JP4871119B2 - Speech synthesis method, speech synthesizer, program, recording medium - Google Patents

Description

  The present invention relates to a speech synthesis method, speech synthesis apparatus, program, and recording medium that enable a user to listen to text information desired by a user.

Recently, along with the downsizing and weight reduction of storage devices and the low price and large capacity of memory, music and voices that users prefer are stored in small portable devices and can be listened to anywhere and at any time. It has become possible to do. In recent years, with the spread of broadband Internet services using ADSL, optical fiber, etc., a site that distributes music data and voice data over a network has been prepared, and these data can be paid for free or for free on personal computers. Network services that are assumed to be downloaded and listened to by portable audio playback devices are becoming widespread. In addition, a service in which such data is directly downloaded to a mobile phone with a voice or music playback function and listened to by the mobile phone is becoming widespread.
As mentioned above, the easy distribution of voice data and music data is largely due to the broadbandization of Internet services. However, the improvement and standardization of voice and music coding technologies, and the spread of platform software as a foundation. As a result, the process of processing, storing, and distributing such data can be realized easily and inexpensively.

As mentioned in the background section, network distribution of voice and music data processed for distribution by voice and music coding technology allows the use of portable devices on hand at times and places that the general public likes. However, there is a problem that only voice / music data that has been processed in advance and prepared for distribution can be used.
Specifically, for example, when a user wants to listen to the news of the day, weather forecast, etc. by voice, it is possible to obtain voice data by accessing a site such as a newspaper company that distributes such news. It is. However, unlike such TV and radio news, such sites usually deliver audio data only once a day, and there is a problem that the news of the day cannot be known.
Also, there are at most several types of voices prepared in advance at such sites, and it is usually difficult to change the type of voice according to the user's preference and mood.

In addition, blogs and SNS messages that have been attracting attention as new contents of the network in recent years are often distributed only in text information, and there is a problem that such information cannot be heard by voice.
As one method for solving such a problem, a service has been proposed in which text-to-speech is converted using speech synthesis technology and listened to with such a device. By converting text information, which is abundant on the Internet and frequently updated, into speech, it is possible to listen to the latest information desired by the user by speech. In addition, there is a reading software for a personal computer for such a purpose, and it has become possible to listen to such text information by voice by converting it into voice by a personal computer and transferring it to a portable device. Yes.

However, with the current text-to-speech synthesis technology, the given text is simply converted to speech as it is, and if the user wants to listen to various information by speech, all of that information will be converted to speech. In this case, the voice becomes very long, and it is difficult for the user to actually listen to it.
To solve this problem, it is easy to use the document summarization method realized in Reference Document 1 as a means for realizing listening in a short time. By appropriately summarizing the text in which the information desired by the user is described, it is possible to listen in a short time by removing redundant portions.
Further, if the maximum amount defining type information editing device realized in Reference Document 2 is used, it is possible to cope with the user's interest by specifying in advance for each user the type of content of interest and the maximum amount of information to be listened to. It is possible to select content preferentially and listen in a short time.
However, the following problems also exist in these cases.

The first problem is that you can't listen to preferences that vary from person to person, for example, speaking speed is slow, and you don't have much time, so if you want to listen to 5 minutes in total, or want to know as much information as possible, the speaking speed is somewhat faster. However, if you want to keep it for about 20 minutes, or if you want to listen to the information that is of interest to users in a relatively detailed manner, or if you want to summarize the information that is not of interest and that is simple, you should listen to it again. It is impossible to respond to various requests for each user, such as the voice and tone of a famous announcer for news, and the style of entertainment reporter for entertainment information. Points are raised.
Furthermore, as a second problem, in order to cope with various preferences that vary from user to user, even if various settings for summarizing and classifying text such as genres and keywords are prepared, In a situation where various contents are flooded, it is difficult for the user to manually set keywords, genre divisions, and degrees of interest that are directly related to the contents that the user really likes.
Because the above problems exist, listening to audio whose content of information is important compared to listening to music that was created in advance and that is the purpose of listening on the portable device The spread of such services and equipment is difficult.

The speech synthesizer according to the present invention selects an article that matches a set condition from an article group storing a plurality of articles composed of text data, and generates a synthesized speech from text data constituting the selected article. Content creation parameter setting means for setting the speech speed, speaker type, total time length T, and the number of required articles to be reproduced, and keyword set storage means for the keyword that the user is interested in and the degree of interest in the keyword For each article i included in the article group and the keyword setting means to be set in the article group, the interest of the article (i) based on the degree of interest of the keyword appearing in the article i among the keywords accumulated in the keyword accumulating means degrees alpha and interest level determining means for determining (i), DOI articles i ranging from 0 to 1 alpha to (i) The summary level A (i) of each article i is obtained by f (α (i)) using a predetermined function f that increases in value, and the article i is ordered in descending order of the degree of interest α (i) of the article i. Article i is extracted and extracted in a range where the sum of the time length L (i) and the summarization level A (i) of the article i is less than the total time length T when the text before being summarized is speech-synthesized The summary level determination means for assigning the summary level A (i) to the article i, and the article i to which the summary level A (i) is assigned corresponds to the article i based on the summary level A (i) Text summarizing means for summarizing the text to be performed, and text speech synthesizing means for synthesizing speech from the summarized text based on the speech speed and speaker type set by the content creation parameter setting means. .

