JPH0962286A - Voice synthesizer and the method thereof - Google Patents

Abstract

PROBLEM TO BE SOLVED: To generate synthesized voice sound of a proper rhythm according to an input information or sentence. SOLUTION: An input sentence is analyzed by a sentense analyzing part 2, and the result is supplied to a rhythm information generation part 14. The part 14 has therein stored several kinds of pause setting rules for setting pause in synthesized sound. Initial information such as the field of input sentence, average length of one sentence (basic length of sentence), application of synthesized sound (to e.g. reading or composition of sentence), any one of the above rules is selected based on the initial information. Pause is set in the synthesized sound according to the selected rule.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech synthesizer and a speech synthesis method. In particular, by selecting one of a plurality of types of prosody information generation rules for generating prosody information of synthetic sound and generating a synthetic sound based on the prosody information generation rule, it is appropriate. The present invention relates to a voice synthesizing device and a voice synthesizing method capable of providing various synthetic sounds.

[0002]

2. Description of the Related Art In a conventional speech synthesizing apparatus for generating a synthetic voice corresponding to an input sentence (text sentence) such as a sentence containing kanji and kana, the input sentence is input by natural language processing. Phonological information and prosody information of a sentence are obtained, and a synthetic sound corresponding to an input sentence is generated based on the phonological information and prosody information.

In such a voice synthesizer, the synthesized voice is
To generate prosodic information in order to approximate human speech,
A predetermined prosody information generation rule is used. The prosody information generation rule is obtained (determined) based on, for example, the position of punctuation marks, or by statistically analyzing various sentences.

[0004]

By the way, in the conventional speech synthesizer, prosody information is generated for any input sentence based on a constant prosody information generation rule incorporated in the device. It was done like this. Therefore, prosody information that matches the characteristics of the input sentence may not be generated, and an appropriate synthesized speech may not be obtained.

That is, when the input sentence belongs to the field of weather forecast, for example, it is better for the user to emphasize a specific word such as fine, rainy, cloudy, or a place name, or a numerical value such as a probability of precipitation. Can accurately convey information to
Furthermore, if the sentence length of the input sentence is long, it is easier to hear if the utterance speed is faster and the pauses are spaced, and if the sentence length of the input sentence is shorter, the utterance speed is not so fast. Is easy to hear.

However, in the conventional speech synthesizer,
Prosody information is generated based on one prosody information generation rule, so information is accurately transmitted for every input sentence,
Moreover, it was difficult to provide a synthetic sound that was easy to hear.

Further, the conventional speech synthesizer does not always generate a synthesized voice having a prosody desired by the user. That is, when the user uses the speech synthesizer for reading a book, for example, it is desirable to provide a smooth synthesized voice with a high speech rate. When used for such a purpose, it is desired to provide a synthetic voice in which the entire sentence is clearly uttered and a specific word is emphasized. However, the conventional speech synthesizer does not always provide the synthesized speech desired by the user as described above.

The present invention has been made in view of such a situation, and it is possible to provide a user with an appropriate synthesized sound.

[0009]

According to a first aspect of the present invention, there is provided a voice synthesizing apparatus, which stores a plurality of types of prosody information generation rules for generating prosody information of synthetic voice.
Selection information input means for inputting selection information for selecting one of a plurality of types of prosody information generation rules stored in the rule storage means, analysis means for analyzing an input sentence, and selection information input Based on the selection information input from the means, any one of the plurality of prosody information generation rules stored in the rule storage means is selected, and the prosody information generation rule is used to analyze the analysis result of the analysis means. , Prosody information generating means for generating prosody information.

According to a seventh aspect of the present invention, there is provided a voice synthesizing device, which stores a plurality of types of prosody information generation rules for generating prosody information of synthesized voice, and a feature for detecting features of an input sentence. Detection means, analysis means for analyzing the input sentence,
Based on the characteristics of the input sentence detected by the characteristic detection means, one of a plurality of types of prosody information generation rules stored in the rule storage means is selected, and the prosody information generation rule is used to analyze means. And a prosody information generating means for generating prosody information from the analysis result of 1.

A speech synthesis method according to a thirteenth aspect analyzes an input sentence, selects one of a plurality of types of prosody information generation rules stored in the rule storage means, and selects the prosody information generation rule. It is characterized by generating prosody information from the analysis result of the input sentence and generating a synthetic sound based on the prosody information.

