JPH08248971A - Text reading aloud and reading device - Google Patents

Abstract

PURPOSE: To provide a text reading-out device which reads out the dialog parts appearing in text having no additive information with different sound quality with each speaker by estimating the speakers of these dialog parts. CONSTITUTION: This device is inputted with the text from a text data input device 206. A compsn. analysis program 101, speaker analysis program 102 and speaker data allotment program 103 stored in a ROM 205 for storing the programs are executed under control of a control program 120 by a CPU 201. The data base stored in a non-volatile memory 203 is referenced in this CPU. To which speaker the respective dialogs of the dialog table 211 and speaker table 212 in the memory 202 correspond and which speaker of the speaker data 113 stored in a speaker data storage device 204 is to be used for each speaker are set. A voice rule synthesis program 104 references the dialog table 211 and the speaker table 212 and forms the synthesized voices while choosing the adequate speaker data for the text. The synthesized voices are outputted from a D/A device 144.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a text-to-speech device for reading out text information, and particularly to a text including dialogue such as a novel and a text structure having a layout.
The present invention relates to a text-to-speech device that reads aloud while changing the sound quality.

[0002]

2. Description of the Related Art In recent years, digitized text has become widespread due to widespread use of word processors and electronic publishing. As one of means for accessing these texts, a reading device that generates and outputs a speech waveform from the texts has been proposed.

In a conventionally proposed reading device, a quoted sentence portion is identified from a symbol in a text, and the quoted sentence portion can be identified by changing its properties from the other portions ( Japanese Unexamined Patent Publication No. Hei 5-28987) or an example of embedding a voicing control code in a text and performing regular speech synthesis according to the control code (Japanese Unexamined Patent Publication No. 5-30787).
396).

[0004]

When a conventional text-to-speech device is used, it is completely impossible to change the voice quality of the dialogue portion of a text including dialogue such as a novel by changing it for each speaker of the dialogue. Or, it is necessary to embed additional information in the text in advance.

An object of the present invention is to provide a text-to-speech device which estimates a speaker of a dialogue portion appearing in a text and reads aloud with a different voice quality for each speaker.

[0006]

According to the present invention, a text data input device for reading text data, and a symbol appearing in the text are used to separate the text into dialogue and ground sentences.
Parsing means for detecting the subject and verb of the sentence in the text, speaker analyzing means for estimating the speaker of the dialogue in the text from the result of the parsing, and speaker data that is the speaker data actually estimated from the text. Speaker data allocating means for deciding whether or not to read aloud, a voice rule synthesizing means for generating a rule synthesized voice waveform from text data and speaker data, a D / A conversion device for outputting the voice waveform as a sound, and overall control The present invention is to provide a text-to-speech device having a control means for performing the above and reading out different contents of different speakers in the text with different synthetic voices.

[0007] In addition, a classifying means for classifying a sentence in the ground of a text into a title, a comment, and a normal sentence from line breaks and parenthesis marks in the text, and a voice quality, a speech speed, a voice pitch, and an intonation pattern for each classification. It has a voice rule synthesizing means that can synthesize synthesized voices by changing it, and changes any or all of voice quality, speech speed, pitch, and intonation pattern for each title, annotation, and ordinary sentence classification, and from the text. (EN) A text-to-speech device that reads a synthetic voice aloud and presents the structure of the text in an easy-to-understand manner by the difference in utterance style.

[0008]

[Function] The text data input means takes in text data, the syntax analysis means divides the text into dialogue and ground sentences, and further extracts the subject and verb of each sentence,
The speaker estimating means identifies the speaker of each dialogue from the specific words appearing in the dialogue and the subject and verb information of the sentence before and after the dialogue, and the speaker appearing in the text in the speaker data allocating means. It is decided by what kind of voice should be read aloud, the text is converted into a voice synthesis waveform by the voice rule synthesis means, and the synthesized voice is outputted by the D / A converter, so that the texts of different speakers in the text are Can be read aloud with synthetic voices with different voice qualities, and the conversation structure contained in the text can be presented in an easily understandable manner.

