JPH08249343A - Device and method for speech information acquisition - Google Patents

Abstract

PURPOSE: To efficiently perform information retrieval, etc., by extracting information which is characteristic to external real time speech data, from the external real time speech data, etc. CONSTITUTION: A speech signal from a sound source for an external real time broadcast, etc., is inputted to a speech signal input part 1 and converted into digital speech data, which are stored as a speech file in a speech data storage part 3. A news item extraction part 4 extracts speech data by news items from the stored speech data on the basis of wording mode patterns, characteristic of news, in a news broadcast pattern storage part 5 and saves them in a news item storage part 6. Various saves news item data are converted by a speech data recognition part 7 into text data, and a key word appearance frequency extraction part 8 extracts a key word interactively and further finds its appearance frequency. An information retrieval part 10 performs retrieval from an external data base by using a key word of high appearance frequency as a retrieval key.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention extracts information peculiar to audio data from external audio data such as a news program or audio data of an existing audio file (audio data file) to retrieve information. The present invention relates to a voice information acquisition device and a voice information acquisition method suitable for use.

[0002]

2. Description of the Related Art In recent years, various types of systems using computers have become popular at the individual level. In particular, multimedia systems on personal computers
Has produced many uses at the individual level.

For example, the expansion of the personal computer software market for individuals such as in the amusement field is remarkable. Even low-priced personal computers can handle media such as graphic animations and images, and audio data in addition to conventional text information.

The audio data is an existing audio file stored in a CD-ROM or the like (for example, a file of digital audio data in a format represented by * .wav, etc.) by a standard equipment of audio data capturing hardware. It is possible to take in from a microphone. These audio data are used as they are in various software. As described above, in recent years, various types of application software have become popular in the market using various types of media.

[0005]

However, conventionally,
For example, as the audio data, the existing static audio file stored in the CD-ROM or the like is simply used. Also, voice data (real-time voice data) that changes in real time (content) can be input as it is, but it is not considered to be used according to the data content, and it is used temporarily. If you try, it will be difficult to achieve.

[0006] Also, as the types of static voice files become huge, it becomes a heavy task to search for the required voice file only by the file name, for example.

The present invention has been made in consideration of the above circumstances, and an object thereof is to extract information peculiar to audio data from external real-time audio data such as news programs or audio data of an existing audio file. Accordingly, it is to provide a voice information acquisition device and a voice information acquisition method that can efficiently perform information search and the like.

Another object of the present invention is to extract voice data for each news item from the scheduled news program information and obtain a keyword used for information retrieval based on the extracted voice data for each news item. An object of the present invention is to provide a voice information acquisition device and a voice information acquisition method capable of efficiently searching for information and the like which is currently a hot topic.

Still another object of the present invention is to extract the attribute of individual data such as a huge existing voice file and add the attribute information to the corresponding voice file to determine the voice file. To provide a voice information acquisition device and a voice information acquisition method that can give a clue to know what the content is and that can narrow down the candidates of the audio file to be searched based on the attribute information. is there.

[0010]

The audio information acquisition apparatus according to the first aspect of the present invention is an audio signal input means for taking in an audio signal from a sound source such as an external real-time broadcast and converting it into digital audio data. And audio data storage means for storing various digital audio data that has been captured by the audio signal input means, converted and filed, and news broadcast pattern storage means for storing broadcast pattern information of scheduled news. A news item for extracting audio data for each news item as news item data from the audio data stored in the audio data storage means based on the broadcast pattern information stored in the news broadcast pattern storage means. The extraction means and the news item data extracted by this news item extraction means are recorded. For interacting with news item storage means, voice data recognition means for recognizing the news item data stored in the news item storage means by voice, and the news item data recognized by the voice data recognition means. To extract keywords,
A keyword appearance frequency extraction means for acquiring the appearance frequency of the extracted keyword, and an information search means for determining a search keyword based on the keyword extracted by the keyword appearance frequency extraction means and the appearance frequency and performing information search It is characterized by having.