According to a fourteenth aspect of the present invention, there is provided a voice synthesizing device, wherein a parameter storage means for controlling a prosody of a synthesized voice stores a plurality of prosody parameters and a plurality of prosody parameters stored in the parameter storage means. A selection information input means for inputting selection information for selecting any one of the above, an analysis means for analyzing an input sentence, and a parameter storage means based on the selection information input from the selection information input means. Selection means for selecting any one of a plurality of stored prosody parameters, and synthesis sound generation for generating a synthesis sound of a prosody corresponding to the prosody parameter selected by the selection means based on the analysis result of the analysis means. And means.

According to a nineteenth aspect of the present invention, in the voice synthesizing method, when selection information for selecting one of a plurality of prosody parameters stored in the parameter storage means is input, based on the selection information, It is characterized in that any one of a plurality of prosody parameters stored in the parameter storage means is selected and a synthetic sound of a prosody corresponding to the prosody parameter is generated.

According to a twentieth aspect of the present invention, there is provided a speech synthesizing apparatus, which stores a plurality of prosody parameters for controlling the prosody of synthesized speech, a feature detecting means for detecting a feature of an input sentence, and a feature detecting means. Analyzing means for analyzing the input sentence; selecting means for selecting one of a plurality of prosody parameters stored in the parameter storing means based on the characteristics of the input sentence detected by the characteristic detecting means; and analyzing means Based on the analysis result of (1), there is provided a synthetic sound generating means for generating a synthetic sound of a prosody corresponding to the prosody parameter selected by the selecting means.

A speech synthesis method according to a twenty-fourth aspect of the present invention detects a feature of an input sentence and selects one of a plurality of prosody parameters stored in the parameter storage means based on the feature of the input sentence. , Producing a synthetic sound of a prosody corresponding to the prosody parameter.

In the voice synthesizing apparatus according to the first aspect, the rule storage means stores a plurality of types of prosody information generation rules for generating prosody information of synthesized voice, and the selection information input means, Selection information for selecting any one of a plurality of types of prosody information generation rules stored in the rule storage means can be input. The analyzing means analyzes the input sentence, and the prosody information generating means, based on the selection information inputted from the selection information inputting means, selects one of a plurality of types of prosody information generating rules stored in the rule storing means. Is selected, and the prosody information generation rule is used to generate prosody information from the analysis result of the analysis means.

In the speech synthesizer according to the seventh aspect, the rule storage means stores a plurality of types of prosody information generation rules for generating prosody information of synthetic speech, and the feature detection means inputs It is designed to detect sentence characteristics. The analysis unit analyzes the input sentence, and the prosody information generation unit selects one of the plurality of types of prosody information generation rules stored in the rule storage unit based on the feature of the input sentence detected by the feature detection unit. Is selected and the prosody information generation rule is used to generate prosody information from the analysis result of the analysis means.

In the speech synthesis method according to the thirteenth aspect, the input sentence is analyzed, and one of a plurality of types of prosody information generation rules stored in the rule storage means is selected,
By using the prosody information generation rule, prosody information is generated from the analysis result of the input sentence, and a synthetic sound is generated based on the prosody information.

In the speech synthesizer according to the fourteenth aspect, the parameter storage means stores a plurality of prosody parameters for controlling the prosody of the synthetic voice, and the selection information input means stores in the parameter storage means. Selection information for selecting any one of the stored prosody parameters can be input. The analysis unit analyzes the input sentence, and the selection unit selects one of the plurality of prosody parameters stored in the parameter storage unit based on the selection information input from the selection information input unit. Has been done. The synthetic sound generation means is adapted to generate a synthetic sound of a prosody corresponding to the prosody parameter selected by the selection means, based on the analysis result of the analysis means.

In the speech synthesis method according to claim 19, when selection information for selecting one of the plurality of prosody parameters stored in the parameter storage means is input, based on the selection information. , Any one of a plurality of prosody parameters stored in the parameter storage means is selected, and a synthetic sound having a prosody corresponding to the selected prosody parameter is generated.

In the speech synthesizer according to the twentieth aspect, the parameter storage means stores a plurality of prosody parameters for controlling the prosody of the synthetic voice, and the feature detection means stores the features of the input sentence. It is designed to detect.
The analysis unit analyzes the input sentence, and the selection unit selects one of the plurality of prosody parameters stored in the parameter storage unit based on the feature of the input sentence detected by the feature detection unit. Has been done. The synthetic sound generation means is adapted to generate a synthetic sound of a prosody corresponding to the prosody parameter selected by the selection means, based on the analysis result of the analysis means.