The speech synthesizer according to the present invention further includes text classification means for classifying a plurality of articles into a plurality of predetermined genres in the speech synthesizer described above, and a desired synthesized speech for each genre in the content creation parameter setting means. The speech speed and speaker type are set.
The speech synthesizer according to the present invention further includes an operation target keyword determining means for determining a keyword related to an article to be operated when the user operates the device and listens to the sound in the speech synthesizer described above, An interest level adjusting means is provided for adjusting the degree of interest related to the determined keyword in accordance with an operation parameter when the device is operated.

By providing means to determine the summarization level for each text content desired by the user according to the requirements such as the length of time and speech speed for the synthesized sound of the user, individual differences and preferences of individual users' hearing ability Based on the time that can be allocated for listening to voice, etc., it is possible to listen to information by voice synthesis adjusted in individual units, and it is possible to provide optimal voice information for each user, which has been difficult until now .
In addition, by providing means for determining the speech speed desired by the user for each type of text information, the summary level of the information content, the type and tone of the synthesized speech speaker, Depending on the content type, listen to specific information slowly and carefully, listen to other information quickly, listen all at once, or adjust to the listening style and requirements of each individual. Is possible.

At the same time, it is possible to speak relatively slowly in the form of a calm announcer like a male if it is a news or tone, depending on the content of the information and personal preferences, for example, in the case of news information, a female voice that is bright and active It becomes possible to select a speaker / speaking style that speaks, and it is possible to solve problems such as tiredness and fatigue that are likely to occur when listening to speech for a relatively long time and to respond to personal preferences. .
Operation information related to operations performed by the user during audio playback, such as fast-forward, skip, and rewind, is fed back to the generation side of the synthesized voice data, and the degree of interest in the genre keyword that is the target of those operations By providing a means that can be adjusted on a per-user basis, the user can accurately listen to it, feel free to listen to it, listen to it quickly because it is not very interested, or skip it in some cases. The degree of interest set by genre unit, such as soccer, baseball, movies, etc., more detailed topics about specific teams, specific players, specific directors, specific actors, etc. , It is possible to provide information based on personal interests of users and likes and dislikes that are not usually conscious .

The speech synthesizer that implements the speech synthesis method according to the present invention can be entirely configured by hardware. However, in order to achieve the simplest implementation, the speech synthesis program according to the present invention is installed in a computer, and the present invention is installed in the computer. An embodiment that functions as a speech synthesizer according to the present invention and executes the speech synthesis method according to the present invention is the best.
When the computer functions as a speech synthesizer according to the present invention, the speech synthesis program according to the present invention is installed in the computer, and a plurality of articles consisting of text data are stored in the computer, and the conditions set from these article groups are met. Article selection means for selecting and taking out articles, content creation parameter setting means for setting the desired speech speed, speaker type, and time length of the synthesized speech, keywords for which the user is interested, and keywords for setting the degree of interest Setting means, interest level determination means for determining the interest level of each article based on the set keyword and degree of interest, correspondence to articles based on the set speech speed and listening time, and the determined interest level A summary level determination means for determining the summary level of the text to be processed and the article based on the summary level Construct text summarizing means for summarizing text and text speech synthesizing means for synthesizing speech from text summarized based on speech speed and speaker type set by content creation parameter setting means and function as a speech synthesizer A first embodiment,
Furthermore, in the speech synthesizer described above, further comprising text classification means for classifying a plurality of articles into a plurality of predetermined genres,
A second embodiment in which content creation parameter setting means functions as a speech synthesizer to which means for setting desired speech speed and speaker type for each genre is added;
Further, in the speech synthesizer described above, an operation target keyword determination means for determining a keyword related to an article to be operated when operating a device and listening to speech, and a degree of interest related to the determined keyword, A third embodiment may be considered that functions as an apparatus including an interest level adjusting unit that adjusts according to an operation parameter when the device is operated.

FIG. 1 shows a conceptual configuration diagram of a speech synthesizer according to Embodiment 1 of the present invention.
The speech synthesizer 100 proposed in the first embodiment includes an article group 105 that stores a plurality of articles made up of text data. When generating synthesized speech from an article selected from the article group 105, a desired synthesized speech is spoken. Based on the set keyword and the degree of interest, the content creation parameter setting means 101 for setting the speed, speaker type, and time length, the keyword setting means 114 for setting the keyword and the degree of interest that the user is interested in Interest level determination means 104 that determines the level of interest of each article, summary level determination that determines the level of text corresponding to the selected article based on the set speech speed and listening time, and the determined level of interest Means 106, text summarizing means 112 for summarizing the text corresponding to the article based on the summarization level, and content creation parameters. And a text-to-speech synthesis means 107 for synthesizing speech from the summary text on the basis of the set speech speed and speaker type by data setting means 101.