By the classifying means for classifying the ground sentence of the text into the title, the annotation, and the ordinary sentence, the text is divided into the title,
Annotations, normal sentences are classified, and voice quality, speech speed, and voice pitch are classified for each sentence by a voice rule synthesizing unit that can synthesize synthesized speech by changing voice quality, speech speed, pitch, and intonation pattern. By changing any or all of the intonation patterns, the synthesized voice is read aloud from the text, and the structure of the text can be presented in an easy-to-understand manner by the difference in utterance style.

[0010]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 shows an embodiment of a hardware configuration diagram of the present invention. In this embodiment, a program storage ROM (205) in which various programs are stored, and a CPU which executes the programs stored in the program storage ROM (205)
(201), a memory for storing various data (20
2), a non-volatile memory (203) that stores a database necessary for speaker analysis, a speaker data storage device (204) that stores speaker data used in rule synthesis, and a text data input device that inputs text data from the outside. (20
6), a user operation input device (207) for inputting an operation from a user, and a D / A conversion device (114) for outputting a reading voice generated from text data.

The overall processing flow of the actual reading processing is shown in FIG. The text data (100) input from the text data input device (206) is subjected to a parsing process (101) with reference to a line feed symbol, a pair of brackets and parentheses, and the text is converted into a ground sentence (181) and a dialogue (182). ), A chapter (180), and an annotation sentence (183), linguistic analysis is performed, the text is decomposed into words, and part-of-speech information and syntactic information such as a subject and a predicate in the sentence are added. The speaker analysis process (102) estimates the speaker of the dialogue and the attribute of the speaker from the relationship between the subject of the dialogue (182) and the sentence of the ground (181) and the arrangement of the dialogue (182). FIG. 3 shows an example of text handled in the present invention. FIG. 4 shows an example of the result of processing the text of FIG. The speaker data allocation processing (103) in FIG. 2 selects speaker data that matches the speaker attributes and sets the speaker data index (1008). The voice rule synthesis processing unit (104) refers to the text with speaker information (191) and the speaker attribute information (192) in FIG. Read out by the speaker.

Next, the details of the processing flow of FIG. 2 will be described. The syntactic analysis (101), for example, performs morphological analysis by the minimum clause number method, divides the text into words,
Part-of-speech is analyzed, case-particle is analyzed as a key, and subject is analyzed.

The speaker analysis process (102) uses the subject of the verb relating to the utterance appearing in the sentence before and after the dialogue, the first-person pronoun appearing in the dialogue, the difference in the sentence end expression of the dialogue, and the dialect appearing in the dialogue. A speaker table (211) showing the characteristics of the speaker by estimating the speaker's name and the age and gender from the specific word appearing during the utterance, and distinguishing the speaker for each dialog. , A speaker table (212) showing characteristics such as sex and age of each speaker is generated. FIG. 5 is an example of the utterance verb verb database (106) used in the process of searching for a verb relating to utterance. In this embodiment, the verb ending type for utterances and idiomatic expressions for utterances are stored. FIG. 6 shows an example of the personal pronoun database (107). In this embodiment,
A word column (401) containing the notation of personal pronouns, a gender attribute column (402) that stores the gender when the personal pronoun gender is estimated, and a personal attribute column (403) that indicates how many personal pronouns are , Age attribute column (404) for storing the age group when the age group can be estimated from the personal pronoun, and dialect attribute storage for storing the dialect attribute when the dialect attribute can be estimated from the personal pronoun It consists of a column (107). Figure 7
Is an example of the personal name attribute database (108).
In many personal names, the gender can be determined from the personal name. Therefore, the personal name attribute database (108) includes a word column (401) for storing a name and a gender attribute column (402) for storing a gender attribute estimated from the personal name. FIG. 8 is an example of the sentence end expression attribute database (109). The sentence end expression attribute database (109) is a sentence end expression that appears only in a specific dialect, a sentence end expression that is used only by a speaker of a specific gender, or a sentence end expression that is used only by a speaker of a specific occupation or background. The end-of-sentence expression considered to be effective for identifying is stored. In this embodiment, the gender attribute column (402) for storing the gender of the speaker predicted by the sentence end expression and the dialect attribute column (for storing the dialect attribute of the speaker whose speaker dialect can be specified by the sentence end expression). 405). FIG. 9 shows an example of the dialect database (110). The dialect database (110) includes a word column (401) for storing words appearing in a specific dialect and a dialect attribute column (405) for storing dialect attributes of the words. FIG. 10 shows an example of the age attribute database (111). In this embodiment, a word column (401) for storing a word used by a speaker of a specific age or age and an age attribute column (406) for storing an age estimated from the word. An example of the speaker table (212) is shown in FIG. Also,
An example of the dialogue table (211) is shown in FIG. The details of the speaker analysis process (102) are shown in FIG. In the speaker analysis process (102), first, all the dialogues of the text are registered in the dialogue table (212) (1201), then the speaker is extracted from the registered dialogues and registered in the speaker table (1203). Next, a process of estimating which speaker registered in the dialog table is the utterance of the speaker registered in the speaker table by the attribute appearing in the dialog (1
203), with respect to the dialogue in which the speaker cannot be determined only by the attribute of the dialogue, the fact that adjacent dialogues often have different speakers is used to perform speaker estimation based on the arrangement of dialogues (1204). Finally, the speaker whose dialogue cannot be estimated automatically is manually set (1205).

