JP2001282815A - Summary reading device - Google Patents

Abstract

(57) [Summary] [PROBLEMS] To provide a summary reading device that enables a user to hear necessary information faster in a text reading system such as a mail reading system or a sightseeing guide system. SOLUTION: In the text-to-speech apparatus having a text input means, a language analysis means, and a speech synthesis means, a summary-speech apparatus according to the present invention is a summary which is a ratio of a length of a synthesized speech read to a length of an input text. A summarization rate input means for inputting a rate, and a length of voice to be synthesized is set to a specified summarization rate compared to a length of the input text.
Partial text selecting means for selecting one or more partial texts from the input text. [Effect] For a text to which no additional information is added, a summary reading device that reads out only important portions by summarizing to a summarization rate specified by a user can be provided, so that the contents of the text can be quickly understood. Become like

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001] 1. Field of the Invention [0002] The present invention relates to a speech synthesis method, and more particularly to a summary speech synthesis method for selecting and reading out a partial text from input text. More specifically, the present invention relates to a summary speech synthesis method for selecting and reading out a partial text based on attribute information added to each partial text using a result of a language analysis process.

2. Description of the Related Art As an e-mail reading system, "Family Electronic Mail" of Flex Pharm Co., Ltd., which is a system that calls a center and reads out e-mails arriving at you, is a personal computer equipped with an e-mail reading function. Hitachi's "Persona" and others. These and other currently-available text-to-speech softwares read all input texts. Also, Japanese Patent Application Laid-Open No. 9-25137
As disclosed in Japanese Unexamined Patent Application Publication No. 3 (1994) -302, a technique for defining a format in which a tag indicating a reading part is added to an input text, or a part including a specific keyword, such as JP-A-5-224688 and JP-A-6-83381, is used. There are techniques to read out aloud by changing the type and speed of the voice.

For this reason, the user is forced to listen to all the contents read out, and a lot of time is spent until necessary messages appear, such as when confirming the contents of many e-mails. There were many. In addition, although there is a technology to read a part of a specific keyword or the like while changing the tone of the voice to call attention of the user, there is no system that automatically skips an unnecessary part. JP-A-9-25137
The use of the format of Japanese Patent Publication No. 3 can realize a text-to-speech system for skipping, but requires manual tagging of input text in advance according to the format. SUMMARY OF THE INVENTION The present invention has been made in view of the above problems, and has as its object to provide a summary reading device that allows a user to quickly hear only necessary information in a voice reading system such as a mail reading system.

To achieve the above object, the present invention provides a text-to-speech apparatus including a text input unit, a language analysis unit, and a speech synthesis unit. Summarization rate input means for inputting a relative ratio of the length of the synthesized speech read out to the readout time or the summarization rate which is the absolute value of the length of the synthesized speech read out (the number of characters or the readout time); Partial text selecting means for selecting one or a plurality of partial texts from the input text so that the length of the voice to be input has a specified summarization rate compared to the length of the input text. Features. According to the text-to-speech apparatus of the present invention, by referring to the result of linguistic analysis of input text and reading out only important phrases or partial texts determined based on information such as part of speech or character type, all conventional text-to-speech devices are used. Compared to a text-to-speech system that reads text, the time required for the user to grasp the outline of the text content can be reduced.