In the audio information acquiring apparatus according to the first aspect, the audio signal input means takes in an audio signal from a sound source such as an external real-time broadcast and converts it into digital audio data. This voice data is made into a file and stored in the voice data storage means. Therefore, by setting a time zone for capturing an audio signal by the audio signal input means, an audio signal from a scheduled news broadcast that is broadcast at a fixed time within a day at a short time can be captured and filed. It is possible to convert the data into a voice data storage means and store it.

The news item extraction means targets the real-time audio data from such filed scheduled news information, and a voice pattern matching the wording pattern peculiar to the scheduled news registered in the news broadcast pattern storage means. By detecting a region and a silent portion for a predetermined time or more, each voice region between the adjacent voice pattern regions (excluding the silent portion) is extracted as one block region of the news item. The extracted voice data of each chunk area is filed as voice data of independent news items (news item data) and stored in the news item storage means.

The voice data corresponding to each news item stored in the news item storage means is converted into text data by the voice recognition processing of the voice recognition means. The keyword appearance frequency extraction means extracts a keyword from this text data through a dialog with the user, and further acquires the appearance frequency of the extracted keyword.

In this way, the news program is broadcast (voiced).
Since the appearance frequency of each keyword is calculated,
The user can know the information that has become a topic in the present world by, for example, display output or voice output.

Further, by using a keyword having a high appearance frequency as a search key, for example, an information search for an external large-capacity database, that is, an information search using news information with real-time property is performed. It is also possible to obtain information related to an event that is a topic in the real world, and information on media with different news sources.

A voice information acquisition apparatus according to a second aspect of the present invention is a voice data storage unit for storing digital voice data of various existing voice files, and each voice stored in the voice data storage unit. A voice attribute information extraction means for extracting attribute information of digital voice data of a file and adding it to the voice file, and an attribute information designated externally of each voice file stored in the voice data storage means are added. The present invention is characterized by comprising an information retrieval means for conducting an information retrieval for an existing audio file.

In the voice information acquisition apparatus according to the second aspect, for an existing voice file, what the voice data of the voice file is, for example, "voice" sound. Or is it a "music" sound,
In the case of the "voice" sound, the attribute relating to the voice data of the voice file, such as the voice of a male voice or the voice of a female voice, is extracted by the voice attribute information extracting means.

Here, as a method for extracting the attribute of the voice data, regarding whether it is a "voice" sound or a "music" sound, the level of the "voice" sound is different from that of the "music" sound. As for the utterance voice, the male voice is 1 when the intense voice is used and it is the male voice or the female voice.
It is used that the frequency of 00 Hz and female voice is almost constant at a frequency of around 200 Hz. In addition, it is also possible to extract attributes such as whether the "voice" sound is a child's voice or an adult's voice.

By adding the attribute information extracted in this way to the corresponding audio file, it is possible to know what the contents of individual data in the existing huge audio file are. It will be a clue. Therefore, for example, when the user specifies desired attribute information, it becomes possible to narrow down the files to be searched for information from among a large number of audio files, and improve the operability in future multimedia usage systems. be able to.

By combining the configuration of the voice information acquisition apparatus according to the first aspect described above and the configuration of the voice information acquisition apparatus according to the second aspect, information retrieval using news information with real time property is possible. Can be performed only for the audio file having the attribute designated by the user, and the necessary information can be easily searched. In the information search targeting the voice file, voice data of the target voice file may be voice-recognized, converted into text data, and a full-text search may be performed on the text data.

[0021]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block configuration diagram showing the overall configuration of a voice information acquisition device according to an embodiment of the present invention.

In FIG. 1, reference numeral 1 designates an external audio signal (real-time audio signal) that changes in real time (content), for example, a radio broadcast, a television broadcast, or an audio signal from a microphone, and converts it into digital audio data. A voice signal input unit 2 for conversion is a time zone setting unit with a clock function for setting a time zone (time zone consisting of a start time and a capture time) for capturing a voice signal by the voice signal input section 1. If the target of real-time audio signal capture by the audio signal input unit 1 is, for example, public scheduled news, the capture time zone set by the time zone setting unit 2 is determined to be 7:00 am, noon, and 7:00 pm. It is a time zone within a certain time range from the time.