Next, details of the processing flow of the embodiment of the speaker analysis processing will be described. First, the dialog table registration processing (1201) will be described. In the dialog table registration processing (1201), the dialog table (21
Register all the words that appear in the text in 1). In the dialogue table registration processing (1201), when the speaker is clearly indicated in the sentence (181) before and after the dialogue, the speaker name (1
002) is registered in the dialogue table. In the present invention, the subject of the speech verb of the ground sentence is treated as the speaker. In the determination of the utterance verb, the verb and the idiomatic expression registered in the utterance verb database of FIG. 5 are determined as the utterance verb. FIG. 15 shows the dialogue and the pattern of the ground sentence that can determine the speaker. There are three types of patterns in which the sentence of the ground including the utterance verb appears before and after the dialogue,
Pattern 1 (1351) in which the sentence of the earth appears before the dialogue, Pattern 2 (1352) in which the sentence of the earth appears after the dialogue, Pattern 3 (1353) in which the dialogue appears as the object of the sentence of the earth
There is. Also, as shown in FIG. 16, the utterance verb (1362)
When the dialogue including the word is sandwiched between two dialogues, if the speaker of dialogue 1 (1371) located before the sentence of the ground is not registered in the dialogue table (211), the subject of the ground sentence ( 1361) is treated as a speaker of dialogue 1 (1371),
When the speaker of the line 1 (1371) is already registered, the subject (1361) of the ground sentence is registered as the line 2 (1372) speaker. If the ground sentence containing the utterance verb does not have a subject, the subject of the ground sentence existing before the ground sentence containing the utterance verb is treated as the speaker, as the subject of the utterance verb. Fig. 1 shows an example of the case where the sentence in the ground containing the utterance verb has no subject.
7 shows. In the example of FIG. 17, the sentence 2 of the ground (1382) has no subject in the utterance verb (1362), but the sentence 1 of the ground before the sentence 2 of the ground has the subject of “Miwako” (1361).
Therefore, the subject of the ground sentence 2 (1382) is "Miwako". The processing procedure of this speech table registration processing will be described with reference to FIG. First, the internal variable "speaker variable"
And "subject variable" are cleared (1301). Each sentence is read, and it is determined whether the sentence is a dialogue (130
2) In the case of dialogue, the dialogue attribute registration processing (1308) is called to register the dialogue attribute in the table. Next, it is checked whether or not a value is set in the "speaker variable". If this value is set, the value of the "speaker variable" is set in the speaker column (1002) of the dialogue table (211). Is set and the "speaker variable" is cleared (1312). This process is shown in FIG.
This corresponds to the registration of Pattern 1 (1351) of 5. Also, 1
When the sentence read in 302 is not a dialogue but a sentence of the ground, the subject of the sentence is taken out (1303), and when the subject is present, a value is assigned to the "subject variable". Next, it is checked whether the utterance verb is included in the sentence (1306), and if it is included, first, as a process of checking whether it is pattern 3 (1353) in FIG. 15, it is checked whether the sentence includes a dialogue ( 1307). If included, the dialogue attribute registration process (130
8) is called, the attribute of the dialogue is registered, and the value of the “subject variable” is set as the speaker (1002) of the dialogue being processed.
If the sentence does not include dialogue, pattern 2 (13
As a process of inspecting whether or not it is 52), it is inspected whether there is a dialogue whose speaker is undecided immediately before the sentence (1310), and a "subject variable" is set as the speaker (1002) of the dialogue just before. If there is no dialogue whose speaker is undecided immediately before the sentence,
Since it may be the pattern 1 (1351), the value of the “subject variable” is set in the “speaker variable”. Finally, it is checked whether all the sentences have been processed (1314). If there are still sentences to be processed, the process returns to 1302.

