JPH1152997A - Speech recorder, speech recording system, and speech recording method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To make it possible to record a summary of conversations as soon as they end or immediately distribute a copy of the record to attendants as a summary of the conference by recording words concerning a trigger word extracted by an extraction means and also a few words preceding and/or following the trigger word. SOLUTION: Speech from a telephone equipment 42 is taken into a speech recognition part 13 through a speech input unit 12. A storage device 14 as a computer device is connected with the speech recognition part 13. Further, the speech recognition part 13 has a speech dictionary part 15 for storing word groups to be recorded from the conversation of the telephone call as a trigger word. Moreover, a speech recorder 11 extracts a word to be recorded from the conversation over the telephone based on the trigger set beforehand, namely, a trigger word. When the trigger word is extracted, it is transmitted to the recorder 14 with a few words preceding and following the trigger word as a recognition result. The recorder 14 records the recognition result, namely, the trigger word and a few words preceding and following the trigger word.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice recording apparatus and system for inputting voice and electronically recording the voice.

[0002]

2. Description of the Related Art Japanese Unexamined Patent Publication No. 4-118800 (name of the invention, a data collection device for plant patrol) discloses a device for collecting data during daily patrols of equipment, devices, and processes in a plant. A voice input device, a voice recognition device for recognizing the input voice, an alarm device for determining the presence or absence of an abnormality in the recognized data and issuing an alarm when there is an abnormality, and a storage for storing the recognized data And a device for collecting data for a plant patrol.

[0003] Japanese Patent Application Laid-Open No. 59-47657 (Title of Invention)
The speech dialogue test apparatus) discloses a device having a plurality of voice input / output units and a voice recognition unit, and capable of performing different voice input / output processes in parallel for a plurality of persons. That is, this publication includes a central processing unit, and a voice interactive test in which a test can be performed on any device under test by repeating a test operation instruction output by voice and a test data input by voice repeatedly. In the apparatus, a plurality of voice interaction input / output processing terminals each having an independent voice recognition function and capable of independently executing a voice interaction type test program are provided in common with the central processing unit, and a plurality of A speech interactive test apparatus is described which is configured to perform a plurality of test operations for test equipment in parallel.

[0004]

Heretofore, there has been no voice recording apparatus for recording a conversation in which participants summarize the conversation at the same time as the end of the conversation.

Conventionally, the minutes of meetings have been created by handwriting or input / output to a personal computer, and confirmation items have been exchanged by facsimile (FAX) or the like.

The present invention provides a voice recording apparatus capable of recording a series of voices at the end of a series of voice inputs or conversations, or distributing a record to participants as a conference summary immediately. , An audio recording system and an audio recording method.

[0007]

According to the present invention, there is provided a voice recording apparatus for digitizing and recording voice, comprising a voice input device for inputting voice, a voice recognition unit for recognizing the input voice, and a trigger specified. A storage device for storing a group of words (trigger words); an extraction unit for extracting a trigger word stored in the storage device from the voice recognized by the voice recognition unit; and a word associated with the trigger word extracted by the extraction unit And a recording device for recording a few words before and / or after the trigger word based on the trigger word, thereby solving the above-mentioned problem.

Further, the present invention relates to a voice recording device for digitizing and recording voice, a voice input device for inputting voice, a voice recognition unit for recognizing the input voice, and storing the recognized voice for a predetermined number of words. A voice storage device, a voice dictionary for recording a word group (trigger word) specified as a trigger, a unit for extracting a trigger word stored in the voice dictionary from the voice recognized, and an extraction unit. An audio recording device is provided that has a word associated with a trigger word and a recording device that also records several words before and / or after the trigger word.

Further, the present invention relates to a voice recording apparatus for digitizing and recording voice, wherein a plurality of voice inputs for inputting conversation voices having a plurality of directions and transmission sources, such as voices communicated through a communication line and voices during a conference. A device, a plurality of voice recognition units for recognizing the input voice, storage means for respectively storing word groups (trigger words) designated as triggers, and a plurality of directions and transmission sources recognized by the voice recognition unit. Extracting means for respectively extracting the trigger words stored in the storage device from the conversation voice held by the user, words associated with the trigger words extracted by the extracting means, and numbers before and / or after the trigger words based on the trigger words And a recording device for recording words.