The audio signal input section 1 is a radio broadcast or a television during a fixed time period set by the time zone setting section with clock function 2 (that is, in the time zone set by the time zone setting section with clock function 2). It takes in audio signals from broadcasting and converts them into digital audio data. This voice data, for example, the public fixed-time news at 7 pm, is recorded in the voice data storage unit 3 as one independent voice file by being taken in and digitally converted for a fixed time from the start. Saved.

Here, the audio signal input section 1 can change the required file capacity by changing the sampling interval at the time of audio acquisition depending on the availability of the storage area of the audio data storage section 3. ing. Of course, if the required file capacity within a certain time is reduced by increasing the sampling interval, the sound quality of the audio data will deteriorate, but it can be used properly according to the purpose of the audio data.

The voice data storage unit 3 stores real-time voice data (consisting of voice files) captured by the voice signal input unit 1 as well as various existing static voice files read from a medium such as a CD-ROM. Data (digital voice data) is stored. These audio files are audio data used by various software. Further, the voice data storage unit 3 also stores a voice data file having a special voice attribute fetched by the voice attribute information extraction unit 13 described later.

The real-time voice data (file thereof) taken in by the voice signal input unit 1 in the set time zone and stored in the voice data storage unit 3 is an extraction target of the news item extraction unit 4. That is, the news item extraction unit 4
Performs data extraction for each predetermined news item from (the file of) the real-time voice data stored in the voice data storage unit 3. For this extraction
The apparatus of FIG. 1 is provided with a news broadcast pattern storage unit 5 in which a wording pattern peculiar to scheduled news is registered.

The news item extraction unit 4 displays various wording patterns registered in the news broadcast pattern storage unit 5 and silent time zones in (the file of) the real-time voice data stored in the voice data storage unit 3. By checking, data is extracted for each news item.

The extraction processing by the news item extraction unit 4 will be described with reference to FIG. First, FIG. 2 shows a voice waveform of public regular news at 7 o'clock that is taken in by the voice signal input unit 1, and a character string surrounded by "" on the right side of the voice waveform is a wording pattern in the regular news. Is shown.

In the news broadcast pattern storage unit 5, a plurality of wording patterns often used in public scheduled news are registered as dictionary patterns. In particular,
Voice patterns such as "It's news at 7 o'clock", "First, first", "Next", "It's sports information", and "I was in charge of Sakurai" are registered.

The news item extraction section 4 selects the news broadcast pattern storage section 5 from the audio files stored in the audio data storage section 3 by the audio signal input section 1 from the public scheduled news at 7 pm. Voice pattern area R1, which matches the voice pattern registered in
Extract R2, R3, R4 and R5. Note that various techniques for extracting a voice that matches a specific voice pattern have been conventionally known, and therefore description thereof will be omitted here.

Next, the news item extraction unit 4 selects the voice (voice data) region between the extracted voice pattern regions, specifically, the region R2 of the voice pattern of "First, First" and the voice pattern of "Next". A voice (voice data) region RF1 between the region R3, and a voice (voice data) region between a voice pattern region R4 of "Is sports information" and a voice pattern region R5 of "I was in charge of Sakurai". RF2
Each is extracted as one block area of news items.

Between the region R2 and the region R3, there is a silent region S1 following the region RF1, that is, a relatively long silent region S1 between the news, but the silent region S1 (voice data ) Is discarded. That is, RF1
+ S1 is not extracted as one block area of the news item, but RF1 is extracted as one block area as described above.

When extracting the lump area described above, the target voice data (voice waveform data) is preprocessed.
It is preferable to apply smoothing processing to smooth the steep part of the voice waveform. The chunk area is extracted by binarizing the smoothed audio data with a predetermined threshold value.

Now, the voice data (from the outside) extracted for each news item by the news item extraction unit 4 (here, the "first, first" voice pattern and the "next" voice pattern in the scheduled news) News item # 1 between
Voice data of the news item # 2 between the voice pattern of "is sports information" and the voice pattern of "I was in charge of Sakurai" is the news item voice file (here, files A and B). ) Data,
It is stored in the news item storage unit 6.

