close' in FIG. 9) 311 is pressed, the editing screen control unit 161 makes the editing area 317 invisible, changes the moving image display area 313B to the thumbnail image display area 313, and returns the display to the segment list screen 31 of FIG.
Returning to FIG. 2, the configuration of the speech recognition error correction assistance device 1 will be further described.

The error correcting means 162 corrects errors in the word string of the segment in the editing area 317 (FIG. 9) through editing operations by the operator of the editing terminal 2. This error correcting means 162 functions as a general text editor (screen editor) in the editing operation of correcting the word string. Therefore, in addition to correcting errors in the word string, it is also possible to add comments on the situation, impressions, etc. at the time. Moreover, the unrecognized section symbol can be edited as a simple word in exactly the same way as other word strings.
Moreover, the error correcting means 162 notifies the editing means 16 (editing means started by a different process) corresponding to the other editing terminal 2 of the correction content, thereby causing the correction content to be reflected in the editing area 317 of the other editing terminal 2. Conversely, when the error correcting means 162 is notified of the correction content from the other editing means 16, the error correcting means 162 causes the correction content to be reflected in the editing area 317 of its own editing terminal 2.

When a correction is made, the error correction means 162 overwrites the word in the speech recognition information Rec (FIG. 4) to correct it, and sets the correction flag f. Even when the error correction means 162 of another editing means 16 corrects the same correction target, the error correction means 162 updates by overwriting (with postscript taking priority), thereby avoiding complex exclusive control.
The error correction means 162 can also delete or insert a word instead of replacing the word. When deleting a word, the error correction means 162 deletes the row of the corresponding word in the speech recognition information Rec (FIG. 4). When inserting a word, the error correction means 162 combines the word with the word closest to the insertion position (immediately before or immediately after the inserted word), replaces the word with a word (word string) combined with the word in the same row as the word closest to the insertion position in the speech recognition information Rec (FIG. 4), and sets the correction flag f.
The word string correction process in the error correction means 162 is also performed on unrecognized section symbols in the same manner.

Then, the error correction means 162 notifies the other editing means 16 that the correction has been made, so that the corresponding error correction means 162 reflects the correction content by referring to the speech recognition information Rec (FIG. 4). Alternatively, the error correction means 162 may notify the word before and after the change, the word that has been deleted, or the word that has been inserted.
At this time, the error correcting means 162 displays the corrected words in a color attribute (for example, red) different from that of the uncorrected words.
This enables the error correcting means 162 to allow the worker to visually confirm the contents corrected at the other editing terminal 2.

For example, as shown in FIG. 10, it is assumed that the same object to be edited is edited at different editing terminals 2 ₁ and 2 ₂ in an editing area 317 .
In this case, as shown in Fig. 10(a), the worker at editing terminal ₂₁ corrects "Kyo" to "Kyo" in the editing area 317. Also, as shown in Fig. 10(b), the worker at editing terminal ₂₂ corrects "Hata" to "Hada" in the editing area 317.
Then, the error correction means 162 corresponding to the editing terminals 2 ₁ and 2 ₂ notify each other of the correction contents, thereby reflecting the correction contents made in the other editing terminal.
This allows the correction status to be shared among all workers, suppressing conflicts among workers and enabling errors in speech recognition to be corrected efficiently.

Moreover, the error correcting means 162 has a function of reproducing audio when correcting a word string.
Specifically, the error correction means 162 plays back audio from a word selected by clicking the mouse or touching the touch panel in the editing area 317 (FIG. 9). Also, when an arbitrary position is selected again during audio playback, the error correction means 162 stops the audio playback.

FIG. 11 is an explanatory diagram for explaining an example of an editing operation in the editing area.
For example, when "March" is selected by clicking the mouse in the editing area 317 in Fig. 11, the error correcting means 162 refers to the time information (time information t in Fig. 4) of the word included in the segment in the content information storage means 15, and instructs the video/audio reproducing means 163 to reproduce the audio from the time of the corresponding word position in the segment. At this time, in conjunction with the audio, a video corresponding to the audio reproduction time may be reproduced in the video display area 313B.

Then, as shown in FIG. 11, the error correction means 162 changes the display attributes of the display portion of each word corresponding to the reproduced sound in conjunction with the reproduction of the sound so as to clearly indicate the reproduction position of the sound and whether or not the sound is being reproduced within the segment. For example, the error correction means 162 displays the words corresponding to the sound in black and white inversion or in a predetermined color.