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a diagram for explaining a basic configuration of a text-to-speech system of the present invention. As shown in the figure, the basic configuration of the text-to-speech system of the present invention is such that the input text 10
Text input means 200 for inputting 0, linguistic analysis means 300 for performing linguistic analysis on the input text, summarization rate input means 500 for inputting the summarization rate 400, and read out from the input text based on the input summarization rate A partial text selecting means 600 for selecting a partial text to be processed;
And a voice synthesizing unit 700 for converting the selected partial text into a synthesized voice 800. Text input means 2
00 is a floppy disk or CD
An information storage medium such as a ROM and a network such as a telephone / wireless / satellite communication line can be used. The linguistic analysis means 300 has at least a morphological analysis process of dividing the input text into words and adding reading and accent information, and preferably has a function of further performing a syntax / semantic analysis process. This process is described in detail in "Natural Language Processing" (by Makoto Nagao, Iwanami Shoten). As the summarization rate input means 500, a numerical input device such as a switch, dial, or telephone push button, a character input device such as a keyboard, or a numeric voice recognition device can be used. The partial text selecting means 600 outputs the input text 10
After dividing and analyzing 0, a certain partial text group is selected from the input text 100 using the linguistic analysis result information given to each of the divided partial texts, and the selected partial text group is sent to the speech synthesis unit 700. Is output. As the speech synthesis means 700, when a reading text to which a prosodic symbol specifying an accent or a prosody is added is input, a conventional speech synthesis technique of performing a unit phoneme waveform selection process, a connection process of the waveform, a prosody process, and the like is used. be able to. A specific embodiment of the text-to-speech system of the present invention having the basic configuration described above will be described in detail below. FIG. 2 is a diagram for explaining an overall configuration diagram in the case where the present invention described in claim 1 is implemented as a mail summary reading system using a telephone line. As shown in the figure, the configuration of the embodiment of the present invention as a mail summary reading system using a telephone line is to present a voice to a user through a telephone line, or to respond to a push button or voice from the user. Telephone line connection means 900 having a function of transmitting the message to the system side, and interactive processing with the user through the telephone line voice connection means 900 to select the input mail text 110 that the user wants to read out. , A mail input unit 210 for inputting the input mail text 110 selected by the mail selection unit 120 to the summary reading device, and a language analysis unit 300 for performing a language analysis process on the input mail text. Dialogue processing with the user through the telephone line voice connection means 900, and the user The summarization rate input unit 510 that uses interactive processing to specify how much the summarization rate is desired to be read out, the input mail text analysis result subjected to language analysis processing by the language analysis unit 300, and the summation rate input unit 510 inputs With reference to the summarization rate, a partial text selecting means 600 for selecting one or a plurality of partial texts to be read out, and a partial text selected by the partial text selecting means 600 are converted into a synthesized voice, and the telephone line voice is converted. The voice synthesizing unit 700 reads out a mail to the user via the connection unit 900. Next, the processing steps in the mail summary reading system according to the present invention will be described. When there is a line connection from the user via the telephone line voice connection means 900, the telephone line voice connection means 9
00 performs user authentication processing and determines the user,
The mail selection unit 120 using the interactive processing is started.
The mail selecting unit 120 using the interactive processing notifies the user of the readable mail by reading out the number of mails currently received to the user and the respective mail titles, and then determines which mail is read out. To the user. As a method of specifying an e-mail from the user, there may be a method of specifying an e-mail number by a push button, or a method of recognizing a mail number by voice. Further, there may be a method of specifying a title or a sender by voice recognition. The mail selecting means 120 outputs information designating the mail that the user wants to read to the mail input means 210. The mail input unit 210 is a mail selection unit 1
Reference is made to the mail designation information inputted from the user 20, and the mail text 110 designated by the user is selected from the mails currently received for the user, and the language analysis means 30 is selected.
Output to 0. When the mail selection process is completed, the summarization rate input unit 510 using the interactive process is activated.
The summarization rate input unit 510 inputs the summarization rate by referring to the summarization rate already registered by the user, or by inquiring a new user. The user can input the summarization rate by a push button, or can recognize the summarization rate by voice. Here, the summarization rate to be specified by the user is specified as a relative value such as a ratio of the number of characters and words of the reading portion to the number of characters and words of the entire mail text or a ratio of a time required for reading by voice synthesis. It can also be specified as a value of within a number of characters, words, or sentences, or an absolute value of within a few seconds reading time by speech synthesis. The summarization rate input means 510 outputs the summarization rate specified in this way to the partial text selection means 600. The language analysis means 300 is provided by the mail input means 21
The language analysis processing is performed on the mail input from 0.
The specific processing method is described in detail in "Natural Language Processing" (by Makoto Nagao, Iwanami Shoten), etc., and will not be described here.
The linguistic analysis means 300 creates linguistic analysis information shown in FIG. 4 as an analysis result. The analysis result of FIG. 4 is merely an example, and a more detailed analysis of the dependency relationship may be performed. In FIG. 4, the word number 305 stores the order in which the word appears in the input mail text 110, and the word 3
10 stores each word character string when the input mail text 110 is divided into words by morphological analysis processing, reading 320 stores a reading character string corresponding to each word character string, and part of speech 330 Is stored, and the character type 340 stores the character type of each word character string. here,
The reading 320, the part of speech 330, and the character type 340 can be acquired using word dictionary information or the like. The linguistic analysis result analyzed by the linguistic analysis means 300 and shown in FIG. 4 is input to the partial text selecting means 600 together with the digest rate output from the digest rate input means 510. FIG. 3 shows an internal configuration diagram of the partial text selection means 600. The partial text selecting unit 600 receives the linguistic analysis result analyzed by the linguistic analyzing unit 300 shown in FIG. 4 and determines the importance corresponding to each of the words 310. Word-importance dictionary 660 in which importance is defined by the user, and importance order sorting means 620 that sorts the linguistic analysis results to which the importance is assigned in order of importance.
And a summary amount determining means 630 for determining the range of words to be read out based on the summary rate input from the summary rate input means 510, and a sentence for sorting the words 310 determined to be read out in the order in which they appear in the input text. An analysis result storage device 650 which is a data storage unit for transferring an analysis result between the sequential sorting unit and each of these units.
And This construction method is one embodiment in which a partial text is selected based on the importance of each word stored in the dictionary, and the part of speech 330 in the language analysis result shown in FIG.
There is also an implementation method of giving importance to each part of speech or character type by referring to information such as the character type and the character type 340. Also, words and parts of speech,
