JP2004309965A - Conference recording/dictation system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To dictate speeches of persons who attend a meeting as the minutes of the meeting in time series. <P>SOLUTION: Microphones as many as persons who attend the meeting are available and allocated to the persons present, one to one. An input selection part selects a predetermined number of microphones which output speeches. Speech recognition processing parts are as many as the microphones which can be selected by the input selection part and a speech recognized by each speech recognition processing part is saved in a speech storage part as it is; and recognized data are converted into a character string, which is saved in a character string storage part. The character string saved in the character string storage part is displayed by a display means, and an editor can write the minutes of the meeting with the recognition result displayed by the display means and speeches. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a conference recording / transcription system that can transcribe utterances of persons attending a conference and the like in the form of minutes. The term “meeting” as used in this specification has a broad meaning, and it is assumed that there are a large number of speakers such as discussions, collective interviews, and collective bargaining in addition to ordinary meetings.
[0002]
[Prior art]
FIG. 3 is a conceptual diagram for writing up minutes in a conventional conference in time series. In FIG. 3, the voice picked up by the microphone a311, microphone b312,..., Microphone n313 is temporarily stored by the voice storage means a314, the voice storage means b315,. Remembered.
[0003]
The contents of each of the voice storage means 314 to 316 are, for example, converted into character strings by a word processor by a stenographer or character string editing means a317,..., N319 such as voice recognition means (not shown), The series editing unit 320 can arrange the character strings in time series.
[0004]
Further, in general, a plurality of speakers are stored in the voice storage means through one microphone, and while listening to the voices of the speakers stored in the voice storage means, for example, a stenographer uses a word processor or the like to proceed with the proceedings. I transcribed it as a record.
[0005]
In addition, the voice document creation apparatus described in Japanese Patent Laid-Open No. 5-12246 selects a plurality of inputs as one, deletes meaningless words to facilitate kanji conversion, It has a function to edit meeting minutes, which is created in a sentence in a predetermined format.
[0006]
[Patent Document 1]
Japanese Patent Laid-Open No. 5-12246
[Problems to be solved by the invention]
In general, meeting minutes are transcribed as character strings by listening to a recording tape recorded by a tape recorder and performing keyboard input on a word processor or personal computer. Further, instead of the human transcription, the recorded voice of the recording tape is converted into a character string by voice recognition.
[0008]
However, since the voice during a conference recorded on the recording tape records a plurality of people's utterances simultaneously with one microphone, there are many problems even with manual or voice recognition technology. Especially when the distance between the speaker and the microphone is large, the recorded content of the speaker recorded on the recording tape overlaps with the content of other speakers along with the surrounding noise, making the content clear and clear. It was difficult to do.
[0009]
As shown in FIG. 3, it was considered to prepare a microphone for each speaker and record the minutes on a recording tape.
[0010]
However, recording tapes recorded using multiple microphones not only make it difficult to create minutes, because it is necessary to listen to all of the multiple voices including silent sections when speech content is written down. There was a problem that the efficiency became very bad. In addition, it is necessary to write down the utterances input to different microphones in time series in the order of the utterances, which causes a problem that the work is complicated.
[0011]
The problem can be solved by, for example, mixing sounds from a plurality of microphones and recording them on a single recording tape. However, when a plurality of speakers speak, the writing becomes difficult.
[0012]
Also, in order to hear and transcribe the voice recorded on the recording tape, it is necessary to perform operations such as cueing a speaker who wants to transcribe or repeatedly listening to an unclear part of the speaker. The operation needs to be performed frequently such as reproduction, fast-forwarding, rewinding, etc., and there is a problem that efficiency is lowered.
[0013]
As a method of solving the above problem, voice recognition processing is performed for each of a large number of microphones. However, it is impossible to provide a single computer with the same number of speech recognition processing means as the number of speakers because it is a heavy load on the computer except for a meeting of two or three people. It is.
[0014]
Further, since the voice recognition processing is usually performed by selecting and inputting one with the maximum volume, a speaker with a low voice may be ignored when there are a plurality of utterances. Furthermore, the minutes of the meeting may be a problem of silent time, but if the conventional transcription method is used, the time series of the meeting cannot be expressed accurately.
[0015]
In order to solve the above-described problems, the present invention provides a voice recognition process using a single personal computer, even when a plurality of speakers speak at the same time, are recorded separately without duplicate recording. It is an object of the present invention to provide a conference recording / transcription system that can easily transcribe and edit.
[0016]
Applicants are not more than 2 to 3 speakers at the same time, and if more speakers speak, there is no need to take minutes because the conference is in conflict. We focused on that. That is, according to the present invention, it is not necessary to provide as many speech recognition processing means as the number of microphones, even if a large number of speakers are provided with microphones.