The news item voice file data stored in the news item storage unit 6 is subjected to voice recognition processing by the voice data recognition unit 7. That is, the voice data recognition unit 7
Performs a voice recognition process on the news item voice file data stored in the news item storage unit 6,
Convert media to text data (consisting of character code string). This text data is stored in the news item storage unit 6 separately from the original news item voice file.

The text data of the news item voice file data converted by the voice data recognition unit 7 is the target of the keyword appearance frequency extraction processing by the keyword appearance frequency extraction unit 8. That is, the keyword appearance frequency extraction unit 8
Displays text data of news item voice file data on a display unit 9 such as a CRT display or a liquid crystal display.
By displaying it on the screen and interacting with the user, the keyword specified by the user is extracted from the text data,
Find the appearance frequency of the keyword. In the present embodiment, this processing is performed in accordance with an instruction given from the user through a keyboard, mouse, etc. (not shown) when the user needs to extract information from an external large-capacity database system. is there.

Therefore, in the apparatus shown in FIG.
Is provided. This information search unit 10 receives the information extraction from the user, and the keyword appearance frequency extraction unit 8
The keyword extracted by is received from the voice data recognition unit 7, and a keyword with a high appearance frequency (for example, three keywords with an appearance frequency of 1st to 3rd) is used as a search key, and an external large-capacity database (Fig. By performing an information search from (not shown), information on events (related to news) that are hot topics in the real world is obtained. The information search result by the information search unit 10 is displayed on the display unit 9.

The number of keywords used in the search can be specified by the user, and the keywords extracted by the keyword appearance frequency extraction unit 8 are displayed on the screen in the order of appearance frequency and used for the search. It is also possible to allow the user to select and specify the keyword.

Further, in the apparatus of FIG. 1, in order to output the voice file data stored in the voice data storage unit 3 and the news item storage unit 6 and further the keywords extracted by the keyword appearance frequency extraction unit 8 by voice. In addition, a voice output unit 11 and a speaker 12 are provided. Audio output unit 1
1 is activated in response to an instruction from the user and is based on the voice file data stored in the voice data storage unit 3 or the news item storage unit 6 or the keyword extracted by the keyword appearance frequency extraction unit 8. The audio signal (acoustic signal) is output to the speaker 12 for audio output. However, in order to output the text data stored in the news item storage unit 6 and the keyword data extracted by the keyword appearance frequency extraction unit 8 by voice, a process of converting the data into voice data is required.

In the apparatus of FIG. 1, a voice attribute information extraction unit for extracting attribute information of voice file data (including existing voice files) stored in the voice data storage unit 3 is targeted. 13 and a voice environment dictionary unit 14 in which information (a kind of dictionary data) for determining the attribute information is registered. The candidate attribute information extracted here is whether the target voice data is voice or music (voice / music distinction), and in the case of voice, is male voice or female voice. Or (male voice /
(Female voice distinction), and further, whether it is a child's voice or an adult's voice (child / adult distinction).

That is, the voice attribute information extraction unit 13 targets the voice file data stored in the voice data storage unit 3 based on the registration information of the voice environment dictionary unit 14 and outputs the voice data (voice file). ) Attribute information,
It distinguishes between "voice" sound and "music" sound, and in the case of "voice" sound, it is a male voice, a female voice, a child voice or an adult voice. Distinguish whether it is voice. Here, for example, the distinction between “voice / music” is that the voice (“voice” sound) being spoken is music (“music” sound).
This is done by utilizing the fact that the level changes more drastically than in. Further, the distinction between male voice / female voice is performed by utilizing the fact that the male voice in the uttered voice is substantially constant at a frequency of 100 Hz and the female voice is substantially constant at a frequency of around 200 Hz.