By the way, not all lines are explicitly written as shown in FIG. Therefore, we extract the features that appear in the dialogue and calculate the matching between the dialogue in which the speaker of the dialogue is not explicitly written and the dialogue in which the speaker is written explicitly, and the dialogue in which the features match is the same. By presuming that the speaker has uttered, we presume a speaker whose dialogue is not written explicitly. In one embodiment of the present invention, as a feature of a dialogue, a specific word appearing in the dialogue or attribute information estimated from the particular word is used as the feature. Specifically, first-person expression (1003), dialect attribute (1004), gender attribute (1005), sentence end expression (1
006), and the age attribute (1007) is used. FIG. 18 shows an example of the actual speaker attribute registration processing. First, the first person column (1003) is set (1401). The first person expression (1003) is the personal pronoun database (107).
Use to search from the dialogue. An example of the personal pronoun database (107) is shown in FIG. A word (401) whose personal name (403) is 1 in the personal pronoun database is searched from the dialogue text and registered when found.
Next, the sentence end expression attribute column (1006) is set (14
02). The sentence end expression (1006) is searched for whether the word (401) registered in the sentence end expression database of FIG. 8 appears in the dialogue text, and registered if found. Next, the dialect column (1004) is set (1403). As for the dialect attribute, when the word (401) registered in the personal pronoun database (107), the sentence end expression database (109) in FIG. 8 and the dialect database (110) in FIG. Retrieves the dialect attribute (109) corresponding to the word and uses the dialogue table (21
It is registered in the dialect attribute column (1) of 1). As shown in FIG. 19, the dialogue may be continuous without a ground sentence between the dialogues. In most cases, when the dialogue is continuous without scissors in the ground sentence, the speakers of the two dialogues appearing consecutively are different. In the present invention, this feature is used for speaker estimation. Therefore, when the dialogue is registered in the dialogue table, information as to whether the preceding sentence is a dialogue or a sentence of the ground is recorded as a dialogue connection flag (1102). First, it is determined whether the sentence immediately before the current dialogue is dialogue (1404),
If it is a dialogue, the dialogue connection flag (1102) is set to 1 (1406), and if the immediately preceding sentence is not a dialogue,
The dialogue connection flag (1102) is set to 0 (140)
5). Next, the gender column (1005) is set (140
7). As for the gender attribute, if the speaker name is registered in the speaker name column (1002), the person name attribute database (1
09), the personal pronoun database (107) of FIG. 6 is searched by the speaker name, and the gender attribute (402) is extracted and registered. Next, set the age column (1007) (14
08). The age attribute (1007) is searched for whether the word (401) registered in the personal pronoun database (107) of FIG. 6 and the age identification word database (111) of FIG. 10 appears in the dialogue text. There are words that appear in the dialogue in the database, and the age attribute (406)
If exists, the age attribute is retrieved from the database and registered.