Also, for example, as shown in Figure 12, when an unrecognized section symbol SC is selected in the editing area 317 by clicking the mouse or the like, in the same manner as in the case of the voice-recognized word described in Figure 11, the error correction means 162 refers to the time information (time information t in Figure 4) of the unrecognized section symbol at the selected position contained in the segment in the content information storage means 15, and instructs the video/audio playback means 163 to play back the audio from the time of the corresponding word position in the segment.
This allows the operator to specify the position at which the audio is to be played back even in an unrecognized section.

In addition, the worker may check the played audio and video, and if the worker can hear the audio, as shown in FIG. 13, select one or more unrecognized section symbols SC (FIG. 12) and replace them with the replacement character string _CC1 that the worker heard.
FIG. 13 shows an example in which, for the 11 consecutive unrecognized section symbols (">") in FIG. 12, the first three unrecognized section symbols are left as they are, the next three unrecognized section symbols are selected and replaced with a replacement character string _CC1 , "A coat was needed.", and the next five unrecognized section symbols are left as they are.
If the predetermined time interval is 2 seconds, after the 6-second unrecognized section comes the voice "I needed a coat," which indicates that the unrecognized section continues after that. For example, if the worker hears a part that could not be recognized by the voice recognition means 11 due to an extremely low-volume voice or excessively overlapping background sound, it is effective to select the unrecognized section corresponding to the audible time and replace the character string in this way.
Incidentally, even if the worker can hear a portion that was not recognized by the voice recognition means 11, in cases where it is considered that matching the timing of the voice and the character string is not that important, such as when the same voice is being repeated, the unrecognized section symbol (">") may be left as it is, and a character string may be added before, after, or in the middle of the unrecognized section identified by the unrecognized section symbol (">". In this way, the length of the continuation of the unrecognized section remains as the number of unrecognized section symbols (">".

Also, as shown in FIG. 14, the worker may select the unrecognized section symbol (">") and replace it with a replacement character string _CC2 that is a comment about the situation, impressions, or the like.
FIG. 14 shows an example in which, for the 11 consecutive unrecognized section symbols (">") in FIG. 12, the first three unrecognized section symbols are left as they are, the next three unrecognized section symbols are selected and replaced with a replacement character string _CC2 , "(Editor's comment: Yamada) There is wind sound in the background," and the next five unrecognized section symbols are left as they are.
If the predetermined time interval is 2 seconds, this indicates that there is a wind sound in the background sound after a 6-second unrecognized section. In addition, it is preferable to add a note that the replacement character string _CC2 is not a speech sound but a comment (in FIG. 14, "Editing comment") and who added the comment (in FIG. 14, the worker, Yamada).
In addition, if the timing between the comment string and the audio is not considered to be particularly important, the unrecognized section symbol (">") may be left as is, and text may be added before, after, or in the middle of the unrecognized section identified by the unrecognized section symbol (">").
In this way, in the editing area 317, the operator can treat the unrecognized section symbol (">") without distinguishing it from the word string resulting from speech recognition.
Returning to FIG. 2, the configuration of the speech recognition error correction assistance device 1 will be further described.

The video/audio reproducing means 163 outputs the video and audio of the content to the editing terminal 2. This video/audio reproducing means 163 reads the video/audio from the content information storage means 15 based on the position specified by the editing screen control means 161 or the error correction means 162, and outputs it to the editing terminal 2.

The correction result output means 164 outputs the speech recognition result (transcription result) corrected by the error correction means 162 to the outside.
The correction result output means 164 reads out the speech recognition result after correction for the content to be corrected, output from the editing screen control means 161 , from the content information storage means 15 , and outputs it to the corresponding editing terminal 2 .

For example, as shown in FIG. 15, the correction result output means 164 generates text data including a title T, time information/characteristic words K, and correction results W. Note that there are multiple pieces of time information/characteristic words K and correction results W, the number of which is equal to the number of segments.

In the title T, for example, the revision date is written.
For example, the start time ts of the segment and the characteristic word k associated with the segment (see FIG. 5) are entered in the time information/characteristic word K. If the segment information is not finalized for the content currently being input, the correction result output means 164 enters the time information t of the first word (see FIG. 4) in the time information and leaves the characteristic word field blank.

For example, a corrected word w string (see FIG. 4) corresponding to the segment is entered in the correction result W. If the segment information is not finalized for the currently input content, the correction result output means 164 enters a word w string (see FIG. 4) up to the current point in time of the voice recognition information corresponding to the currently input content.
However, in the correction result W, the unrecognized section symbols of the word w are omitted.