  The content creation parameter setting means 101 is a means for the user to set content creation parameters in the content creation parameter storage means 102. The setting means can be realized using, for example, a GUI screen shown in FIG. The content creation parameters include at least the total length of speech, the speed of speech, and the type of speaker. Depending on how the summary level determination means 106 is implemented, the content creation parameters may include the number of essential playback articles. The total length of time and the number of required playback articles can be entered in a form that fills in the numerical values. For the speaking speed, a slider for specifying the degree of speaking speed is presented. For the speaker type, several options are presented. The setting should be made using radio buttons.

The keyword setting unit 114 is a unit in which a user sets a keyword related to an article of interest to the keyword set storage unit 111. The keyword setting unit 114 sets at least the keyword and the degree of user interest in the keyword in the keyword set storage unit 111. The setting means can be realized by using, for example, a GUI screen shown in FIG. A keyword can be set by using a form for freely inputting text (text data is input in the input field 111A), and the degree of interest can be realized by using several stages of radio buttons 111B. The keyword set by the user is not limited to one, and a plurality of keywords can be set according to the user's interest.
FIG. 4 shows an operation flow of the speech synthesizer 100.
For each of the articles stored in the article group 105 in the interest level determination step SP <b> 1, an interest level is assigned using the interest level determination means 104 and accumulated in the interest level added article accumulation means 103.

Next, for each article to which the degree of interest is given in the summary level determination step SP2, a summary level is determined using the summary level determination means 106, this summary level is assigned to each article, and a summary level-added article storage means 109 is provided. To accumulate.
In the text summarization step SP3, the text summarizing means 112 is used to summarize the text for each article to which the summarization level is given, and is stored in the summary article accumulating means 108.
In the speech synthesis step SP4, the text speech synthesis means 107 synthesizes speech for each text of the summarized articles stored in the summary article storage means 108, and the user is connected via the synthesized speech storage means 113 and the communication network 200. To the terminal 300.

Hereinafter, the constituent operations of each means will be described in more detail.
The interest level determination means 104 assigns an interest level α (i) to each article i included in the article group 105. For example, for the article i, the interest level determination unit 104 examines the keywords stored in the keyword set storage unit 111 that appear in the article i, and has the highest value among the interest levels corresponding to the keywords included in the article i. A large degree of interest can be realized by determining the degree of interest α (i) of article i.
The summary level determination means 106 determines an article to be summarized from the set of articles to which the degree of interest stored in the interest level-added article storage means 103 is given, and simultaneously sets the summary level for each target article. decide. Here, the summarization level is a value obtained by dividing the time length when the synthesized text is synthesized by speech by the speech length when the synthesized text is synthesized by speech. That is, the summarization level is expressed by a numerical value between 0 and 1, and the closer the value is to 1, the more the original text is not summarized, and the closer the value is to 0, the shorter the summarization is.

Here, T is the total audio time length specified by the content creation parameter, MinArt is the number of required playback articles, and L (i) is the time length when the text before being summarized for each article i is synthesized. The degree of interest is α (i). The degree of interest α (i) is assumed to be realized by a numerical value in the range of 1 to 5.
At this time, the summary level A (i) of the article i can be obtained by the following summary level determination method 1 or summary level determination method 2, for example.

Summarization level determination method 1
First, it is assumed that the interest level weight w (α (i)) takes a value in the range from 0 to 1 that increases in accordance with the interest level. For example, w (α (i)) is obtained by dividing (α (i) -1) by 4.
Step 1, articles i are extracted in descending order of interest α (i) until the sum of L (i) * w (α (i)) exceeds T. If the number of articles extracted is less than the required number of articles to be reproduced MinArt, MinArt articles are extracted in descending order of interest α (i). Let S be the total sum of L (i) * w (α (i)) for the extracted articles.
Procedure 2, summarization A (i) is determined by the following equation.
A (i) = w (α (i)) * (T / S)
The summarization level A (i) obtained by the above-described method 1 becomes higher as the degree of interest is higher, and is obtained by multiplying the audio length L (i) corresponding to each article by A (i). The sum of the speech lengths L (i) * A (i) when the synthesized text is speech synthesized is T.