[0017]
[Means for Solving the Problems]
(First invention)
The conference recording / transcription system according to the first aspect of the invention is capable of editing a plurality of speaker's utterances as character strings, and includes a plurality of microphones and a predetermined number or less of the microphones among the microphones. An input selection unit for selecting voice output, the predetermined number of voice recognition processing units, a voice storage unit and a character string storage unit for storing voices recognized by the respective voice recognition processing units and their character strings And a character string editing unit that edits a time-series character string based on the voice of the voice storage unit and the character string of the character string storage unit.
[0018]
(Second invention)
In the conference recording / transcription system according to the second aspect of the present invention, the input selection unit selects zero signals, one signal, or a plurality of signals input from the microphone, and the selected signals are subjected to different voice recognition processing. It outputs to each part, It is characterized by the above-mentioned.
[0019]
(Third invention)
In the conference recording / transcription system according to a third aspect of the invention, the input selection unit outputs a signal input from the microphone to a different voice recognition processing unit as a digital or analog signal.
[0020]
(Fourth invention)
In the conference recording / transcription system according to a fourth aspect of the present invention, the input selection unit selects, in a first-come-first-served basis, a signal whose signal strength continuously exceeds a predetermined level for a certain period of time.
[0021]
(Fifth invention)
In the conference recording / transcription system according to the fifth aspect of the present invention, the input selection unit is computer software.
[0022]
(Sixth invention)
In the conference recording / transcription system according to the sixth aspect of the invention, the voice recognition results in the respective voice recognition processing units are displayed side by side in time series, and at the same time, have a function of reproducing the voice corresponding to the voice recognition results. .
[0023]
DETAILED DESCRIPTION OF THE INVENTION
(First invention)
The first invention is a conference recording / transcription system in which the content of a speech can be edited in time series as an accurate character string even if the speech of a plurality of speakers overlaps at the same time. There are as many microphones as there are attendees in the conference, and one microphone is assigned to each attendee. The input selection unit is configured to select a predetermined number or less of the microphones having a sound output among the microphones. Also, the number of microphones is not necessarily the same as the number of attendees. At that time, the attendee speaks after saying his name, or is identified by the voiceprint of the speaker.
[0024]
In a normal meeting, for example, it is assumed that three or more people rarely speak at the same time. The two speakers need to write down because they start and end at the end of a statement and duplicate statements occur in the hands of others. In this case, even if there are many participants in the conference, only two speech recognition processing units are required.
[0025]
The selection of microphones in the input selection unit is based on the premise that only two or three people speak at the same time, even in a conference in which many attend. If more people speak at the same time, the meeting is in a state of conflict and it makes no sense to transcribe those statements.
[0026]
The number of speech recognition processing units is the same as the number of microphones that can be selected by the input selection unit. The voices recognized by the respective voice recognition processing units are stored in the voice storage unit as they are. The data recognized by each of the voice recognition processing units becomes a character string and is stored in the character string storage unit.
[0027]
As for the voice stored in the voice storage unit and the character string stored in the character string storage unit, a voice waveform and a character string as a recognition result are displayed on the display means of the computer. Audio is output as recorded from the computer speakers as needed. The person who transcribes the meeting minutes adopts the speech recognition result displayed on the computer display means if it is correct, and adopts it as it is, and if the speech recognition result is determined to be incorrect, The character string is corrected while listening to the voice.
[0028]
For example, if the person who transcribes the minutes of the meeting determines that the recognition result of the sound is unknown, the person who clicks the button of the waveform portion of the sound displayed on the display means or the vicinity thereof clicks on the meeting. The content of the utterance is reproduced by the speaker. The person who transcribes the minutes of the meeting creates a correct character string in time series based on the recognition result of the voice and the actual recording content.
[0029]
The person who transcribes the minutes of the meeting has the recognition result stored in the voice recognition processing unit stored in the character string storage unit, and the recognition result is displayed on the display means, and at the same time the voice stored in the voice storage unit is heard. However, the character string editing unit can not only make an accurate character string but also edit it correctly in time series.
[0030]
(Second invention)
The second invention considers that there is silent input even when no one speaks during the meeting. That is, the input selection unit selects zero signals (silence), one signal, and a plurality of signals input from the microphone, and outputs the selected signals to different speech recognition processing units. When there are one or two overlapping utterances, different speech recognition processing units use the recognition results as character strings for different speakers.
[0031]
(Third invention)
In the third aspect of the present invention, the signal input from the microphone is converted into a digital signal or an analog signal, so that software can be used for the input selection unit or hardware used for the audio device can be used. Each signal selected by the input selection unit is recognized by a different speech recognition processing unit. The analog signal input from the microphone may be A / D converted before the voice recognition processing unit.