The correction result output means 164 may output the corrected speech recognition result as text data to the correction terminal 2, or may output words (excluding unrecognized section symbols) for which a correction flag f (see FIG. 4) is set in the speech recognition information as data in a predetermined format with added color information, such as PDF (Portable Document Format).

The information deletion means 17 periodically deletes information stored in the content information storage means 15. For example, the information deletion means 17 is configured to store information for a predetermined number of days, and deletes information that has exceeded that number of days (content Con, voice recognition information Rec, segment information Seg (see FIG. 3)).

By configuring the speech recognition error correction assistance device 1 as described above, the speech recognition error correction assistance device 1 can quickly reflect the worker's corrections without performing exclusive control by validating the last correction made by multiple correction terminals 2, even if the same correction target is made by the same person. Furthermore, since the speech recognition error correction assistance device 1 can notify the other correction terminals 2 of the correction made by one correction terminal 2, it is possible to prevent the same correction from being made by multiple correction terminals 2, shorten the overall correction time, and correct speech recognition errors in input content in real time.
Furthermore, the speech recognition error correction assistance device 1 allows the user to correct speech recognition errors through simple text editing operations while checking the words of the recognition result and the original speech.
Furthermore, the speech recognition error correction assistance device 1 can specify a playback position of the speech during a time period during which the speech recognition result is not output, and the worker can check the speech from the specified position. This allows the speech recognition error correction assistance device 1 to make corrections such as adding character strings even in unrecognized sections.
The speech recognition error correction assistance device 1 can be operated by a speech recognition error correction assistance program for causing a computer to function as each of the above-mentioned means.

[Operation of the speech recognition error correction support device]
Next, the operation of the speech recognition error correction assistance device 1 according to the embodiment of the present invention (speech recognition error correction assistance method) will be described with reference to Figures 16 to 18. Note that, as the operation of the speech recognition error correction assistance device 1, a segment information generation operation for generating recognition results by speech recognition for content on a segment-by-segment basis, a segment information presentation operation for presenting the recognition results on a segment-by-segment basis to the correction terminal 2, and a segment correction operation for correcting the recognition results while playing back the speech will be described.

(Segment information generation operation)
First, the segment information generating operation of the speech recognition error correction assistance device 1 will be described with reference to FIG. 16 (and also with reference to FIG. 2 as necessary).

In step S1, the content input unit 10 inputs the content to be subjected to voice recognition. At this time, the content input unit 10 writes and stores the input content in the content information storage unit 15.
In step S2, the voice recognition means 11 recognizes the voice of the content input in step S1, associates the recognition result, which is text data, with time information for each word that constitutes the recognition result, notifies the segment division means 12, and writes and stores it in the content information storage means 15.

In step S3, the segment division means 12 divides the word string recognized in step S2 into segments based on predetermined criteria, such as video cut points, silent periods in audio, etc.

In step S4, the unrecognized section division means 13 divides the unrecognized sections of the speech that was not recognized into segments divided in step S3. At this time, the unrecognized section division means 13 divides the unrecognized sections into predetermined time intervals (e.g., 2 seconds), and writes and stores in the content information storage means 15 a symbol indicating the divided unrecognized section (unrecognized section symbol), as well as its start time and duration.

In step S5, the segment information extraction means 14 extracts characteristic words contained in each segment divided in step S3, and extracts thumbnail images from the video corresponding to the segment. At this time, the segment information extraction means 14 writes and stores the extracted characteristic words and thumbnail images in the content information storage means 15 in association with the segment.
If the input of the content is not completed (No in step S6), the speech recognition error correction assistance device 1 returns to step S1 to continue the input of the content.
On the other hand, if the input of the content is completed (Yes in step S6), the speech recognition error correction assistance device 1 ends the operation.
Through the above operations, the speech recognition error correction assistance device 1 associates the content with the speech recognition information and segment information in the content information storage means 15 as shown in FIG.

(Segment information presentation operation)
Next, with reference to Fig. 17 (and also with reference to Fig. 2 as necessary), a description will be given of the segment information presentation operation of the speech recognition error correction assistance device 1. Note that this segment information presentation operation is an operation of the editing means 16 after the connection with the correction terminal 2 is started.