Summarization level determination method 2
For example, a summary level function f (α) that uniquely determines a summary level for each degree of interest α (i) is prepared. f (α) takes a value in the range from 0 to 1, and for example, the following function can be used.
f (α) = ((α−1) / 8) +0.5
If the summarization level A (i) regarding the article (i) is set to f (α (i)) using such a function f (α), the summarization level is 0.5 when the degree of interest α is 1, When the degree of interest is 5, the summarization level is 1.0. Therefore, articles with high interest are not summarized very much, and articles with low interest are summarized into short text. Using such a function, if an article is extracted from a set of articles to which the degree of interest is given in descending order of interest, the sum of L (i) * A (i) is less than T. The target article can be determined. In the case of the summary level determination method 2, the parameter of the number of essential reproduction articles is not used as the content creation parameter.
In the embodiments of the above two summary level determination methods 1 and 2, it is necessary to know the speech length L when the text is speech-synthesized. Can be obtained by synthesizing. Of course, the method of realizing the summary level determination method is not limited to these methods.

The text summarizing means 112 summarizes the article text to which the summary level is given according to the summary level. For example, it can be realized by using the “document summarization method” realized in Reference Document 1.
The text-to-speech synthesizer 107 converts the summarized article text into speech corresponding to the speaker type and speech speed specified by the content creation parameters. For example, the “speech synthesis method” according to Reference 3 can be realized by preparing a reference codebook or a difference codebook for a plurality of speakers and selectively operating them.

Here, the operation of the above embodiment will be described using specific articles and keyword examples. Assume that the keyword set stored in the keyword set storage unit 111 is as shown in FIG. In addition, the article group 105 includes six articles as shown in FIG. 6, articles 1 to 4 each include one keyword, article 5 includes two keywords, and article 6 includes keywords. Suppose that it was not included. Further, it is assumed that the content creation parameters designated by the user and stored in the content creation parameter storage unit 102 are as shown in FIG.
At this time, the interest level determination means 104 determines the interest level for each article. Since the interest level of the article is the interest level of the keyword having the highest interest level among the keywords included in the article, the interest level granting article is as shown in FIG.

The summary level determination means 106 determines a summary level for each article. Hereinafter, although an example of operation is shown, it is assumed that the voice length L (i) before summarization when the text of each article is voice-synthesized is a time as shown in FIG.
At this time, the operation when the summary level determination method 1 is used is as follows.
First, the interest level weight w (α) is determined. Since article 3 has an interest level of 5, dividing (5-1) by 4 results in an interest degree weight of 1.0. Similarly, articles 1 and 5 are 0.75, and articles 2 and 4 are 0.5 and Article 6 is 0.
Next, articles are extracted in descending order of interest as long as L (i) * w (α (i)) does not exceed 10 minutes. Here, for example, articles 1 and 5 and articles 2 and 4 having the same degree of interest may be selected at random.

  When only the article 3 is selected, the sum of L (i) * w (α (i)) is 6.0 (6 * 1.0). Next, when article 1 is selected, the total is 6 * 1.0 + 7 * 0.75 = 11.25. Here, the total exceeds the total audio time length, but the number of mandatory playback articles specified by the content creation parameter is 4, so two more articles need to be extracted. Therefore, article 5 and article 2 are selected in descending order of interest. In this case, the sum of L (i) * w (α (i)) is 6 * 1.0 + 7 * 0.75 + 10 * 0.75 + 6 * 0.5 = 21.75. If the formula of procedure 2 is applied here, the summary level of each article is determined. Specifically, the summary level of article 3 is 1.0 * (10.0 ÷ 21.75) = 0.46, and the summary level of article 1 and article 5 is similarly 0.34, The summarization level is 0.23. FIG. 10 shows an article with a summary level obtained in this way.

The operation when the summary level determination method 2 is used is as follows.
First, for the article i, the value of the summary level A (i) is calculated using the summary level function. Since article 3 has an interest level of 5, ((5-1) / 8) + 0.5 = 1.0. Articles 1 and 5 are similarly ((4-1) / 8) + 0.5 = 0.875, articles 2 and 4 are 0.62, and article 6 is 0.5. Here, articles are extracted in a descending order of interest so long as the sum of L (i) * A (i) does not exceed T. When article 3 is extracted, 6 * 1.0 = 6, but when article 1 is further extracted, 6 * 1.0 + 7 * 0.875 = 12.12, which exceeds the total time length T. Therefore, when the summary level determination method 2 is used, only the article 3 is extracted, and the summary level is 1.0.

11 to 15 show examples of the speech synthesizer according to the second embodiment of the present invention.
The second embodiment is characterized in that a text classification unit 121 and a genre-specific article group 122A shown in FIG. 12 are newly provided. That is, a plurality of articles included in the article group 105 are classified by genre by the text classifying unit 121 based on, for example, the keyword appearance rate determined for each genre, and sorted and stored in the genre-specific article group 122A. In the keyword setting means 114, the user sets the keyword and the degree of interest together with the genre belonging to the keyword. Also, the content creation parameter and the interest level are set together with the genre to which the keyword belongs. The content creation parameter setting unit 101 also sets a part of the content creation parameters for each genre.
Therefore, the interest-giving article accumulating means 103 shown in the first embodiment is the genre-specific interest-giving article accumulating means 103A, the content creation parameter accumulating means 102 is the genre-specific content creating parameter accumulating means 102A, and the keyword set accumulating means 111 is The genre-specific keyword set storage means 111A, the summary level-added article storage means 109 are changed to the genre-specific summary level-added article storage means 109A, and the summary article storage means 108 are changed to the genre-by-genre summary article storage means 108A.