[0032]
(Fourth invention)
In the fourth invention, the selection order in the input selection unit is given. That is, the selection order is a first-come-first-served basis in which the strength of the input signal continuously exceeds a predetermined level for a predetermined time or more. In the fourth aspect of the invention, even when multiple speakers are speaking in duplicate, priority is given with a certain level of strength and continuous time. However, it can be prevented from being selected.
[0033]
(Fifth invention)
According to a fifth aspect of the present invention, when a plurality of speakers are simultaneously recorded by using computer software for the input selection unit, the content recorded by each microphone is recorded separately, and then voice recognition is made accurate. be able to.
[0034]
(Sixth invention)
The sixth aspect of the invention has a function of displaying the voice recognition results in the respective voice recognition processing units in time series and reproducing the voice corresponding to the voice recognition results. The person who writes the minutes of the meeting can easily create the minutes by modifying the voice recognition result in time series while listening to the reproduced voice as necessary.
[0035]
【Example】
FIG. 1 is a schematic block diagram for explaining a system for writing up minutes of a meeting in a meeting room according to an embodiment of the present invention. 1, microphones 1, microphones 2,..., Microphones m are installed in the conference room 11 by the number of speakers (conference participants). Each microphone is connected to a computer input terminal 114 that can be connected to an audio input terminal of the computer 12.
[0036]
The computer 12 includes an input selection unit 121 connected to the computer input terminal 114, a voice recognition processing unit (1) 122 that can recognize the voice selected by the input selection unit 121, and a voice recognition processing unit (2). ,..., A speech recognition processing unit (n) 124, a speech storage unit 125 that stores speech as it is, and a character string that stores the character strings recognized by the speech recognition processing units (1) to (n). The storage unit 126, the speaker output unit 127 that outputs the audio stored by the audio storage unit 125, the display unit 128 that displays the character string stored by the character string storage unit 126, and the output of the speaker output unit 127 And / or a character string editing unit 129 for editing a character string displayed on the display unit 128 and an edited character string 130.
[0037]
Number of attendees of meeting = number of microphones m >> voice recognition processing unit n. Then, the speech recognition processing units (1) to (n) adopt a speaker whose speech signal exceeds a certain level, and ignore those below a certain level. The voice recognition processing units (1) to (n) can simultaneously select n of the speakers 1 to m according to the priority order.
[0038]
FIG. 2 is a diagram for explaining an example of editing using a display unit of a computer according to an embodiment of the present invention. In FIG. 2, the display unit 128 (see FIG. 1) includes an editing area 21, an audio pattern display unit 22, and a conference information display unit 23.
[0039]
In the editing area 21, the recognition result 233 recognized by the voice recognition processing unit by selecting a part of the conference information display unit 23 is displayed. The conference information display unit 23 displays a speaker display unit 231 in which the name of the speaker is displayed, a time display unit 232 in which the speech time is displayed, and a result in which the speech is recognized by the voice recognition processing units 1 to n. A recognition result display unit 233 for displaying the result of editing in the editing area 21.
[0040]
With reference to FIG. 1 and FIG. 2, the editing of the character string based on the speech recognition of the speaker in the meeting will be described. The conference is attended by m people, each equipped with microphones 1, 2,. Each microphone is provided with a computer input terminal 114 and is connected to a terminal (not shown) of the computer 12.
[0041]
Now, it is assumed that two speakers speak simultaneously through the microphone 1 and the microphone 2. The input selection unit 121 selects the outputs of the microphone 1 and the microphone 2 simultaneously, and assigns the output of the microphone 1 to the speech recognition processing unit (1) 122 and the output of the microphone 2 to the speech recognition processing unit (2) 123.
[0042]
The voices recognized by the voice recognition processing unit (1) 122 and the voice recognition processing unit (2) 123 are stored in the voice storage unit 125, respectively. The character strings recognized by the voice recognition processing unit (1) 122 and the voice recognition processing unit (2) 123 are stored in the character string storage unit 126, respectively.
[0043]
The voices and character strings stored in the voice storage unit 125 and the character string storage unit 126 are displayed in the editing area 21 in the display unit 128 of the computer 12 when editing (FIG. 2). For example, when the editor selects a message at 14:54:24 (2341), the speaker Yasumai recognizes that the recognition result “I forgot the big discussion” is displayed in the editing result display unit 234. Is displayed. At the same time, Abe Yuko is recognized as having laughed "haha".
[0044]
Since the content of the recognition result of Yasumoto's utterance stored in the character string storage unit 126 is unknown, the editor clicks on the voice pattern display unit 22 or a button in the vicinity thereof, thereby the voice storage unit 125. The audio output stored in is output as an audio signal from the speaker output unit 127 of the computer 12. In addition, when any one of the speaker display unit 231, the time display unit 232, the recognition result display unit 233, and the editing result display unit 234 is selected, a voice can be automatically output.