In the speech synthesis method according to the twenty-fourth aspect, the feature of the input sentence is detected, and based on the feature of the input sentence,
Any one of a plurality of prosody parameters stored in the parameter storage means is selected, and a synthetic sound having a prosody corresponding to the prosody parameter is generated.

[0023]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiments of the present invention will be described below, but before that, in order to clarify the correspondence between each means of the invention described in the claims and the following embodiments. The features of the present invention are described as follows by adding a corresponding embodiment (however, an example) in parentheses after each means.

That is, the speech synthesizer according to claim 1 is
A rule synthesizing device for generating a synthetic voice corresponding to an input sentence, which stores a plurality of types of prosody information generation rules for generating prosodic information of the synthetic voice (for example,
(Pose setting rule dictionary 38A, etc. shown in FIG. 11) and selection information input means for inputting selection information for selecting any one of a plurality of types of prosody information generation rules stored in the rule storage means. For example, based on the initial information setting unit 1 shown in FIG. 1), analysis means for analyzing an input sentence (for example, language analysis unit 2 shown in FIG. 1), and selection information input from the selection information input means. , A prosody information generation means for selecting any one of a plurality of types of prosody information generation rules stored in the rule storage means, and using the prosody information generation rule to generate prosody information from the analysis result of the analysis means. (For example, the pose setting unit 38 shown in FIG. 11)
And a synthetic sound generation unit that generates a synthetic sound based on the prosody information generated by the prosody information generation unit (for example, FIG. 1).
Sound processing unit 15 shown in FIG.

The speech synthesizer according to a fourth aspect of the present invention is an emphasized word determining means (for example, the emphasized word shown in FIG. 1) that determines a word to be emphasized based on selection information which is information about a field to which an input sentence belongs. Further includes a selection unit 11),
It is characterized in that the prosody information generation means generates prosody information so that the word determined by the emphasis word determination means is emphasized.

A speech synthesizer according to a seventh aspect is a speech synthesizer for generating synthetic speech corresponding to an input sentence, and has a plurality of types of prosody information generation rules for generating prosody information of synthetic speech. The stored rule storage means (for example, FIG. 11).
The pose setting rule dictionary 38A shown in FIG. 2), feature detecting means for detecting the features of the input sentence (for example, the field determining unit 6 or the reference sentence length determining unit 9 shown in FIG. 1), and analyzing means for analyzing the input sentence. (For example, the language analysis unit 2 shown in FIG. 1)
Based on the characteristics of the input sentence detected by the characteristic detection means, any one of a plurality of types of prosody information generation rules stored in the rule storage means is selected, and using the prosody information generation rules, Based on the analysis result of the analysis means, a synthetic sound is generated based on the prosody information generation means (for example, the pose setting unit 38 shown in FIG. 11) which generates prosody information and the prosody information generated by the prosody information generation means. It is characterized by comprising a synthetic sound generation means (for example, the sound processing unit 15 shown in FIG. 1).

A speech synthesis apparatus according to a ninth aspect of the present invention is an emphasized word determining means (for example, an emphasized word shown in FIG. 11) that determines a word to be emphasized based on the field to which the input sentence detected by the feature detecting means belongs. Further, the prosody information generating means is characterized in that the prosody information generating means generates prosody information so that the word determined by the emphasis word determining means is emphasized.

A speech synthesizer according to a tenth aspect of the invention further comprises keyword detecting means (for example, the keyword extracting section 3 shown in FIG. 1) for detecting a predetermined keyword from the input sentence, and the feature detecting means is a keyword. It is characterized in that the field to which the input sentence belongs is detected based on the keyword detected by the detecting means.

A speech synthesis method according to a thirteenth aspect is a rule storage means (for example, FIG. 11) storing a plurality of types of prosody information generation rules for generating prosody information of synthesized speech corresponding to an input sentence. A pose setting rule dictionary 38A shown in FIG. 2) and the like, wherein the input sentence is analyzed and any one of a plurality of prosody information generation rules stored in the rule storage means is selected. Then, using the prosody information generation rule, prosody information is generated from the analysis result of the input sentence, and a synthetic sound is generated based on the prosody information.