FIG. 13 shows a processing flow of the speech synthesizer according to the second embodiment.
First, in the text classification step SP11, a plurality of articles included in the article group 105 are classified for each predetermined type of genre.
Next, in the interest level determination step SP12, the degree of interest is given to the articles classified by genre using the interest level of the keyword specified in association with the corresponding genre.
Next, in the summary level determination step SP13, the articles classified by genre are summarized according to the summary level assigned to the article.
Finally, in the speech synthesis step SP15, the articles classified and summarized for each genre are converted into speech according to the content creation parameters and presented to the user terminal 300.

Hereinafter, embodiments of each means will be described.
The content creation parameter setting means 101 is a means for the user to set content creation parameters for each genre and parameters relating to the overall content configuration. For example, it is realized using a GUI screen shown in FIG. The content creation parameters include at least the total length of speech, the speaking speed in each genre, and the type of speaker, and may include the number of essential playback articles depending on the method of realizing the summary level determination method. .
In addition, it includes a genre playback order that is the order in which audio is finally presented to the user. The total length of time may be entered on a form for entering numerical values. For GUIs that specify the speakers, speaking speeds, and the number of required playback articles for each genre, the GUIs similar to those illustrated in FIG. 2 can be written together with the GUI that specifies the target genre in a selection list or the like. realizable. Also, for the genre playback order, prepare a screen in which genres are displayed side by side, select an arbitrary genre with the mouse pointer, etc., and then operate the buttons to advance or delay the playback order to move the playback order back and forth. This can be realized by using such a GUI.

The keyword setting unit 114 sets keywords that the user is interested in for each genre. For example, using the GUI screen shown in FIG. 15, FIG. 3 can be realized by combining a similar GUI and a GUI for specifying the target genre by a selection list or the like.
The text classification unit 121 automatically classifies a plurality of articles included in the article group 105 into a predetermined number of genres. For example, if a large number of articles classified and published according to an existing classification system on the Web or the like are collected and used as an article group, it can be realized by using the destination where the article was originally classified as a genre.
The interest level determination means 104 can be realized by operating the interest level determination means 104 in the speech synthesizer 100 according to the first embodiment independently for each genre.
The summary level determination means 106 can be realized by causing the summary level determination means 106 in the speech synthesizer 100 according to the first embodiment to operate the summary level determination method 2 independently for each genre. This can also be realized by obtaining the summary level A (i) of the article i by the summary level determination method 3 in FIG.

Summarization level determination means 3
First, it is assumed that the interest level weight w (α (i)) takes a value in the range from 0 to 1 that increases in accordance with the interest level. For example, w (α (i)) is obtained by dividing (α (i) -1) by 4.
Step 1, until the sum of L (i) * w (α (i)) exceeds T, articles i are extracted in descending order of interest α (i) regardless of the genre. When the number of articles extracted for each genre is less than MinArt set for each genre by the content creation parameter setting means at the time of completion of extraction, the degree of interest α ( Articles are additionally extracted in order from the largest of i) until the number reaches MinArt.
Let S be the total sum of L (i) * w (α (i)) for the extracted articles.
Procedure 2, summarization A (i) is determined by the following equation.
A (i) = w (α (i)) × T / S
The summarization level A (i) obtained by the above method has a higher summarization rate as the degree of interest is higher, and further, the summation obtained by multiplying the speech length L (i) corresponding to each article by A (i). The sum of the voice length L (i) * A (i) when the subsequent text is voice-synthesized is T and always includes the number of MinArt articles set for each genre.

The text summarization unit 112 can be realized in the same manner as the text summarization unit 112 in the speech synthesizer 100 according to the first embodiment.
The text-to-speech synthesizer 107 converts the summarized article into speech for each genre. At this time, the sound is converted into sound corresponding to the speaker type and speaking speed for each genre specified by the content creation parameter. In addition, the audio is sequentially presented to the user's terminal 300 in accordance with the genre playback order specified in the content creation parameter.