[0045]
While listening to the audio signal, the editor determines that “I forgot the big discussion” was “I forgot to write something important.” Edit correctly in area 21. At the same time as the above-mentioned remarks, Abeyuko's remarks of completely different “haha” are processed by the voice recognition processing unit (2) 123 via another microphone 2, and thus are accurately recognized.
[0046]
In the screen shown in FIG. 2, at 14:54:43, Yasmotomai selected the recognition result “Please make all of these attempts to Oshima”. In the editing area 21, “Create all the specifications”. The idea is ... "and the edited state is shown.
[0047]
The correspondence between the speaker and the microphone can be identified by voiceprint from the next time by referring to the name as a self-introduction first, so there is no need to say the name for each statement. Alternatively, the microphone 1 can be associated with Mr. XX in advance, and the microphone 2 can be associated with Mr. □□ in advance.
[0048]
Although the present embodiment has been described in detail above, the present invention is not limited to the present embodiment. The present invention can be modified in various ways without departing from the scope of the present invention described in the claims. The specific technology of the block configuration diagram of the speech recognition processing unit and the like of the present invention is not described in detail because a known or publicly known technology can be used.
[0049]
【The invention's effect】
According to the present invention, paying attention to the fact that the number of people who speak at the same time is about 2 to 3 in the conference, the output from the microphone corresponding to the number of attendees is selected and 2 to 3 people are selected. By simply including the voice recognition processing section, it is possible to easily write up the duplicated speech clearly.
[0050]
According to the present invention, by providing the voice storage unit and the character string storage unit, the character string can be written in time series while listening to the voice.
[0051]
According to the present invention, the name of the speaker can be identified not only by the number or voiceprint of the corresponding microphone, but also the editing result is transcribed along with the speaking time. As for the speech time, the silent time in the venue is known, and an accurate editing result is transcribed as a time series.
[0052]
According to the present invention, since the number of speakers and the number of microphones are the same, even if there are duplicate utterances, the edited result is transcribed accurately and easily. According to the present invention, even if the number of microphones is large, the input selection unit can select only the output of the microphone that has made a statement, so that the number of voice recognition processing units can be processed by a single personal computer. be able to.
[0053]
According to the present invention, since the utterances of a large number of speakers are edited in time series, editing can be performed while paying attention to the utterance contents of a specific speaker. In addition, according to the present invention, editing suitable for various purposes is possible, such as editing the opinions of the pros and cons for the theme separately.
[Brief description of the drawings]
FIG. 1 is a schematic block diagram for explaining a system for writing up minutes of a meeting in a meeting room according to an embodiment of the present invention.
FIG. 2 is a diagram for explaining an example of editing using a display unit of a computer according to an embodiment of the present invention.
FIG. 3 is a conceptual diagram for writing up minutes in a conventional conference.
[Explanation of symbols]
11 ... Conference room 111 ... Microphone 1
112 ... Microphone 2
113 ... Microphone m
114: Computer input terminal 12: Computer 121 ... Input selection unit 122 ... Speech recognition processing unit 1
123... Voice recognition processing unit 2
124... Voice recognition processing unit n
125 ... Voice storage unit 126 ... Character string storage unit 127 ... Speaker output unit 128 ... Display unit 129 ... Character string editing unit 130 ... Character string 21 ... Editing area 211 ..Editing character string 22 ... voice pattern display unit 23 ... conference information display unit 231 ... speaker display unit 232 ... time display unit 233 ... recognition result display unit 234 ... Edit result display portion 2341... Recognition result selection portion

Claims (6)

In a conference recording and transcription system that can edit the speech of multiple speakers as a character string,
Multiple microphones,
Among the microphones, an input selection unit that selects the audio output of a predetermined number of microphones or less;
A predetermined number of speech recognition processing units;
A voice storage unit and a character string storage unit for storing voices recognized by the respective voice recognition processing units and their character strings;
A character string editing unit that edits the time-series character string based on the voice of the voice storage unit and the character string of the character string storage unit;
Conference recording and transcription system characterized by having at least. The input selection unit selects zero signals, one signal, or a plurality of signals input from the microphone, and outputs the selected signals to different speech recognition processing units, respectively. The meeting recording and transcription system described in 1. The conference recording / transcription system according to claim 1 or 2, wherein the input selection unit outputs a signal input from the microphone to a different voice recognition processing unit as a digital or analog signal. 4. The input selection unit according to any one of claims 1 to 3, wherein the input selection unit selects a signal whose signal strength exceeds a predetermined level continuously for a predetermined time in order of arrival. The conference recording and transcription system described. The conference recording / transcription system according to any one of claims 1 to 4, wherein the input selection unit is computer software. 6. The method according to claim 1, further comprising: a function of displaying the voice recognition results in each of the voice recognition processing units in time series, and a function of playing back the voice corresponding to the voice recognition results. Conference recording and transcription system as described in section.

Images