In step S10, the correction object selection means 160 displays on the correction terminal 2 a selection screen 30 (see FIG. 6) for selecting a target content for which a voice recognition error stored in the content information storage means 15 is to be corrected.
In step S11, the correction target selection means 160 waits until the start time and end time of the content to be corrected are set on the screen and the correction start button is pressed (No in step S11). When the correction start button is pressed (Yes in step S11), the correction target selection means 160 transfers control to the editing screen control means 161 that performs control from step S12 onward.

In step S12, the editing screen control means 161 displays on the editing terminal 2 a segment list screen 31 (see FIG. 8) that includes, for each segment, a selection button for specifying whether or not to display characteristic words and word strings contained in the segment, based on various information stored in the content information storage means 15.

In step S13, the edit screen control unit 161 waits until the selection button (open) is pressed on the segment list screen (No in step S13).
On the other hand, if the selection button (open) is pressed (Yes in step S13), in step S14, the editing screen control means 161 displays an editing area 317 corresponding to the selected segment as shown in Fig. 9, and expands the word string that is the recognition result corresponding to the segment stored in the content information storage means 15 in the editing area 317. At this time, the editing screen control means 161 displays an unrecognized section symbol (">") SC together with the word that is the recognition result.

After this operation, the editing means 16 transitions to a state in which the worker can edit the editing result on the screen of the editing terminal 2. When the editing area 317 is displayed by pressing the selection button (open), the editing screen control means 161 can hide the editing area 317 at any timing by pressing the selection button (close), but this operation of hiding is not shown in the figure. Also, the operation of playing back the content in the video display area 313B of the segment list screen 31B (see FIG. 9) is not explained here.
By the above operation, the speech recognition error correction assistance device 1 becomes able to correct speech recognition errors in content on a segment-by-segment basis.

(Segment correction operation)
Next, with reference to Fig. 18 (and also with reference to Fig. 2 as necessary), a description will be given of the segment correction operation of the speech recognition error correction assistance device 1. Note that the segment correction operation is an optional procedure performed by the operator, so here, an example of an operation in which speech playback and correction operation are performed together will be described.

In step S20, the error correcting means 162 selects a word or word string for which audio is to be played back in the editing area 317 (FIG. 9) by the worker clicking the mouse or touching the touch panel. At this time, the error correcting means 162 outputs audio of a time corresponding to the word or word string to the editing terminal 2 via the video/audio playing means 163, with reference to the time information of the segment in the content information storage means 15. This causes the audio to be played back in the editing terminal 2, allowing the worker to compare and confirm the audio with the word string recognized by voice.
In addition, in this step S20, even if an unrecognized section symbol or an unrecognized section symbol string in the editing area 317 (Figure 9) is selected, the error correction means 162 outputs the corresponding audio to the correction terminal 2 via the video/audio reproduction means 163, in the same way as for a word or word string.

In step S21, the error correction means 162 accepts the specification of the position of the correction portion by the worker clicking the mouse or touching the touch panel. At this time, if the audio is being played back, the error correction means 162 stops the output to the editing terminal 2. This stops the playback of the audio.

In step S22, the error correcting means 162 displays a cursor at a specified position in the editing area, and corrects the recognition error by the operator's editing operation such as deleting characters, inserting characters, etc. Here, the error correcting means 162 updates the word or unrecognized section symbol in the content information storage means 15 with the correction result.
In step S23, the error correcting means 162 notifies the other editing means 16 of the correction content, so that the error correcting means 162 of the other editing means 16 reflects the content to be displayed on the corresponding editing terminal 2.

In step S24, the error correction means 162 accepts the specification of the position of the correction made by the worker by clicking the mouse or touching the touch panel. At this time, the error correction means 162 references the time information of the segment in the content information storage means 15 via the video/audio playback means 163 and outputs audio of the time corresponding to the word or word string to the editing terminal 2. This causes the audio to be played back on the editing terminal 2, allowing the worker to check whether the correction result is correct or not.

Although not shown in the drawing, if the worker checks in step S24 and finds that the correction portion has not been corrected correctly, the process returns to step S22 and the operation is repeated.
By the above operation, the speech recognition error correction assistance device 1 can quickly correct speech recognition errors at a plurality of correction terminals 2.

Although the embodiment of the present invention has been described above, the present invention is not limited to this embodiment.
Here, the content has been described as including video and audio, but the content may be audio only.
In this case, the intra-segment information extraction means 14 may extract only characteristic words without extracting thumbnail images, and the video/audio reproduction means 163 may be an audio output means that outputs only audio.