A voice synthesizer according to a fourteenth aspect is a voice synthesizer which generates a synthesized voice corresponding to an input sentence.
A parameter storage unit (for example, the prosody parameter dictionary 35 shown in FIG. 8) that stores a plurality of prosody parameters for controlling the prosody of the synthetic sound, and a plurality of prosody parameters stored in the parameter storage unit. Selection information input means (for example, the initial information setting unit 1 shown in FIG. 1) for inputting selection information for selecting one of them, and analysis means for analyzing an input sentence (for example, shown in FIG. 1) Selection means for selecting any one of a plurality of prosody parameters stored in the parameter storage means based on the selection information input from the language analysis unit 2) and the selection information input means (for example, in FIG. 8). Based on the analysis result of the analyzing means and the prosody parameter selecting section 37, etc.) and a synthetic sound of a prosody corresponding to the prosody parameter selected by the selecting means. Sound generation means (for example, the acoustic processing unit 15 shown in FIG. 1), characterized in that it comprises a.

A speech synthesis method according to a nineteenth aspect of the present invention is a parameter storage means (for example, the prosody parameter shown in FIG. 8) that stores a plurality of prosody parameters for controlling the prosody of a synthesized voice corresponding to an input sentence. A voice synthesizing method for a voice synthesizing device including a dictionary 35, etc., wherein when selection information for selecting one of a plurality of prosody parameters stored in the parameter storage means is input, Based on this, any one of the plurality of prosody parameters stored in the parameter storage means is selected, and a synthetic sound of a prosody corresponding to the prosody parameter is generated.

A speech synthesizer according to a twentieth aspect is a speech synthesizer which generates synthetic speech corresponding to an input sentence,
Parameter storage means (for example, the prosody parameter dictionary 35 shown in FIG. 8) that stores a plurality of prosody parameters for controlling the prosody of the synthetic voice, and feature detection means (for example, the prosody parameter dictionary 35 shown in FIG. 8). Field determination unit 6 shown in FIG.
Alternatively, based on the features of the input sentence detected by the feature detection unit, the reference sentence length determination unit 9 and the like), analysis means for analyzing the input sentence (for example, the language analysis unit 2 and the like shown in FIG. 1), and parameter storage. Selecting means for selecting one of the plurality of prosody parameters stored in the means (for example, the prosody parameter selecting unit 37 shown in FIG. 8) and the selecting means based on the analysis result of the analyzing means. It is characterized by comprising a synthesized sound generating means (for example, the acoustic processing unit 15 shown in FIG. 1) which generates a synthesized sound of a prosody corresponding to the prosody parameter.

A speech synthesis method according to a twenty-fourth aspect of the present invention is a parameter storage means for storing a plurality of prosody parameters for controlling the prosody of a synthesized voice corresponding to an input sentence (for example, the prosody parameters shown in FIG. 8). A voice synthesis method for a voice synthesizer including a dictionary 35, etc., wherein any one of a plurality of prosody parameters stored in the parameter storage means is detected based on the feature of the input sentence. Is selected, and a synthetic sound of a prosody corresponding to the prosody parameter is generated.

Of course, this description does not mean that each means is limited to the above.

FIG. 1 shows the configuration of an embodiment of a speech synthesizer to which the present invention is applied. In this speech synthesizer,
By reflecting the characteristics of the input sentence, it is possible to generate a synthesized voice that meets the user's wishes.

That is, the initial information setting unit 1 stores, for example, a pause setting rule (priority rule and narrowing rule) as a plurality of types of prosody information generation rules stored in a pose setting rule dictionary 38A (FIG. 11) described later. ), Initial information (selection information) for selecting any of the above can be input. Therefore, the user operates the initial information setting unit 1 to obtain the initial information,
For example, information about the field to which the input sentence belongs (for example, information that the input sentence is related to weather forecast, politics, etc.), information about the reference sentence length of the input sentence, and output from this speech synthesizer. It is possible to input information about the intended use of the synthetic voice (for example, the information that the synthetic voice is used for reading aloud or proofreading a document).

Here, the reference sentence length of the input sentence means the average sentence length of the input sentence input to the language analysis unit 2 described below. The sentence length means the prosodic length of the input sentence, and here, it is assumed to match the mora number (beat number) of the input sentence.

The initial information input by operating the initial information setting unit 1 includes the field determination unit 6, the reference sentence length determination unit 9, the prosody information generation rule selection unit 13, and the prosody information generation unit 14.
To be supplied.