Next, the speaker table registration process (120)
Details of the embodiment 2) will be described with reference to FIG. The outline of the speaker table registration processing is that the person whose speaker name (1002) is the same as the speaker in the dialogue table, or the speaker of the first person personal pronoun is regarded as the same speaker, and the list of speakers is the speaker. Generate in table (212). First, it is checked whether a personal name or a first-person pronoun is set in the speaker name column (1002) of the dialogue table (1501). If the personal name or the first-person pronoun is registered, the corresponding speaker is registered in the speaker table. Inspected (150
2) If the speaker is already registered, a value is set for the speaker attribute in the dialogue table, but is there an attribute for which a value is not set for the speaker attribute in the speaker table? If the speaker attribute in the dialogue table is set, but the speaker attribute in the speaker table does not have a value, the attribute value in the dialogue table is changed to the speaker table attribute. Substitute in the value. If the corresponding speaker is not registered in the speaker table in the inspection of 1502, the speaker is registered in the speaker table. Then, at the end of the processing, it is checked whether all the speech tables have been processed (1505). If there is any unprocessed speech, the processing is repeated from 1501.

Details of the speaker estimation processing (1203) based on the dialogue attribute will be described with reference to FIG. In the speaker estimation process (1203) based on the dialogue attributes, the dialogue speakers are estimated in the dialogue index (1101) order of the dialogue table (211) shown in FIG. 12, and the estimated speakers are the speaker table. It is represented by the value of the speaker index (1001) of (212) and is set in the speaker index column (1103) of the speech table (211). An example of specific processing will be described below. First, when the speaker name is set in the speaker name column (1002) of the dialogue table (212) and the speaker name is registered in the speaker table (1601), and it is registered Sets the speaker index number (1001) of the corresponding speaker of the speaker table (212) in the speaker index column (1103) of the speech table (211). If no speaker name is set in the speaker name column (1002) of the dialogue table,
First person (1103), dialect (1004), gender (100
A speaker having a matching attribute of 5) is searched from the speaker table (212) (1602). Here, the attribute matching check is performed in the first person (1103), dialect (1004), and gender (1
005) Of the attributes, only the attributes whose values are set in both the speaker table (212) and the speech table (212) are targeted. It is checked whether the number of speaker candidates having an attribute that matches the dialogue attribute is one (1604). If only one speaker matches, the speaker index (100
1) is set in the dialogue speaker index column (1103). Next, it is checked whether the processing has been completed for all the dialogues (1606), and the dialogue to be processed is indicated in the dialogue table (2).
If it remains in 12), the processing from 1601 is repeated.

Next, the outline of the speaker estimation processing (1204) based on the line of dialogue and the details of the processing flow will be described. First, the outline of the processing will be described with reference to FIG. If a dialogue is adjacent to another dialogue, the speakers of the two dialogues are different. Therefore, as shown in FIG. 23, when there is a speech adjacent to the speech currently being processed (1754), when a speaker of the speech that is adjacent to the speech is set, the speech from the speaker candidate of the speech currently being processed is adjacent. It is possible to narrow down the speaker candidates excluding the speakers who speak the dialogue. In the example of FIG. 23, the dialogue adjacent to the front (1753) and the dialogue adjacent to the back (1755)
Is removed from the speaker candidates of the dialogue (1754) being processed. When the speaker candidates are narrowed down to one, the one speaker candidate is registered as the speaker of the dialogue being processed. Here, the sentence (1752) of the place before the dialogue is changed from the sentence of the closest place existing before the dialogue (1755) being processed to
It is defined as a series of continuous texts (see FIG. 23).
In addition, the ground sentence (1756) after the dialogue is defined as a series of ground sentences (1756) that continuously follow the closest sentence existing after the dialogue (1755) being processed (FIG. 2).
3). When the number of speaker candidates cannot be narrowed down to one, the sentence before the dialogue (1752) and the sentence after the dialogue (175
The speaker candidates are narrowed down by utilizing the fact that the speaker of the dialogue appears as the subject in 6). Speaker candidates other than the subject of the ground sentence (1752) before the dialogue and the ground sentence (1756) after the dialogue are removed. When the number of speaker candidates is narrowed down to one, the one speaker remaining after being narrowed down is registered as the speaker of the dialogue being processed. Next, the processing flow of the speaker estimation processing (1204) based on the line of dialogue will be described in detail with reference to FIG. First, a variable i for storing the index number of the dialogue to be processed first is initialized to 1 (1701). Then, the speaker table (211) is searched for speaker candidates in which the first-person expression of the dialogue of the index number i, the dialect attribute, and the gender attribute match. The connection flag (1102) of the dialogue table is referred to, and it is checked whether or not it is adjacent to the previous dialogue (1703). If it is adjacent, the speaker candidate is set if the previous dialogue is set to the speaker. Remove the speaker of the adjacent speech before (1704). Similarly, by referring to the connection flag (1102) in the dialogue table, it is checked whether or not it is adjacent to the next dialogue (1705). If it is, the speech is spoken if the previous dialogue is set to the speaker. The speaker of the adjacent speech is removed from the candidate candidates (1706). It is checked whether the number of speaker candidates is one (1707). If the number of speaker candidates is one, the speaker index corresponding to the speaker candidate is the speaker index column (110) of the dialogue table (212).
It is registered in 3) (1710). When the number of speaker candidates is not one, the sentence before and after the dialogue (1752, 175)
Speaker candidates not included in the subject of 6) are excluded (170
8) Inspect whether the number of speaker candidates is one (170)
9). The speaker index in which the number of speaker candidates corresponds to the speaker candidate is registered in the speaker index column (1103) of the speech table (212) (1710). Then, the speaker index number being processed is incremented by 1 (1711), it is checked whether all the lines have been processed (1712), and if all the lines have not been processed, the processes from 1702 are repeated. After processing all the lines, it is checked whether there is a line for which a new speaker index is set (171).
3). If there is a new line for which a speaker has been set, it is possible to reduce one speaker candidate for the line that is adjacent to the line for which a new speaker has been set, so it may be possible to set a line for a speaker. Come out. Therefore, if there is a dialogue for which a speaker index is newly set, the processing from 1701 is repeated.