In addition, although the objects of correction of speech recognition errors are described here as contents for which speech recognition has already been completed and contents currently being recognized, it is also acceptable to only correct one of them. For example, when only contents currently being recognized are to be the objects of correction of speech recognition errors, the speech recognition error correction assistance device 1 may be configured simply by omitting the segment division means 12, the unrecognized section division means 13, the intra-segment information extraction means 14, and the correction object selection means 160.
In addition, here, the correction target selection means 160 selects the content to be corrected based on the date and time when the content was stored, but the selection criteria are not limited to the date and time. For example, if the content is associated with an electronic program guide genre such as news or sports as a broadcast material content, the genre may be specified to select the content. Alternatively, if event information when the content was recorded is added to the content, the event may be specified to select the content.

The speech recognition error correction assistance device 1 may also include the speech recognition means 11 externally.
For example, the configuration of a speech recognition error correction assistance device 1B shown in Fig. 19 may be used. The speech recognition error correction assistance device 1B includes the speech recognition means 11 of the speech recognition error correction assistance device 1 (Fig. 2) as an external speech recognition device. In this case, the recognition result input means 18, which is an input interface, stores the words and time information in association with each other in the content information storage means 15 and notifies the segment division means 12.
The speech recognition error correction assistance device 1B can also be operated by a speech recognition error correction assistance program for causing a computer to function as each of the above-mentioned means.

The error correcting means 162 may also select the editing operation from a context menu.
For example, as shown in FIG. 20, the error correcting means 162 displays a context menu 318 in response to a right-click key of the mouse or the like.
The context menu 318 displays functions (e.g., copy) and shortcut keys (e.g., Ctrl+C) together, and when a function is selected by, for example, left-clicking the mouse, the error correction means 162 executes the corresponding editing operation. Alternatively, when a shortcut key corresponding to a function is pressed, the error correction means 162 executes the corresponding editing operation.
This reduces the time required for workers to learn the operation method and minimizes the number of operation procedures.

Also, here, the editing screen control means 161 displays an unrecognized section symbol (">") SC in the editing area 317 (see FIG. 9) in a section where speech is not recognized.
However, the unrecognized section symbol may be switched between displayed and hidden by an operator's operation.
For example, in the initial state, the unrecognized section symbol is hidden, and the editing screen control means 161 displays the unrecognized section symbol when the user selects “Display unrecognized sections >>>” from the context menu 318 shown in FIG. 20 or when the shortcut key (here, Ctrl+D) is pressed.
In addition, when the unrecognized section symbol is displayed, the editing screen control means 161 hides the unrecognized section symbol by selecting “Hide unrecognized section >>>” (not shown) from the context menu or by pressing the corresponding shortcut key.
This allows the editing screen control means 161 to switch between displaying and not displaying the unrecognized section symbol for each editing terminal 2 .

Here, an example of switching between display and non-display of the unrecognized section symbol (">") has been described, but the display and non-display of the word string of the unrecognized section (including the unrecognized section symbol) may be switched depending on whether or not it is an unrecognized section. For example, in the voice recognition information Rec (FIG. 4), not only the word W with the unrecognized section symbol (">") but also the word string including the unrecognized section symbol (">") may be the target of switching between display and non-display. In addition, in order to determine whether or not it is a word of the unrecognized section, in addition to the presence or absence of the unrecognized section symbol (">"), a recognition flag (not shown) may be provided in the voice recognition information Rec (FIG. 4) in association with the word w, and whether or not it is a word of the unrecognized section may be determined based on the value (for example, recognized word "1", unrecognized word "0"). The value of this recognition flag can be set by the voice recognition means 11 and the unrecognized section division means 13. For example, the speech recognition means 11 sets a value indicating a recognized word (e.g., "1") to the recognition flag for a recognized word, and the unrecognized section division means 13 sets a value indicating an unrecognized word (e.g., "0") to the recognition flag for a word that corresponds to an unrecognized section symbol.

In addition, the speech recognition error correction assistance device 1, 1B corrects speech recognition errors using a plurality of correction terminals 2.
However, there may be only one correction terminal 2. In this case, the speech recognition error correction assistance device 1, 1B may be configured to include one editing means 16.