An input sentence, which is a synthesized voice, is input to the language analysis unit 2, for example, in the form of text data. The input sentence input to the language analysis unit 2 does not necessarily have to be in the form of text data, but may be in the form of another (for example, output data from another program).

The language analysis unit 2 processes the input sentence by natural language processing,
That is, the language is analyzed. The analysis result of the input sentence in the language analysis unit 2 is supplied to the keyword extraction unit 3, the sentence length acquisition unit 7, and the prosody information generation unit 14.

The keyword extraction unit 3 detects (extracts) a keyword included in the input sentence based on the analysis result from the language analysis unit 2 and outputs it to the field determination unit 6. The keyword dictionary 4 stores keywords and fields in which the keywords are used in association with each other. The keyword history table 5 includes the keyword extraction unit 3
The history of keywords extracted from the input sentence (hereinafter, appropriately referred to as keyword history) is stored. The field determination unit 6 is configured to search the keyword supplied from the keyword extraction unit 3 from the keyword dictionary 4 and store the field associated with the keyword in the keyword history table 5. Then, the field determination unit 6 determines 1 of the characteristics of the input sentence input to the language analysis unit 2 based on the field of the keyword history stored in the keyword history table 5 and the initial information from the initial information setting unit 1. As an example, the field to which the input sentence belongs is determined, and the determination result is supplied to the emphasized word selection unit 11, the prosody information generation rule selection unit 13, and the prosody information generation unit 14.

The sentence length acquisition unit 7 calculates the sentence length as one of the features of the input sentence based on the analysis result of the input sentence from the language analysis unit 2, and outputs it to the reference sentence length determination unit 9. It is done like this. The sentence length history table 8 stores a sentence length history supplied from the sentence length acquisition unit 7 (hereinafter, appropriately referred to as a sentence length history). Standard sentence length determination unit 9
Is a sentence length history table 8 that calculates a reference sentence length based on the sentence length history stored in the sentence length history table 8. The information is output to the prosody information generation rule selection unit 13 and the prosody information generation unit 14. Note that the sentence length information does not need to be three ranks, that is, whether the reference sentence length is long, short, or intermediate, and can be information of two ranks or four ranks or more.

The emphasized word dictionary 10 stores words to be emphasized and their parts of speech for each field to which the input sentence belongs. The emphasized word selection unit 11 determines a word and a part of speech to be emphasized based on the field of the input sentence (the field to which the input sentence belongs) supplied from the field determination unit 6, and supplies the word and the part of speech to the prosody information generation unit 14. Has been done.

The parameter dictionary 12 stores parameters for generating prosody information. The parameter selection unit 13 uses the initial information from the initial information setting unit 1, the sentence length information from the reference sentence length determination unit 9, and the field of the input sentence from the field determination unit 6 to extract prosodic information from the parameter dictionary 12. Are selected and read out and supplied to the prosody information generation unit 14.

The prosody information generation unit 14 includes an initial information setting unit 1
From the reference sentence length determination unit 9, the sentence length information from the reference sentence length determination unit 9, the field of the input sentence from the field determination unit 6, the words and parts of speech to be emphasized from the emphasized word selection unit 11, the parameters from the parameter selection unit 13, Based on the analysis result of the input sentence from the language analysis unit 2, prosody information of the input sentence is generated and supplied to the sound processing unit 15.

That is, the prosody information generation unit 14 poses based on the initial information from the initial information setting unit 1, the sentence length information from the reference sentence length determination unit 9, and the field of the input sentence from the field determination unit 6. The language analyzing unit 2 is selected by selecting one of a plurality of types of pose setting rules stored in the setting rule dictionary 38A (FIG. 11) and using the parameters from the parameter selecting unit 13 according to the pose setting rule. The position of the pose which is one of the prosody information is set in the structure tree from. Further, the prosody information generation unit 14 is also configured to determine the size (size and position) of the accent, which is one of the prosody information, corresponding to the word and the part of speech to be emphasized from the emphasis word selection unit 11. ing. further,
The prosody information generation unit 14 is also configured to determine, for example, the speech rate and the size of a phrase as other prosody information based on the parameters from the parameter selection unit 13.

The acoustic processing unit 15 generates a synthetic sound based on the prosody information from the prosody information generation unit 14, and the speaker 15
It is designed to be supplied to A and output.