When a trigger word is extracted by the extracting means, it is desirable to record the trigger word and several words before and / or after the trigger word based on the trigger word.

Further, the present invention relates to a voice recording method for digitizing and recording voice, wherein voice having a plurality of directions and transmission sources is input, the input voice is recognized, and a date, a participant, and a conference are used as triggers. Stores words that correspond to the name, words that are classified into specialized fields that are rarely used in ordinary conversation, and specified words (trigger words), and stores them from conversation voices that have recognized multiple directions and sources. The present invention provides a voice recording method for extracting recorded trigger words, recording a plurality of words related to the extracted trigger words in the order of extraction, and sequentially outputting the recorded words.

[0012]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment according to the present invention will be described below with reference to the drawings. FIG. 1 shows an interactive speech recognition system. This two-way voice recognition system has two voice recording devices 11 and 21 which are
They are tied with one. The figure shows each audio recording device 1
1, 21 are telephone devices 42, 52, respectively, the caller 41,
Reference numeral 51 denotes a state in which voice recording is performed by using.

The sound recording device 11 will be described.

The voice from the telephone device 42 is taken into the voice recognition unit 13 via the voice input devices 12 and 22. The voice recognition unit 13 has a recording device 1 as a computer device.
4 are connected. The voice recognition unit 13 has a voice dictionary unit 15 that stores, as a trigger word, a group of words to be recorded from a conversation during a call.

The voice recording device 11 extracts a word to be recorded, that is, a trigger word, from the conversation of the caller based on a preset trigger. When the trigger word is extracted, it is transmitted to the recording device 14 as a recognition result together with several words before and after the word. The recording device 14
The recognition result, that is, the trigger word and several words before and after the trigger word are recorded. In this example, the trigger word and several words before and after the trigger word are recorded. However, there may be cases where the trigger word and some words before or after the trigger word can be satisfied.

As described above, the voice recording device 11 can be used alone, but exerts its power when used as a two-way voice recognition system as shown in FIG. That is, the trigger word and several words before and after the trigger word are stored in the recording devices 14 and 2.
4 will be recorded simultaneously. The same recognition result is simultaneously recorded in all of the plurality of recording devices on the communication line 31. By doing so, a common call minutes is created in the hands of all correspondents who participated in the call.

The two-way voice input system is a system that provides a recognition function for conference voices having a plurality of directions and transmission sources, such as voices communicated through a communication line and voices during a conference. The voice recognition unit connected to the communication line (or conversation space) where the conference is held always recognizes the conversation, and when a word that triggers during the conference is extracted, the phrase,
And the single word before and after that is recognized and recorded as proceedings. The word to be used as a trigger is registered in advance by the user in the speech dictionary of the recognition device. As a result, only the key points that match the user are recorded without leaking. The word to be used as a trigger is not limited to a fixed word. For example, by specifying "date", "company name", "power plant name", etc., all voices of the date, company name, and power plant name can be recorded. That is, several words before and / or after the trigger word can also be recorded.

If the word used as a trigger is unlikely to appear in ordinary conversation, the speaker can be conscious and say it to increase the efficiency of the system. For example,
When using a system in which triggers such as "accident", "fire", "smoke", and "oil leak" are registered, the speaker can leave the following proceedings by speaking as follows.

"It is a _accident_occurrence at the Maruha Power Station.
The accident was _fire_ and the cause was _oil leak_. [Record]: Accidents, occurrences and details of accidents are fires, causes, and oil leaks at the Maruha Power Station. In some cases, all conversation recognition is performed by the device. In particular, when a plurality of recognition devices are installed, a single recognition record is used, so that all the recognition devices operate in cooperation, and adjustment can be made so that all the recognition records are the same. In this case, an inference algorithm such as probability comparison or majority decision may be used.

The dictionary used in the voice recognition unit will be described.

In the speech recognition system, words to be recognized need to be registered in a dictionary in advance. This dictionary is created and configured as follows.

1. It is advantageous to make it as compact as possible in terms of recognition rate and responsiveness → If you make a huge dictionary,
False recognition increases, and responsiveness also worsens.