Further, in this embodiment, the speech recognition error correction assistance device 1, 1B is configured to include the unrecognized section division means 13.
However, the unrecognized section dividing means 13 may be provided in a voice recognition device having the voice recognition means 11 .
For example, as shown in Fig. 21, a speech recognition device 3 for recognizing speech may be configured to include a speech recognition means 11 and an unrecognized section division means 13, and may be substituted for the speech recognition means 11 of the speech recognition error correction assistance devices 1 and 1B. Note that the speech recognition means 11 and the unrecognized section division means 13 have the same configuration as those of the speech recognition error correction assistance device 1 described in Fig. 2, and therefore a description thereof will be omitted.

1, 1B Speech recognition error correction support device 10 Content input means 11 Speech recognition means (speech recognition device)
12 Segment division means 13 Unrecognized section division means 14 In-segment information extraction means 15 Content information storage means (storage means)
16 Editing means 160 Editing object selection means 161 Editing screen control means 162 Error correction means 163 Video/audio reproducing means (audio output means)
164 Correction result output means 17 Information deletion means 18 Recognition result input means 2 Correction terminal 3 Speech recognition device

Claims (6)

A speech recognition error correction assistance device for correcting speech recognition errors, comprising:
An editing means is provided,
The editing means includes:
an editing screen control means for displaying an editing screen including words constituting the speech recognition result and predetermined symbols indicating unrecognized sections of speech for which no recognition result has been generated in the speech recognition;
a voice output means for outputting a voice corresponding to time information of a word or a symbol designated on the editing screen;
an error correction means for updating the word or the symbol with the correction content corrected on the editing screen ,
The speech recognition error correction assistance device according to claim 1, wherein the edit screen control means switches between displaying and not displaying the symbols of the unrecognized section by selecting from a menu screen or by pressing a predetermined shortcut key . A speech recognition error correction assistance device that corrects speech recognition errors at a plurality of correction terminals, comprising:
a storage means for storing predetermined symbols indicating words of the speech recognition result and unrecognized segments of speech for which no recognition result was generated in the speech recognition, in association with respective time information;
A plurality of editing means corresponding to the plurality of editing terminals,
The editing means includes:
an editing screen control means for displaying an editing screen including words constituting the speech recognition result and a predetermined symbol indicating the unrecognized section on the corresponding editing terminal;
a voice output means for outputting a voice corresponding to time information of a word or symbol designated on the editing screen to the corresponding editing terminal;
an error correction means for updating the word or the symbol stored in the storage means with correction contents corrected on the editing screen in a postscript priority manner ;
The speech recognition error correction assistance device according to claim 1, wherein the edit screen control means switches between displaying and not displaying the symbols of the unrecognized section by selecting from a menu screen or by pressing a predetermined shortcut key . The speech recognition error correction support device according to claim 1 or 2, characterized in that the symbol of the unrecognized section is a predetermined symbol indicating unrecognized for each time interval, the unrecognized section being divided into predetermined time intervals, and time information is associated with the symbol. A speech recognition error correction assistance program for causing a computer to function as the speech recognition error correction assistance device according to any one of claims 1 to 3 . A speech recognition error correction assistance method for correcting a speech recognition error, comprising:
displaying an edit screen including words constituting the speech recognition result and predetermined symbols indicating unrecognized sections of speech for which no recognition result was generated in the speech recognition by an edit screen control means;
outputting, by a voice output means, a voice corresponding to time information of the word or symbol designated on the editing screen;
updating the word or the symbol with the correction content corrected on the editing screen by an error correction means;
Including,
The speech recognition error correction support method according to the present invention, wherein the editing screen control means switches between displaying and not displaying the symbols of the unrecognized section by selecting from a menu screen or by pressing a predetermined shortcut key . A speech recognition error correction support method for correcting speech recognition errors using a plurality of correction terminals, comprising:
storing in a storage means, in association with time information, predetermined symbols indicating words of the speech recognition result and unrecognized sections of speech for which no recognition result was generated in the speech recognition;
In the plurality of editing terminals,
displaying an edit screen including words constituting the speech recognition result and predetermined symbols indicating unrecognized sections of speech for which no recognition result was generated in the speech recognition by an edit screen control means;
outputting, by a voice output means, a voice corresponding to time information of the word or symbol designated on the editing screen;
updating said word or said symbol stored in said storage means with correction contents corrected on said editing screen by said error correction means, with postscript taking priority;
Including,
The speech recognition error correction support method according to the present invention, wherein the editing screen control means switches between displaying and not displaying the symbols of the unrecognized section by selecting from a menu screen or by pressing a predetermined shortcut key .

Images