As described above, whether the sound is a "sound" sound with respect to the individual data of the existing huge sound file and the external sound data which is taken in by the sound signal input unit 1 and stored as a file. Alternatively, if it is a "music" sound or a "voice" sound, it may be a male voice or a female voice, and further, whether it is a child voice or an adult voice. Attribute information is extracted.
By adding this extracted attribute information to the corresponding audio file, it can be used as a clue information for knowing what the contents of individual data in a huge audio file are. . For example, it is possible to efficiently narrow down the necessary file candidates, such as selecting only a female voice audio file from among a huge number of audio files. This makes it possible to improve the operability of future multimedia utilization systems.

Further, the narrowed-down file candidates can be the object of information retrieval by the information retrieval unit 10 as in the above-mentioned external database. In this case, the voice data of each target voice file is read from the voice data storage unit 3 to the voice data recognition unit 7, voice-recognized and converted into text data, and the text data is stored in, for example, the news item storage unit 6. After that, the information retrieval unit 10 targets the corresponding text data in the news item storage unit 6
You can do a full text search at.

The voice attribute information extraction unit 13 extracts the attribute information from the voice file data stored in the voice data storage unit 3 (that is, the voice data once filed), and the voice signal input unit. 1, the real-time voice data input momentarily is directly fetched from the voice signal input unit 1, and the attribute information is extracted, for example, for voice data of a certain time. This is applied, for example, to the purpose of extracting only music sound data or only male sound from the data captured in real time from radio broadcasting or television broadcasting and storing it in the audio data storage unit 3 as an audio file.

As an example, for a real-time sound signal from a radio broadcast or a television broadcast, a function of distinguishing "voice / music" as attribute information is used, and a "music" sound is input from a sound signal that is constantly input. The case of making a file as music sound data will be described with reference to FIG.

First, FIG. 3A shows a sound data acquisition start time t set by the time zone setting unit 2 with a clock function.
During the time width from 0 to time t1, the audio signal input unit 1
Shows a real-time sound signal captured by. The voice attribute information extraction unit 13 extracts the real-time sound signal (voice data) continuously input from the sound data acquisition start time t0, based on the registration information of the voice environment dictionary unit 14,
The attribute information extraction process for identifying the "music" sound is started.

If it is determined that the sound is a "music" sound for the first time at the time ts between the time t0 and the time t1, the voice attribute information extraction unit 13 determines the elapsed time from the time t0 to the time ts. Ts is held and this time information T
Based on s, start to file "voice" sound.

The specific contents of this filing process are as follows. First, in the present embodiment, the audio data continuously input from the audio signal input unit 1 from time t0 is input and held in a shift buffer (not shown) included in the audio attribute information extraction unit 13 by a first-in first-out (FIFO) method. To be done. This shift buffer can input and hold audio data for a time length of 10 seconds, for example.

Therefore, as described above, the voice attribute information extraction unit 13 does not perform "music" for the first time after the time Ts has elapsed from the time t0.
If the sound is judged to be sound, the time from the shift buffer Ts
From the time when the amount of audio data corresponding to is output, the process of converting the output data of the subsequent shift buffer into a file is started.

The voice attribute information extraction unit 13 continues to determine that the real-time sound input after the time ts is the "music" sound, and eventually determines that it is not the "music" sound at the time te shown in FIG. 3B. It is assumed that In this case, the voice attribute information extraction unit 13 sets the time point (time te) when it is determined that the sound is not the "music" sound as the end time point of the "music" sound.

Therefore, the voice attribute information extracting section 13 determines at time t.
The elapsed time Te from 0 to the time t e is held, and when the shift buffer outputs the audio data of the amount corresponding to the time T e, the filing process of the output data of the shift buffer is ended.

Further, the voice attribute information extraction unit 13 determines that the time t
When the end of the "music" sound is judged by e, the time t e is set to the start time t of the sound data acquisition of the new real-time sound signal.
As 0, the identification check of the “music” sound that may be heard is repeated.

In this way, it is possible to automatically capture only the "music" sound (sound data having a desired attribute) from a sound source such as an external real-time broadcast and make it a file. The filed data is stored in the voice data storage unit 3 with the attribute information added. The filed data, that is, the “music” sound taken in from the external sound source is taken out by the voice output unit 11 according to an instruction from the user, and reproduced and output from the speaker 12.