In the speech synthesizer configured as described above, first, the user operates the initial information setting section 1 to input the initial information. Here, the initial information setting unit 1 has a monitor, and the monitor has, for example, FIG.
Lists (application list, field list, reference) for setting the use of the synthesized voice as the initial information, the field of the input sentence, or the reference sentence length of the input sentence as shown in (A) to (C) of FIG. Sentence length list) is displayed. The user can input initial information by selecting one of the items in each list.

That is, when the user uses, for example, a synthetic voice for proofreading a document and the document is related to politics and the reference sentence length is short, the user list (FIG. 2A). "Proofing" from the field list (Fig. 2
From (B)), select "Politics" from the standard sentence length list (Fig. 2
By selecting "short" from (C)), the initial information can be input.

If the purpose, field, or standard sentence length is unknown, "unknown" is selected from each list. Further, the initial information does not necessarily have to be input, and when the initial information is not input, it is handled as "unknown" is selected.

Of the initial information, the field or the reference sentence length is supplied to the field determination unit 6 or the reference sentence length determination unit 9, and the use is supplied to the parameter selection unit 13 and the prosody information generation unit 14. To be done.

On the other hand, when an input sentence to be speech-synthesized is input to the language analysis unit 2, the input sentence is analyzed there. Here, FIG. 3 illustrates a configuration example of the language analysis unit 2. As shown in the figure, the language analysis unit 2 includes a morpheme analysis unit 21, a word dictionary 22, a connection rule dictionary 23, and a structure tree generation unit 24.

The morpheme analysis unit 21 describes a word dictionary 22 in which headings of words (morphemes), readings (phonological information), parts of speech, and the like are stored, and connection rules defining connection relationships between morphemes. The input sentence is morphologically analyzed with reference to the connection rule dictionary 23. As a result, the input sentence is divided into morphemes.

That is, for example, the input sentence "July 25, 1995"
The Social Democratic Party suffered a major defeat in the House of Representatives election on the day, and the position of Prime Minister Murayama in the coalition ruling party became even more difficult. Is input, the morphological analysis unit 21 outputs the following morphological analysis result, for example.

(Heisei Heise noun) (7 shichi number) (year nen number) ・ ・ ・ (, null reading point) (Council-selection sangiinsen noun) ・ ・ ・ (ta ta auxiliary verb) (. Null phrase) ... (1)

The above morphological analysis results (A, B,
In C), A means the morpheme heading, B means the phonological information (reading) of the morpheme, and C means the part of speech of the morpheme. In addition, the column of null for the phoneme information of the morpheme means that there is no reading.

The morpheme analysis result output from the morpheme analysis unit 21 is supplied to the structure tree generation unit 24, and also to the keyword extraction unit 3 and the sentence length acquisition unit 7.

In the structure tree generator 24, the morpheme analyzer 21
Based on the morpheme analysis result from, a structure tree representing the grammatical structure of the morphemes forming the input sentence is generated. That is,
In the structure tree generation unit 24, a structure tree in which the relation between clauses in the input sentence is expressed by a tree structure is generated according to a predetermined syntax rule. Specifically, the structure tree generation unit 2
When the morphological analysis result (1) is input to 4, the following structure tree is generated.

((Adverb phrase (Heisei Heise noun) (7 Shichi number) (Yearn number classifier) ・ ・ ・ (, null reading))) ((Consecutive phrase (Council election Sangiinsen noun) (de Particles)) (Consecutive phrases (Social Party Shakaito nouns) (Haha particles)) ・ ・ ・

The structure tree generated by the structure tree generator 24 is
It is supplied to the prosody information generation unit 14.

Returning to FIG. 1, the keyword extraction unit 3 extracts the keyword included in the input sentence based on the morphological analysis result from the language analysis unit 2 (morpheme analysis unit 21). It should be noted that in the present embodiment, a word whose word part is a noun (morpheme)
Is extracted as a keyword. Therefore, with respect to the morphological analysis result (1) as described above, the following keyword extraction result is obtained.

(Heisei Heise noun) (Council elected Sangiinsen noun) (Social Party Shakaito noun)

The above keywords are used for the field determination unit 6
Is output to The field determination unit 6 determines the field of the input sentence including the keyword supplied from the keyword extraction unit 3 by referring to the keyword dictionary 4.

Here, FIG. 4 shows an example of the structure of the keyword dictionary 4. As shown in the figure, the keyword dictionary 4
, Each keyword is associated with the field in which the keyword is used, and is also associated with the accuracy (probability) in estimating the associated field from the keyword as the field of the input sentence. Are remembered. In the present embodiment, it is assumed that the higher the accuracy, the more likely it is that the field of the input sentence is estimated to be the field associated with the keyword.