2. Glossary is likely to be insufficient in a small dictionary →
Therefore, a certain size is necessary. Therefore, in this system, the dictionary size which is not too large is based on 1000 words, and a plurality of dictionaries up to this size (up to 100) can be built in the system. These dictionaries can be freely and dynamically switched and used even during use of the speech recognition device so that an optimum combination can be selected for each aspect of the application. As a result, while handling a large vocabulary of up to 100,000 words as a whole, this system selects the most appropriate dictionary for each situation to ensure high-speed responsiveness and a good recognition rate.

An example of a dictionary is shown below.

Example 1 [Text], [Reading] Ibaraki Prefecture, Ibarakiken Tochigi Prefecture, Tochigiken Gunma Prefecture, Gunmaken Saitama Prefecture, Saitamaken Chiba Prefecture, Chibaken Tokyo Metropolis, Tokyo, Tokyo Is a simple text file that can be easily edited by anyone using ordinary text editing software. In addition, a designation as to whether or not to make a voice response can be included in the dictionary. In this case, for example, the description is as follows.

Example 2 [Text], [Reading], [Response] Ibaraki prefecture, Ken Ibaraki, 1 {UP}, cursor, 0 Gunma prefecture, Gunmaken, 1 Saitama prefecture, Saitamaken, 1 ＾ {ESC} r check . exe, balance of KENKEN, 0０ {ESC} r completed. In this case, the item where 1 is specified in [Response] is the item where the system repeats the user's voice command, and 0 is added to the item where 0 is specified. The system does not repeat. By using this function, it is possible to make the system make a voice response for only necessary items, and this is particularly effective when registering an operation command to the personal computer in the dictionary as described above.

In the implementation of the dictionary, each line of the dictionary file is indexed by a line number, [Index] + [Reading] is stored in the recognition device, and [Index] + [Text] is stored in the personal computer. Part is stored. By doing so, when the recognition result is transmitted from the recognition device, the corresponding text can be passed to the application by sending only the [index], so that the communication time can be reduced as compared with the case where the [text] itself is transmitted. At the same time, since there is no need to have [text] on the recognition device side, the storage area can be small. However, it is necessary to always synchronize the dictionary active on the personal computer side and the dictionary active on the recognition device side in order to notify the recognition result only by [index]. This part is the [dictionary switching function]. In the case of our system, it can be switched from an external program or by voice command.

In addition, since only one set of dictionaries can be activated at any time in [dictionary switching], it is anticipated that switching of commonly used command words such as the application start text (tenken no jikko) as in Example 2 above will be performed. There is an inconvenience of having to register in all dictionaries. To improve this, the system manages the control dictionary and the general dictionary separately. In this way, the user can easily use the common dictionary by putting the desired phrase in the control dictionary, and switch and put the desired phrase in the general dictionary.

FIG. 2 shows the sound recording units 11 and 12 shown in FIG.
FIG. 3 is an example of a flowchart showing the trigger extraction operation of FIG. 1, and FIG. 3 is an example of a flowchart of the recording output processing portion.

The speech recognition apparatus recognizes a word registered in advance in a dictionary when a speech corresponding to the word is input. If the result is the one designated as a trigger, the speech recognition apparatus transmits the word to the recording apparatus. Perform the recording operation. At this time, not only the word specified as the trigger but also several words before and after the word are recorded at the same time, so the apparatus is provided with a temporary storage area (buffer) having a first-in first-out mechanism. If the number of words to be recorded before the trigger is P, up to P words are recorded in the buffer. Let Q be the number of recorded words of the trigger word, and let C be the trigger word word counter.

A voice signal input from a voice input device (step 1: hereinafter referred to as ST1) is compared with a dictionary by a recognition process: ST2. If a corresponding word exists in the dictionary, it is extracted and sent to ST3. Passed. Whether or not the word is recognized as a trigger (this word is referred to as a trigger word) is determined in ST3. If this word is not a trigger word, first, ST4 checks whether C is zero.

If C is not zero, it indicates that several words following the previously detected trigger word are being recorded and output, so that word is sent to the recording output ST6 as it is. In ST6, the sent word is recorded and output processing: ST12
, And subtracts C in ST7 to return to the beginning. C in ST4
Is zero, the trigger operation is not in progress, and the data is stored in the buffer prepared for P rows in preparation for the next trigger word detection: ST5.