There may be an implementation method for calculating more detailed importance using mathematical formulas from information on character types. Next, the processing steps in the partial text selection means 600 will be described with reference to FIG. The linguistic analysis result analyzed by the linguistic analysis means 300 and shown in FIG. 4 is temporarily stored in the analysis result storage device 650. Next, the importance assigning means 610 is activated. The importance assigning means 610 refers to the word-importance dictionary 660 storing the word-importance information shown in FIG. 5 and, for each of the words 310 in the linguistic analysis result shown in FIG.
The importance of the word is obtained, and the importance analysis information shown in FIG. 6 is stored in the analysis result storage unit 650. Where the word-
The contents of the importance dictionary 660 differ depending on what words are regarded as important. For example, when the invention described in claim 2 is implemented, words having part of speech information to be regarded as important have a high importance. The word-importance dictionary is switched to the given word-importance dictionary, and in the case of implementing the invention according to claim 3, a word-importance dictionary in which words having character type information to be regarded as important are given high importance Will be switched to. Of course, in addition to the parts of speech and character types, it is also possible to use a dictionary that gives the user its own importance in consideration of the frequency of use of each word, etc. From these various importances,
The importance may be comprehensively calculated and used based on a certain mathematical expression. Word-importance dictionary 66 thus determined
0 is a word that each user of the service considers to be important to each individual because it is necessary for the provider of the mail reading service to determine in advance. , You need to specify a general importance. That is, the word "discussion"
The designation must be such that the importance of the word "conference" is higher than that of the word "conference", or the importance of the part of speech "proper noun" is higher than that of the part of speech "ordinary noun". As a method of constructing such a word-importance dictionary 660 specifying a general importance, for example, a document frequency or a document frequency from a Japanese corpus collecting a large amount of Japanese sentences is used.
What is necessary is just to use the importance calculation method for words using an inverted document frequency or the like. The details of this are described in the above-mentioned document “Natural Language Processing” (edited by Makoto Nagao). Furthermore, words that each user of the service thinks-
It is also possible to separately prepare a word-importance dictionary in which importance information is collected and use it together. In this case, for example, a word, part-of-speech, and character type for which the user wants to assign importance to the user by using input means such as word, part-of-speech, and character type using word voice recognition and the like, and importance value input means using a push button or the like And its importance, and record it as a word-importance user dictionary separately from the word-importance dictionary 660 relating to general languages. The partial text selecting means 600 first refers to the word-importance user dictionary, and if the importance of the desired word / part of speech / character type is specified, uses it. If not, the general word-importance is used. Dictionary 6
This can be realized by utilizing the importance of 60.
Next, the importance order sorting unit 620 is activated. The importance sorting means 620 extracts the importance analysis information shown in FIG. 6 from the analysis result storage device 650 and sorts the information in the descending order of the importance 613. By this process, the importance analysis information shown in FIG. 6 is sorted as shown in FIG. The sorted importance analysis information is stored in the analysis result storage device 650. Next, the summary amount determination means 630 is activated. The summary amount determination unit 630 extracts the sorted importance analysis information shown in FIG. 7 from the analysis result storage unit 650, and adds the number of characters of the word 612 in descending order of importance. The word whose position is the closest to the summarization rate input to the summarization amount determining means 630 is extracted from the sum of the number of characters of the input mail text 110 and stored in the analysis result storage device 650. For example, a summary rate of 50%
Is input, after extracting the sorted importance analysis information shown in FIG. 7 from the analysis result storage device 650,
The number of characters in the word is added in order from the word having the highest importance, and the ratio of the input mail text 110 to the number of characters (24 in the example shown in FIG. 4) is obtained as shown in FIG.
Then, the word "line" which is closest to the input summary rate of 50% is selected, and a portion having a higher importance than that is extracted as shown in FIG.
Is stored in When the relative ratio in the reading time by speech synthesis is specified as the summarization rate, the words are read in descending order of importance with reference to the correspondence table between the number of characters of the word and the reading time shown in FIG. The reading time is added, and a word at a position where the ratio to the time required to read the entire input mail text 110 is closest to the designated summary rate is extracted, and the analysis result storage device 6
50. If an absolute value of the reading time within seconds is specified as the summarization rate, referring to the correspondence table of the number of characters of words and the reading time shown in FIG. The time is added, and words up to the position closest to the designated reading time are extracted and stored in the analysis result storage device 650. Here, the calculation of the time required for reading out the selected word / phrase by speech synthesis was performed simply with reference to the correspondence table of the number of characters and the reading time, but more precisely, for each word / phrase, It is better to calculate the length of the voice data generated after actually performing the voice synthesis processing. This is because the reading time differs depending on the phoneme included in the word even if the number of characters is the same, or depending on the reading speed specified by the user. However, if speech synthesis processing is performed on each word / phrase in this way, considerable processing time may be required. As a slightly simplified method, the reading time information when the word is synthesized in the word-importance dictionary is used. May be stored together, and the value thereof may be referred to. Next, the text order sorting unit 640 is activated. The sentence order sorting unit 640 extracts the importance analysis information output by the summary amount determination unit shown in FIG. 9 from the analysis result storage device 650, and sorts using the word number 611 as a key. As a result, the importance analysis information shown in FIG. 10 is output and passed to the speech synthesis unit 700.
Finally, the speech synthesis unit 700 is activated, and generates a synthesized speech in which the words 611 are read out in order based on the importance analysis information shown in FIG. At this time, the word may be read aloud as it is, or in the case of a word that does not make sense if read aloud as it is, such as a part of a verb such as the word "line" in this example, a language analysis means It is also possible to use the word dictionary referred to by 300 to read back after returning to the original form of the word. Further, referring to the linguistic analysis result of the input mail shown in FIG. 4, a certain word in the importance analysis information shown in FIG. 10, for example, "Next week" is read out by speech synthesis, and then the word of the word "Next week" read out now is read out. From number 3,
A search is made in the linguistic analysis result of FIG. 4 up to word number 5 of the next reading word “meeting”, and if there is a period “,”, a silent section for a certain period of time is set, and the reading point “.”
If there is, processing such as placing a silent section longer than a period can be taken. By performing such processing, the advantages of the present invention alone can be obtained as compared with the combination of the sentence summarization technique for extracting words based on words, parts of speech, and character types and speech synthesis processing. Finally, the voice synthesis means 700
The synthesized speech generated by the telephone line speech connection means 900
Is transmitted to the user via. As described above, in the e-mail summary reading system according to the present invention, e-mail can be summarized and read out so that the summarization rate specified by the user each time or registered by the user is the summarization rate.