16 and 17 are functional configuration diagrams of the speech synthesizer according to the third embodiment of the present invention.
In the speech synthesizer 100 according to the third embodiment, when the articles stored in the article group 105 are classified by the text classification unit 121, a unique article ID is assigned to each article so that each article can be identified. Each time an article is selected, operation information such as the user's fast forward and skip rewind is accumulated using the article ID as a clue, and the interest level of the keyword used to select each article based on this operation information is determined. It is characterized in that it is configured to automatically adjust.
For this reason, as the structural changes, the operation target keyword determination means 131 and the interest level adjustment means 132 are newly provided, and the genre-specific article group 122A shown in FIG. The genre-specific article group 122B is changed, the genre-specific interest-added article storage means 103A is changed to the genre-specific genre-specific interest-added article storage means 103B, and the genre-specific summary level-added article storage means 109A is classified by genre with article ID. The summary level-added article storage means 109B is changed, the genre-by-genre summary article storage means 108A is changed to the genre-by-genre summary article storage means 108B, and the genre-by-genre synthesized speech storage means 113A is genre-by-genre-by-genre synthesized speech storage means. This is a change to 113B.
With these structural changes, operation parameters of a device such as the terminal 300 that is operated when the user listens to the voice are input to the operation target keyword determination unit 131 provided in the voice synthesizer 100. Further, using the keyword determined by the operation target keyword determination unit 131 and the operation content operated by the user, the interest level adjustment unit 132 automatically calculates the interest level of the keyword set held in the genre-specific keyword set storage unit 111A. To adjust.

FIG. 18 shows a processing flow of the speech synthesizer 100 according to the third embodiment.
From the text classification step SP21 to the speech synthesis step SP25 is similar to the steps SP11 to SP15 in the speech synthesis apparatus 100 according to the second embodiment, but the features of the third embodiment are classified in the text classification step SP21. A unique article ID is assigned to each article. Further, the article ID is inherited to the corresponding article in the interest level determining step SP22, the summarizing level determining step SP23, and the text summarizing step SP24. In the text speech synthesizing step SP25, the text speech synthesizing means 107 converts the text corresponding to the article into speech. After the conversion, the voice is presented to the user together with the article ID.
In the voice listening / apparatus operation step SP26, the user listens to the voice presented from the speech synthesizer 100, and the user performs equipment operations such as fast forward, rewind, and reproduction skip according to his / her request.

In the operation target keyword determination step SP27, the operation target keyword determination means 131 identifies which article is the target voice for which the user has operated the device, and which keyword the article is extracted by. To decide.
In the interest level adjustment step SP28, the interest level adjustment unit 132 increases or decreases the interest level of the keyword determined in the operation target keyword determination step SP27 in accordance with the user's device operation. For example, if fast-forwarding or skipping operations are performed, the interest level is decreased, and when a rewinding operation is performed, the interest level is increased. Even for keywords that are strengthened or faded, the degree of interest at the time of listening to the user can be reflected in the keyword set, and an appropriate summarization level can be provided in the subsequent speech synthesis.

Hereinafter, examples of the text-to-speech synthesis unit 107, the operation target keyword determination unit 131, and the interest level adjustment unit 132 will be described.
In the text-to-speech synthesizer 107, the text corresponding to the summarized article is converted into speech as in the second embodiment, but when presenting speech to the user, the data related to the article ID is embedded and presented together with the speech data. . This can be realized by using an audio format that can be tagged, such as mp3 format or WAV format.
It is assumed that the terminal where the user listens to the voice can perform operations such as fast-forwarding, rewinding, and reproduction skip, for example. Further, it is assumed that the terminal can communicate the operation information and the article ID embedded in the voice to the voice synthesizer 100 through the network communication network 200 or the like.

The operation target keyword determination unit 131 can extract an article corresponding to the article ID transmitted from the terminal 300 by referring to the genre-specific article group 122B with article ID. Furthermore, the operation target keyword can be determined by extracting the keyword included in the article that is the operation target from the genre-specific keyword set.
The interest level adjustment unit 132 reduces the interest level by 0.1 when the operation information is fast-forwarded with respect to the keyword interest level determined by the operation target keyword determination unit 131, and the operation information is a reproduction skip. The interest level can be adjusted by reducing the interest level by 0.2 and adding 0.1 when the operation information is rewind. When the interest level becomes a value smaller than 1 or a value larger than 5 as a result of the operation, the adjustment of the interest level can be limited so as not to exceed the corresponding range.
Of course, the method for adjusting the interest level is not limited to this, and any method may be used as long as the operation information and the interest level are associated with each other.

Each of the speech synthesizers 100 described in the first embodiment, the second embodiment, and the third embodiment described above is realized by installing a speech synthesis program according to the present invention in a computer and causing a CPU provided in the computer to execute the speech synthesis program. The
The speech synthesis program according to the present invention is described in a computer-readable program language, and is recorded on a recording medium such as a magnetic disk, CD-ROM, or semiconductor memory readable by the computer. It is installed in a computer from these recording media or through a communication line, and functions as a speech synthesizer.
[Reference 1] Japanese Patent Laid-Open No. 11-338867 [Reference 2] Japanese Patent Laid-Open No. 5-225255 [Reference 3] Japanese Patent No. 3444395

  It is used for the operation of information service sites.