If it is determined in ST3 that the word is a trigger word, it is checked whether or not C is zero in ST8. If C is zero, it means that there are several words before the trigger in the buffer, so that in ST9. First, the word in the buffer is ST1
Output to 2. If C is not zero, there is no word in the buffer, and it is determined that the operation has continued since a new trigger was detected during the previous trigger operation. Therefore, the process proceeds to ST10 and the word is sent to ST12. Finally, Q is set to a counter C for outputting several words of the trigger detection word, and the process returns to the beginning.

The recording output process ST12 in FIG.
This is a subroutine called from ST9 and ST10. The word sent from each caller is transmitted to all the recording devices on the communication line, and returns to the caller.

In the above-described embodiment, taking a telephone as an example, one-to-one
However, the same applies to one-to-N information transmission such as general conferences and video conferences.

[0036]

According to the present invention, by inputting a series of voices or conversations, the summary can be recorded at the same time as the end, or the recording can be immediately distributed to the participants as the conference summary. it can.

According to the two-way speech recognition system of the present invention, the same conversation record can be left for all the participants at the same time as the end of the conversation, and the labor and confirmation items for creating the minutes by hand can be confirmed by facsimile (FAX) or the like Can save time and effort to exchange. This is particularly effective when the time for writing a facsimile (FAX) during an emergency response or an emergency response is regrettable. Further, since the record is digitized, it can be easily registered in a database or the like, and can be left semi-permanently, so that it will be a more reliable reference material in the future.

[Brief description of the drawings]

FIG. 1 is a block diagram of an embodiment of the present invention.

FIG. 2 is a flowchart for an embodiment of the present invention.

FIG. 3 is a flowchart for a part of FIG. 2;

[Explanation of symbols]

11, 21 ... voice recording device, 12, 22 ... input device, 1
3, 23 ... voice recognition unit, 14, 24 ... recording device, 15,
25: Voice dictionary, 31: Communication line.

Claims (5)

[Claims] An audio recording apparatus for digitizing and recording a voice, a voice input device for inputting a voice, a voice recognition unit for recognizing the input voice, and storing a word group (trigger word) designated as a trigger. A storage device to perform, extraction means for extracting a trigger word stored in the storage device from the voice recognized by the voice recognition unit, a word associated with the trigger word extracted by the extraction means,
A recording device for recording a few words before and / or after the trigger word based on the trigger word. 2. A voice recording apparatus for digitizing and recording voice, a voice input device for inputting voice, a voice recognition unit for recognizing the input voice, and a voice storage for storing the recognized voice for a predetermined number of words. A device, a voice dictionary for recording a word group (trigger word) specified as a trigger, a unit for extracting a trigger word stored in the voice dictionary from the voice recognized, and a trigger word extracted by the extracting unit. Related words,
A recording device for recording a few words before and / or after the trigger word based on the trigger word. 3. A voice recording apparatus for digitizing and recording voice, comprising: a plurality of voice input devices for inputting conversation voices having a plurality of directions and transmission sources, such as voices communicated through a communication line and voices during a conference; A plurality of voice recognition units for recognizing the input voice; storage means for respectively storing a group of words (trigger words) designated as triggers; and a conversation voice having a plurality of directions and transmission sources recognized by the voice recognition unit. Extraction means for extracting the trigger words stored in the storage device from the storage device, words associated with the trigger words extracted by the extraction means, and several words before and / or after the words based on the trigger words, respectively. An audio recording system, comprising: a recording device for recording. 4. The method according to claim 3, wherein when the trigger word is extracted by the extracting means, the trigger word is recorded together with several words before and / or after the trigger word based on the trigger word. Characteristic voice recording system. 5. A voice recording method for digitizing and recording voice, comprising inputting voice having a plurality of directions and transmission sources, recognizing the input voice, and corresponding to a date, a participant, and a meeting name as a trigger. Triggered words that are stored in conversational voices that have recognized multiple directions and sources, storing words that are classified, words that are classified into specialized fields that are rarely used in ordinary conversation, and specified words (trigger words). A voice recording method comprising: extracting a word; recording a plurality of words associated with the extracted trigger word in the order in which the words are extracted; and sequentially outputting the recorded words.