The field determination unit 6 searches the keyword extracted from the keyword extraction unit 3 from the keyword dictionary 4, reads out the keywords obtained as a result of the search together with the fields and the degrees of accuracy associated with them, and searches the keyword history. Store in Table 5. It should be noted that the field determination unit 6 cannot retrieve the keyword from the keyword extraction unit 3 from the keyword dictionary 4, that is, when the keyword from the keyword extraction unit 3 is not registered in the keyword dictionary 4. A message indicating that fact (hereinafter, appropriately referred to as “estimation impossible” message) is stored in the keyword history table 5.

FIG. 5 shows the stored contents of the keyword history table 5. The keyword history table 5 stores a predetermined number of latest information from the field determination unit 6. That is, the information from the field determination unit 6 is sequentially stored in the keyword history table 5 up to a predetermined number, and thereafter, when new information is input, the oldest information is deleted from the keyword history table 5. , New information is stored. Note that "point" in the rightmost column of the keyword history table 5 in FIG.
Corresponds to the accuracy of the field mentioned above.

Each time the field determination unit 6 receives a keyword from the keyword extraction unit 3, the field, the field, and the point (accuracy) (or the “estimation impossible” message) in the keyword history table 5 (hereinafter referred to as appropriate) Are collectively stored as keyword information). Then, the field determination unit 6 totals the points stored in the keyword history table 5 for each field, and the field having the largest total value is set as the field of the input sentence, and the emphasized word selection unit 11
And output to the parameter selection unit 13. That is, of the total values obtained as a result of totaling the points stored in the keyword history table 5 by field, for example, the field “politics”
When is the largest, the field “politics” is output to the emphasis word selection unit 11 and the parameter selection unit 13.

If the keyword stored in the keyword history table 5 and the keyword stored in the past are treated equally, the field of the input sentence currently input in the speech synthesizer is input in the past. If the field is different from the field of the input sentence that has been input, an error may occur in the determination (estimation) of the field of the input sentence that is currently being input due to the influence of the keywords stored in the past. Therefore, in order to prevent such an error, the field determination unit 6 reduces the points of already stored keyword information at a constant rate each time new keyword information is stored in the keyword history table 5. Thus, the latest keyword has the greatest influence on the determination of the field of the input sentence.

Further, the field determination unit 6 is the initial information setting unit 1
Therefore, when the field (however, not "unknown") of the initial information input by the user is received, the initial information is preferentially used. Therefore, in this case, from the field determination unit 6, the initial information setting unit 1
The field supplied from is output as it is. However, even when the field is supplied from the initial information setting unit 1, it is possible to determine the field based on the keyword information stored in the keyword history table 5 and output the field. It is also possible to add the field supplied from the initial information setting unit 1 to the keyword information stored in the keyword history table 5 to determine and output the field of the input sentence.

Upon receiving the field of the input sentence from the field determination section 6, the emphasized word selection section 11 refers to the emphasized word dictionary 10 and
Detect words (emphasized words) and parts of speech to be emphasized in the field. That is, in the emphasized word dictionary 10, as shown in FIG. 6, for example, words (emphasized words) and parts of speech that are preferably emphasized in each field are stored in association with the fields. In the emphasized word selection unit 11, the emphasized word and the part of speech associated with the field of the input sentence from the field determination unit 6 are searched from the emphasized word dictionary 10 and output to the prosody information generation unit 14. That is, for example, in the case where the emphasized word dictionary 10 is configured as shown in FIG. 6, when the field “weather forecast” is supplied from the field determination unit 6, the emphasized word selection unit 11 is used.
Then, "fine", "cloudy", and "rain" are determined as emphasis words, and the numbers, particles, proper nouns (prefecture name)
Is determined as the part of speech to be emphasized, and the prosody information generation unit 1
4 is output.

The emphasized word dictionary 10 stores a thesaurus in which words are hierarchically structured based on their meanings, in addition to emphasized words and parts of speech associated with the fields as shown in FIG. After the emphasizing word is obtained as described above, the emphasizing word selecting unit 11 can search for a synonym using a thesaurus, and the word obtained as a result can also be an emphasizing word. is there.