The function block diagram for demonstrating Example 1 of this invention. The figure for demonstrating the detail of the content creation parameter setting means shown in FIG. The figure for demonstrating the detail of the keyword setting means shown in FIG. The flowchart for demonstrating the operation | movement flow of the speech synthesizer shown in FIG. The figure for demonstrating an example of the content of accumulation | storage of the keyword set storage means shown in FIG. The figure for demonstrating the content of the article group shown in FIG. The figure for demonstrating an example of the storage content of the content creation parameter storage means shown in FIG. The figure for demonstrating an example of the accumulation | storage content of the interest degree provision article | item accumulation means shown in FIG. The figure for demonstrating the mode in the article group shown in FIG. The figure for demonstrating the storage content of the summary level provision article storage means shown in FIG. The function block diagram for demonstrating a part of Example 2 of this invention. FIG. 12 is a functional configuration diagram for explaining the continuation of FIG. 11. FIG. 13 is a flowchart for explaining the operation of the second embodiment shown in FIGS. 11 and 12. FIG. The figure for demonstrating the detail of the content creation parameter setting means shown in FIG. 11 and FIG. The figure for demonstrating the detail of the keyword setting means shown in FIG. 11 and FIG. The function block diagram for demonstrating a part of Example 3 of this invention. FIG. 17 is a functional configuration diagram for explaining the continuation of FIG. 16. 18 is a flowchart for explaining the operation of the third embodiment shown in FIGS. 16 and 17.

Explanation of symbols

100 Speech synthesizer
101 Content creation parameter setting means
102 Content creation parameter storage means
102A Genre-specific content creation parameter storage means
103 Means for adding articles of interest
103A Genre-specific interest level article storage means
103B Article ID-added interest level-accumulated article accumulation means
104 Interest level determination means
105 articles
106 Summarization level determination means
107 Text-to-speech synthesis means
108 Summary article storage means
108A Summary article storage means by genre
108B Summary article storage means by genre with article ID 109 Summary level assignment article storage means 109A Summary level assignment article storage means by genre 109B Summary level assignment article storage means by genre with article ID 111 Keyword set storage means 111A Keyword set storage means by genre 112 Text summarizing means 113 Synthetic voice accumulating means 113A Genre-specific synthetic voice accumulating means 113B Article ID-added genre synthesized voice accumulating means 114 Keyword setting means 121 Text categorizing means 122A Genre-specific article group 122B Article-ID-genre article group 131 Operation target Keyword determining means 132 Interest level adjusting means 200 Communication network 300 Terminal

Claims (10)