Similarly, from the sound source such as the external real-time broadcast, only the audio of the attribute instructed by the user, such as only the "male" audio, the "female" audio, and the "child" audio, is output. It can be automatically taken in, made into a file, and saved in the voice data storage unit 3.

In this case, for example, only the voice data of the file having the attribute of "male" sound extracted from the regular news program is voice-recognized by the voice-data recognizing unit 7 and converted into text,
By extracting a keyword from the keyword appearance frequency extraction unit 8 and obtaining the appearance frequency of the keyword, it is possible to narrow down the search key used for information search in the information search unit 10, that is, narrow down the search target. Becomes

The present invention is not limited to the above embodiment. For example, in the above-described embodiment, the target of external audio signal acquisition by the audio signal input unit 1 is a scheduled news program, but any program may be used. Also, as the information source, in addition to audio signals from radio broadcasts and television broadcasts, recording media handled by VTR (video tape recorder) devices, CD (compact disc) devices, and audio data from telephone voices are used. It is also possible.

[0057]

As described above in detail, according to the present invention, voice data for each news item is extracted by detecting a wording pattern peculiar to news from the regular news program information, and the extracted news data for each news item is extracted. By converting the voice data into text and interactively extracting the keyword used for information retrieval from it, and acquiring the appearance frequency of the keyword, for example, a keyword with a high appearance frequency with real-time property is used as a search key. Since it is possible to search for information in an external database or the like, it is possible to efficiently obtain currently popular information and information related to the information.

Further, according to the present invention, by extracting the attribute of individual data such as a huge existing audio file and adding the attribute information to the corresponding audio file,
It can give a clue as to what the audio file is about. Therefore, based on the attribute information attached to the audio file, it is possible to narrow down the candidates for the audio file to be searched from among the vast amount of existing audio files, etc. It is possible to improve the sex.

As described above, according to the present invention, the information peculiar to the audio data is extracted from the external real-time audio data such as a news program or the audio data of the existing audio file and the information is used. Thus, information retrieval and the like can be efficiently performed.

[Brief description of drawings]

FIG. 1 is a block configuration diagram showing an overall configuration of a voice information acquisition device according to an embodiment of the present invention.

FIG. 2 is a diagram for explaining an extraction process of a news item extraction unit 4 in the same embodiment, showing a state in which a news item included in scheduled news is extracted using a wording pattern often used in scheduled news. .

FIG. 3 is a diagram for explaining the operation of the voice attribute information extraction unit 13 in the embodiment when a “music” sound is identified from the real-time sound signal input by the voice signal input unit 1 and filed.

[Explanation of symbols]

1 ... Voice signal input section, 2 ... Time zone setting section with clock function (input time zone setting means), 3 ... Voice data storage section, 4 ...
News item extraction unit, 5 ... News broadcast pattern storage unit, 6 ... News item storage unit, 7 ... Voice data recognition unit,
8 ... Keyword appearance frequency extraction unit, 9 ... Display unit, 10 ... Information retrieval unit, 11 ... Voice output unit, 12 ... Speaker, 13 ...
Voice attribute information extraction unit, 14 ... Voice environment dictionary unit.

Claims (10)