Next, the details of the processing flow of the manual dialogue speaker designation processing (1205) will be described with reference to FIG. First, a variable i for storing the index number of the speech to be processed first is initialized to 1 (1801). Next, it is checked whether a speaker is set in the speaker index column (1103) of the i-th dialogue (1802), and if not set, the text around the dialogue and a list of speakers are displayed (1803), The user is made to select the speaker corresponding to the dialogue, the speaker of the dialogue is determined, and registered in the speaker index column (1103) of the dialogue table (212) (1803).
Then, increase the speaker index number being processed by 1 (17
11) and check whether all the lines have been processed (171
2), 170 if all the lines have not been processed
The process from 2 is repeated.

Next, the speaker data allocation process (103) will be described with reference to FIG. In the speaker data allocation processing, when generating the synthesized sound of each speaker extracted from the text, it is determined which speaker data is used to generate the synthesized sound. FIG. 25 shows the speaker data attribute (11
This is an example of the contents of 2). Speaker data attribute (11
2), the dialect attribute (10) of each speaker in the speaker data (113) prepared for the voice rule synthesis (104).
04), gender attribute (1105), age attribute (1107)
Is stored. The attribute of each speaker in the speaker table is matched with the speaker data attribute (112), and the speaker data index (1008) of the speaker data having the closest attribute and not already assigned to another speaker. The speaker data index column (100
Set to 8).

In the voice rule synthesizing process (104), the speech table and (211), the speaker table (212), the speaker data (113) and the text are input and a voice waveform is generated. When reading the text of the dialogue, the speaker index (1103) corresponding to the dialogue read in the dialogue table (211) is acquired. Next, the speaker table (212) is referenced using the obtained speaker index number, and the speaker data index number (1008) corresponding to the speaker index (1001) is extracted. Determine the speaker data to be used when reading the dialogue. The determined speaker data is taken out from the speaker data storage means (113) and a voice waveform is generated. The generated voice waveform is
It is converted into voice by the D / A device (114).

[0022]

As described above, according to the present invention, it is possible to provide a text-to-speech device which estimates a speaker of a dialogue portion appearing in a text in a text to which additional information is not added and reads aloud with a different voice quality for each speaker. , Help understand the text. Further, according to the present invention, by reading aloud with a voice having a different sound quality for each character attribute of the text, it is possible to help the voice understand the character attribute.

[Brief description of drawings]

FIG. 1 is an example of a hardware configuration of the present invention.