In the speech synthesis method for generating synthesized speech from text data constituting the selected article by selecting an article that matches the set condition from an article group storing a plurality of articles consisting of text data,
A content creation parameter setting step for setting the desired synthesized speech speed, speaker type, total time length T, and number of required playback articles ;
A keyword setting step for setting the keyword and the degree of interest in the keyword in the keyword set storage means ;
For each article i included in the article group, the degree of interest α (i) of the article i is determined based on the degree of interest of the keyword appearing in the article i among the keywords stored in the keyword set storage unit. Interest level determination step,
The interest level weight w (α (i)) for each article i is determined so that the value increases in accordance with the degree of interest α (i) in the range from 0 to 1, and the text before the article i is summarized The sum of multiplication of the time length L (i) and the interest weight w (α (i)) of the article i exceeds the total time length T, and the number of extracted articles is the required number. The article i is extracted in descending order of the degree of interest α (i) of the article i so as to satisfy the number of reproduced articles, the time length L (i) for the extracted article i and the interest degree weight w (α ( From the sum S of multiplications of i)) and the total time length T, the summary level A (i) of each article i extracted is determined as the interest weight w (α ((i)) * (total time) of the article i. A summarization level determination step that determines the summation level A (i) to the extracted article i, determined by the length T / sum S) ;
A text summarizing step for summarizing text corresponding to the article i based on the summary level A (i) for the article i given the summary level A (i) ;
A text-to-speech synthesis step of synthesizing speech from the summarized text data based on the speech speed and speaker type set by the content creation parameter setting means;
A speech synthesis method comprising:   In the speech synthesis method for generating synthesized speech from text data constituting the selected article by selecting an article that matches the set condition from an article group storing a plurality of articles consisting of text data,
  A content creation parameter setting step for setting a desired synthesized speech speed, speaker type, and total time length T;
  A keyword setting step for setting the keyword and the degree of interest in the keyword in the keyword set storage means;
  For each article i included in the article group, the degree of interest α (i) of the article i is determined based on the degree of interest of the keyword appearing in the article i among the keywords stored in the keyword set storage unit. Interest level determination step,
  Using a predetermined function f that increases in accordance with the degree of interest α (i) of the article i in the range from 0 to 1, the summary level A (i) of each article i is obtained by f (α (i)). , The product of the time length L (i) and the summary level A (i) of the article i when the text before the article i is summarized is synthesized in the descending order of the degree of interest α (i) of the article i. A summary level determination step of extracting an article i in a range where the sum total is less than the total time length T and assigning a summary level A (i) to the extracted article i;
  A text summarizing step for summarizing text corresponding to the article i based on the summary level A (i) for the article i given the summary level A (i);
  A text-to-speech synthesis step of synthesizing speech from the summarized text data based on the speech speed and speaker type set by the content creation parameter setting means;
  A speech synthesis method comprising: The speech synthesis method according to claim 1 or 2, further comprising a text classification step of classifying a plurality of articles into a plurality of predetermined genres,
A speech synthesizing method comprising a step of setting a desired synthesized speech speed and speaker type for each genre in the content creation parameter setting step. The speech synthesis method according to any one of claims 1 to 3 ,
An operation target keyword determination step for determining a keyword related to an article to be operated when listening to sound;
An interest level adjustment step for adjusting the degree of interest related to the determined keyword according to the operation parameters when operating the device;
A speech synthesis method comprising: In a speech synthesizer that selects an article that matches a set condition from an article group that stores a plurality of articles composed of text data, and generates synthesized speech from text data that constitutes the selected article.
Content creation parameter setting means for setting the desired synthesized speech speed, speaker type, total time length T, and number of required playback articles ;
A keyword setting step for setting the keyword and the degree of interest in the keyword in the keyword set storage means ;
For each article i included in the article groups, based on the degree of interest of keywords appearing in the articles i among the keywords stored in the keyword set storage means, DOI serial things i alpha a (i) Interest level determination means to determine;
The interest level weight w (α (i)) for each article i is determined so that the value increases in accordance with the degree of interest α (i) in the range from 0 to 1, and the text before the article i is summarized The sum of multiplication of the time length L (i) and the interest weight w (α (i)) of the article i exceeds the total time length T, and the number of extracted articles is the required number. The article i is extracted in descending order of the degree of interest α (i) of the article i so as to satisfy the number of reproduced articles, the time length L (i) for the extracted article i and the interest degree weight w (α ( From the sum S of multiplications of i)) and the total time length T, the summary level A (i) of each article i extracted is determined as the interest weight w (α ((i)) * (total time) of the article i. Summarization level determination means that determines the summation level A (i) to the extracted article i, determined by the length T / sum S) ,
Text summarizing means for summarizing the text corresponding to the article i based on the summary level A (i) for the article i given the summary level A (i) ;
Text-to-speech synthesis means for synthesizing speech from the summarized text based on the speech speed and speaker type set by the content creation parameter setting means;
A speech synthesizer comprising:   In a speech synthesizer that selects an article that matches a set condition from an article group that stores a plurality of articles composed of text data, and generates synthesized speech from text data that constitutes the selected article.
  Content creation parameter setting means for setting the desired synthesized speech speed, speaker type, and total time length T;
  Keyword setting means for setting the keyword and the degree of interest in the keyword in the keyword set storage means;
  For each article i included in the article group, the degree of interest α (i) of the article i is determined based on the degree of interest of the keyword appearing in the article i among the keywords stored in the keyword set storage unit. Interest level determination means to
  Using a predetermined function f that increases in accordance with the degree of interest α (i) of the article i in the range from 0 to 1, the summary level A (i) of each article i is obtained by f (α (i)). , The product of the time length L (i) and the summary level A (i) of the article i when the text before the article i is summarized is synthesized in the descending order of the degree of interest α (i) of the article i. Summarization level determination means for extracting an article i in a range where the total sum is less than the total time length T and assigning the summarization level A (i) to the extracted article i;
  Text summarizing means for summarizing the text corresponding to the article i based on the summary level A (i) for the article i given the summary level A (i);
  Text-to-speech synthesis means for synthesizing speech from the summarized text data based on the speech speed and speaker type set by the content creation parameter setting means;
  A speech synthesizer comprising: The speech synthesizer according to claim 5 or 6, comprising text classification means for classifying a plurality of articles into a plurality of predetermined genres,
A speech synthesizer characterized in that, in content creation parameter setting means, a desired synthesized speech speed and speaker type are set for each genre. The speech synthesizer according to any one of claims 5 to 7 ,
An operation target keyword determining means for determining a keyword related to an article to be operated when listening to the sound of operating the device;
Interest level adjustment means for adjusting the degree of interest related to the determined keyword according to the operation parameter when operating the device,
A speech synthesizer comprising: Computer is described by a readable programming language, the speech synthesis program to function as the speech synthesis device according to claim 5 or 8, wherein the computer. A recording medium comprising a computer-readable recording medium, wherein the speech synthesis program according to claim 9 is recorded on the recording medium.

Images