According to the present invention, it is possible to provide a summary reading device that summarizes text to which no additional information is added so as to have a summarization rate designated by a user and reads only important portions. , Will be able to understand the contents of the text quickly.

[Brief description of the drawings]

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

FIG. 2 is an embodiment of a processing flow of the present invention.

FIG. 3 is an embodiment of a partial text selecting means of the present invention.

FIG. 4 is an example of a language processing result of the present invention.

FIG. 5 is an embodiment of word-importance information of the present invention.

FIG. 6 is an embodiment of importance analysis according to the present invention.

FIG. 7 is an embodiment of a sorted importance analysis of the present invention.

FIG. 8 is an embodiment of the sorted importance analysis of the present invention.

FIG. 9 is an embodiment of the importance analysis extracted by the summary amount determination means of the present invention.

FIG. 10 is an embodiment of importance analysis sorted in the order of word numbers according to the present invention.

FIG. 11 is an embodiment of a correspondence table between the number of word characters and the reading time according to the present invention.

[Explanation of symbols]

100 ... input text, 110 ... input mail text,
120 ... Email selection means using interactive processing, 200 ...
Text input means, 210 ... Email input means, 300 ...
Language analysis means, 305: word number, 310: word, 32
0: reading, 330: part of speech, 340: character type, 400: summary rate, 500: summary rate input means, 510: summary rate input means using interactive processing, 600: partial text selection means, 610: importance assigning means, 611: word number, 61
2 ... word, 613 ... importance, 620 ... importance order sorter, 630 ... summarization amount determiner, 640 ... text order sorter, 650 ... analysis result storage device, 660 ... word-importance dictionary, 661 ... word, 662: importance, 700: voice synthesis means, 800: synthesized voice, 900: telephone line voice connection means.

Claims (3)

[Claims] 1. A reading apparatus for generating a synthesized speech corresponding to an input text and performing speech synthesis, wherein a relative ratio of a length of the synthesized speech for reading the text to a length of the text, or a synthesized speech to be read. A summarization rate input means for inputting a summarization rate which is an absolute value of the length of the input text; And a partial text selecting means for selecting one or a plurality of partial texts from a summary text. 2. The summary reading device according to claim 1, further comprising: a language analyzing unit that performs a language analysis of the input text; and a reading part-of-speech list that describes part-of-speech information to be read. As a result of linguistic analysis of the input text by the linguistic analysis means, referring to the part of speech information added to each partial text in the input text, only the partial text having the part of speech information included in the above part of speech list is read out. Summary reading aloud. 3. The summary reading device according to claim 1, further comprising: a language analysis unit for performing a language analysis of the input text; and a reading character type list in which character type information to be read is described. As a result of language analysis of the input text by the language analysis means, referring to the character type information added to each partial text in the input text,
A summary reading device, which reads out only a partial text having character type information included in the reading character type list.