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

FIG. 3 is an example of input text of the present invention.

FIG. 4 is an example of an analysis result of speaker analysis processing of the present invention.

FIG. 5 is an example of a speech verb database of the present invention.

FIG. 6 is an example of a personal pronoun database of the present invention.

FIG. 7 is an example of a personal name attribute database of the present invention.

FIG. 8 is an example of a sentence end expression attribute database of the present invention.

FIG. 9 is an example of a dialect database of the present invention.

FIG. 10 is an example of an age identification word database of the present invention.

FIG. 11 is an example of a speaker table of the present invention.

FIG. 12 is an example of a speech table of the present invention.

FIG. 13 is an example of a processing flow of speaker analysis processing of the present invention.

FIG. 14 is an example of a processing flow of a speech table setting process of the present invention.

FIG. 15 is an explanatory diagram of a speaker estimation method of the present invention.

FIG. 16 is an explanatory diagram of a speaker estimation method of the present invention.

FIG. 17 is an explanatory diagram of a subject estimation method of the present invention.

FIG. 18 is an example of the processing flow of the dialog attribute registration processing of the present invention.

FIG. 19 is an explanatory diagram of a speaker estimation method of the present invention.

FIG. 20 is an example of a processing flow of speaker table registration processing of the present invention.

FIG. 21 is an example of a processing flow of speaker estimation processing based on a dialogue attribute of the present invention.

FIG. 22 is an example of a processing flow of speaker estimation processing based on a line of dialogue according to the present invention.

FIG. 23 is an example of a processing flow of speaker estimation processing based on the line of speech according to the present invention.

FIG. 24 is an example of a processing flow of manual speech speaker specification processing of the present invention.

FIG. 25 is an example of the speaker data attribute of the present invention.

[Explanation of symbols]