[Claims] 1. An audio signal input means for taking in an audio signal from an external sound source such as a real-time broadcast and converting it into digital audio data, and various digital signals taken in by the audio signal input means, converted and filed. A voice data storage unit for storing voice data, a news broadcast pattern storage unit for storing broadcast pattern information of scheduled news, and a news broadcast pattern from the digital voice data stored in the voice data storage unit. News item extraction means for extracting audio data for each news item as news item data based on the broadcast pattern information stored in the storage means, and the news item data extracted by the news item extraction means are stored. News item storage means for this news item A voice data recognition means for voice-recognizing the news item data stored in the storage means and converting it into text, and a keyword is interactively extracted from the news item data recognized by the voice data recognition means, and the extracted keyword A keyword appearance frequency extraction unit that acquires the appearance frequency, and an information search unit that determines a search keyword based on the keyword extracted by the keyword appearance frequency extraction unit and the appearance frequency and performs information search are provided. A characteristic voice information acquisition device. 2. Audio data storage means for storing digital audio data of various existing audio files, and attribute information of digital audio data of each audio file stored in the audio data storage means is extracted. A voice attribute information extraction unit to be added to the voice file, and an information search for performing an information search on a voice file to which externally designated attribute information is added from the voice files stored in the voice storage unit An audio information acquisition apparatus comprising: 3. An audio signal input means for taking in an audio signal from an external sound source such as a real-time broadcast and converting it into digital audio data, and various digital signals taken in by the audio signal input means, converted and filed. Audio data storage means for storing audio data and digital audio data of various existing audio files, and at least the digital audio data of the existing audio files of the digital audio data stored in the audio data storage means. Audio attribute information extraction means for extracting attribute information and adding it to the audio file, news broadcast pattern storage means for storing broadcast pattern information of scheduled news, and the audio signal stored in the audio data storage means From the voice data captured and converted by the input means News item extraction means for extracting audio data for each news item as news item data based on the broadcast pattern information stored in the news broadcast pattern storage means, and news items extracted by the news item extraction means News item storage means for storing data, voice data recognition means for voice-recognizing news item data stored in the news item storage means, and news item recognized by the voice data recognition means A keyword appearance frequency extracting means for interactively extracting a keyword from data and acquiring the appearance frequency of the extracted keyword, and a keyword extracted by the keyword appearance frequency extracting means and a search keyword determined based on the appearance frequency , External database, or the sound Audio information acquisition apparatus characterized by comprising an information retrieval unit which performs information retrieval in a subject an audio file outside the specified attribute information is added among the audio files stored in the storage means. 4. The information search means, when a search keyword is given from the outside, uses the search key,
4. The voice information acquisition apparatus according to claim 3, wherein the voice information acquisition device performs the information search for a voice file to which externally designated attribute information is added among the voice files stored in the voice storage means. 5. The voice attribute information extracting means extracts from the voice data at least a “voice” sound or a “music” sound, and in the case of a “voice” sound, a male voice or a female voice. The voice information acquisition apparatus according to claim 2 or 3, wherein it is identified whether the voice information is voice and corresponding attribute information is extracted. 6. The voice attribute information extraction means inputs the digital voice data fetched and converted by the voice signal input means from the input means, and the voice data portion having an externally designated attribute from the input voice data. 4. The voice information acquisition apparatus according to claim 3, wherein the voice information acquisition unit stores the voice data in the voice data storage unit by adding the attribute information to the extracted voice data. 7. The voice information acquisition apparatus according to claim 1, further comprising an input time zone setting means for setting a time zone for capturing the voice signal by the voice signal input means. 8. A wording pattern peculiar to news is detected from an audio signal from an external sound source such as an external real-time broadcast, converted into digital audio data, saved as a file, and the saved audio data. By doing so, the voice data for each news item is extracted and saved as a file. By performing voice recognition on the saved voice data for each news item, each voice data is converted into a text. Converting to data, interactively extracting keywords from each text data, obtaining the appearance frequency of the extracted keywords, determining the search keyword based on the extracted keywords and their appearance frequency, and performing information retrieval A voice information acquisition method characterized by the above. 9. Attribute information of digital audio data of various existing audio files is extracted and added to the audio file, and an audio file to which externally specified attribute information of each of the audio files is added is extracted. A method for acquiring voice information, characterized in that information retrieval is performed on a target. 10. The attribute information of existing digital audio data of various audio files is extracted and added to the audio file, while an audio signal is taken from an external sound source such as real-time broadcasting and converted into digital audio data. Then, the data is saved as a file, and the voice data for each news item is extracted by detecting the wording pattern peculiar to the news from the saved voice data and saved as a file. By performing voice recognition on the voice data for each news item, each voice data is converted into text data, and keywords are interactively extracted from each text data, and the appearance of the extracted keywords Calculate the frequency, determine the search keyword based on the extracted keyword and its appearance frequency, Parts database or voice information acquisition method characterized in that to perform the subject information retrieval an audio file outside the specified attribute information is added one of the audio file.