100 ... Text data, 101 ... Morphological analysis processing, 1
02 ... Speaker analysis processing, 103 ... Speaker data allocation processing, 1
04 ... Voice rule synthesis processing, 106 ... Utterance verb DB, 107
... Personal pronoun DB, 108 ... Person name attribute DB, 109 ... Sentence expression attribute DB, 110 ... Dialogue DB, Age identification word DB, 112 ... Speaker data attribute, 113 ... Speaker data, 114 ... D / A device, 120 … Control program, 180… Chapter, 181…
Ground sentence, 182 ... Dialogue, 183 ... Annotation sentence, 191 ... Text with speaker information, 192 ... Speaker attribute information, 193 ... Speaker attribute, 201 ... CPU, 202 ... Memory, 203 ... Non-volatile memory, 204 ... Speak Person data storage device, 205 ... ROM for program storage, 206 ... Text data input device,
207 ... User operation input device, 211 ... Dialogue table,
212 ... speaker table, 213 ... work area, 401
... Word finding, 402 ... Gender attribute, 403 ... Person, 40
4 ... age attribute, 405 ... dialect attribute, 406 ... age attribute,
1001 ... Speaker index number, 1002 ... Speaker name column, 1003 ... First person expression column, 1004 ... Dialect column, 10
05 ... gender column, 1006 ... sentence expression column, 1007 ... age column, 1008 ... speaker data index, 1101 ... speech index number, 1102 ... speech connection flag, 11
03 ... speaker index number, 1201 ... dialogue table registration process, 1202 ... speaker table registration process, 1203
... Speaker estimation processing based on dialogue attributes, 1204 ... Speaker estimation processing based on dialogue arrangement, 1205 ... Manual dialogue speaker designation processing, 1301 ... Variable initialization processing, 1302 ... Speech portion detection Processing, 1303 ... Subject extraction processing, 1304 ... Subject extraction result determination processing, 1305 ... Detected subject saving processing, 1306 ... Utterance verb check processing, 1
307 ... In-sentence speech detection processing, 1308 ... Speech attribute registration processing, 1309 ... Speech table speaker column setting processing, 1310
... Speech detection processing, 1311 ... Speaker variable setting processing, 131
2 ... Speaker variable setting check, 1313 ... Dialog table speaker column setting processing, 1314 ... Processing end check, 1351
... pattern 1 example, 1352 ... pattern 2 example, 135
3 ... pattern 3 example, 1454 ... pattern 4 example, 13
71 ... Dialogue 1, 1372 ... Dialogue 2, 1401 ... Dialogue table first-person column setting processing, 1402 ... Dialogue table end expression column setting processing, 1403 ... Dialogue table dialect column setting processing, 1404 ... Dialogue consecutive appearance check processing, 1405 ...
Dialog table dialogue connection flag column setting processing 1, 1406 ...
Dialog table dialogue connection flag column setting processing 2, 1407 ...
Speech table dialogue gender column setting processing, 1408 ... Speech table age column setting processing, 1501 ... Effective speaker name setting check processing, 1502 ... Speaker table already registered check processing,
1503 ... Speaker table blank setting processing, 1504 ... Speaker setting processing for speaker table, 1505 ... Speaker registration processing completion check processing, 1601 ... Processing for searching speaker table by speaker name associated with dialogue , 1602 ... Speaker table search processing by matching characteristics associated with dialogue, 1603 ... Speaker index number setting processing for dialogue, 1604 ... Speaker candidate narrowing check processing, 1605 ... Speaker index number for dialogue Setting processing, 1606 ... Processing for checking whether all dialogues have been processed, 1701 ... Initialization processing for dialogue index numbers to be processed, 1702 ... Speaker candidate search processing,
1703 ... A process of checking whether the dialogue set with the speaker index is adjacent to the speech before the current dialogue, 170
4 ... Processing for removing a speaker having an adjacent line of speech from a candidate speaker of the current line of speech, 1705 ... Process of checking whether a line having a speaker index is adjacent after the current line of speech, 1706 ... Adjacent line of speech Processing for removing the speaker of the above from the current speaker candidates of the dialogue, 1707 ... Check processing of the number of speaker candidates, 1708 ... Processing for eliminating the speaker candidates other than the subject of the sentence before and after the dialogue, 1709 ... Speaker item Complement checking process,
1710 ... Process of registering speaker index in speech table, 1711 ... Process of updating index number of speech to be processed, 1712 ... Process end check process, 17
13 ... Processing for checking whether the setting of the speaker index has changed, 1751 ... Dialogue 1, 1752 ... Text before the dialogue, 1753 ... Speech adjacent to the front, 1754 ... Speech being processed, 1755 ... Adjacent to the back Dialogue, 1756 ... ground sentence after dialogue, 1757 ... dialogue 2, 1801 ... initialization processing of dialogue index number to be processed, 1802 ... processing for checking whether speaker index is set for dialogue to be processed, 1803 ... Processing for displaying text around a dialogue for manually selecting a speaker and displaying candidate speakers, 1804 ... Processing for accepting speaker designation input from the user, 1805 ... Processing end check processing.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Toshiyuki Arizuka 1-280 Higashi Koikekubo, Kokubunji, Tokyo Inside the Central Research Laboratory, Hitachi, Ltd.

Claims (3)

[Claims] 1. In a text-to-speech device, a plurality of speaker data storage means, a speaker estimation means for estimating a speaker in a text, and a rule-synthesized speech while switching speakers according to the estimated speaker information. A text-to-speech device having a regular voice synthesizing means for generating and outputting, and presenting the structure of text such as conversation in an easy-to-understand manner by distinguishing the text by a plurality of speaker-synthesized sounds. 2. A dialogue symbol in a text separates a dialogue of a text from a ground sentence, and further, a personal pronoun appearing in the dialogue, a suffix expression of the dialogue, a dialect appearing in the dialogue, a subject and a utterance of sentences before and after the dialogue. Speaker estimation method for estimating the speaker of each part of the text from one or more pieces of verb information related to. 3. The text-to-speech device according to claim 1, wherein a classifying unit that classifies the text into a title, a comment, and a normal sentence according to a line feed and a symbol appearing in the text, and a different person for each classification is used, or a speaking rate. , A voice rule synthesizing means capable of reading text with different voice qualities by changing the utterance style such as voice pitch and intonation pattern, or by changing both the speaker and the utterance style.
A text-to-speech device characterized by presenting the structure of text in an easy-to-understand manner.