On the other hand, the sentence length acquisition unit 7 calculates the number of mora for each morpheme supplied from the language analysis unit 2, and further calculates the sum of the number of mora for the input sentence (one sentence). Then, the sentence length of the input sentence is required. This sentence length is supplied to the reference sentence length determination unit 9, and is output and stored in the sentence length history table 8. Here, FIG. 7 shows the stored contents of the sentence length history table 8. The sentence length history table 8 is similar to the keyword history table 5 described above.
The sentence length from the reference sentence length determination unit 9 is stored. That is, in the embodiment of FIG. 7, the sentence length history table 8
Is N from the latest input sentence supplied from the sentence length acquisition unit 7.
The sentence length of the input sentence up to the sentence before (sentence history for N sentences) can be stored.

Based on the sentence lengths of the N sentences stored in the sentence length history table 8, the reference sentence length determination unit 9 calculates the reference sentence length as an average value thereof. Further, the reference sentence length determination unit 9 determines whether the reference sentence length is long, short, or an intermediate level. When the reference sentence length is long, short, or in the middle thereof, “long”, “short”, or “medium” is used as the sentence length information, respectively, and the parameter selection unit 13 and the prosody information generation unit. 14 are supplied.

The number of moras, which serves as a reference for determining whether the reference sentence length is long, short, or an intermediate value, is calculated in advance by, for example, a statistical method, and the reference sentence length determination unit 9 Be set to.

In the standard sentence length determination unit 9, the sentence length information is not calculated based on the sentence length of N sentences, but only the sentence length of the latest input sentence supplied from the sentence length acquisition unit 7. It is also possible to ask based on it. However, the sentence length of the input sentence is usually different from one sentence to another. Therefore, if the sentence length information is obtained only by the sentence length of the latest input sentence, the sentence length information may change frequently. Become. As a result, the parameters selected by the parameter selection unit 13, which will be described later, and the rules for generating the prosody information (the pose setting rules in this embodiment) used by the prosody information generation unit 14 are frequently changed. However, such a state is not preferable. Therefore, in order to prevent such frequent changes of parameters and rules, it is desirable to obtain the sentence length information based on the sentence length of N sentences, as described above.

Furthermore, until the sentence length of N sentences is stored in the sentence length history table 8, the sentence length information may change frequently due to the influence of the sentence length of the latest input sentence. Therefore, in the sentence length history table 8, as an initial value,
For example, it is desirable to store the average number of mora per sentence obtained statistically.

Further, the reference sentence length judgment unit 9 includes information about the reference sentence length in the initial information input by the user from the initial information setting unit 1 (however, it is not "unknown").
When it receives, the initial information is preferentially used. Therefore, in this case, the reference sentence length determination unit 9 outputs the initial information (information about the reference sentence length) supplied from the initial information setting unit 1 as sentence length information. However, even when the initial information is supplied from the initial information setting unit 1, the sentence length information may be obtained and output based on the sentence length stored in the sentence length history table 8. It is also possible to add the initial information supplied from the initial information setting unit 1 to the sentence length stored in the sentence length history table 8 and obtain and output the sentence length information.

In the parameter selection unit 13, the information about the use of the initial information from the initial information setting unit 1, the sentence length information from the reference sentence length determination unit 9, and the field of the input sentence from the field determination unit 6 are selected. Based on the parameter dictionary 12,
For example, a pose setting parameter and a prosody parameter as parameters for generating the prosody information are selected and supplied to the prosody information generation unit 14.

That is, FIG. 8 shows a configuration example of the parameter dictionary 12 and the parameter selection unit 13. The parameter dictionary 12 stores a pose setting parameter dictionary 34 that stores the pose setting parameters used to determine the position of the pose to be inserted in the synthesized voice, and a parameter that determines other prosody information of the synthesized voice, for example, the synthesized voice. Parameters related to the speed of speech, pause length, phrase command size, accent command size, etc.
And a prosody parameter dictionary 35 that stores the prosody parameters. The parameter selecting unit 13 selects a pose setting parameter from the pose setting parameter dictionary 34.
6 and a prosody parameter selection unit 37 that selects a prosody parameter from the prosody parameter dictionary 35.

In the pose setting parameter selection unit 36, the pose setting parameters are selected from the pose setting parameter dictionary 34 based on the information about the use, the sentence length information, and the field of the input sentence. Here, in the pose setting parameter dictionary 34, for example, as shown in FIG. 9, the pose setting parameters are stored in association with the intended use of the input sentence, the field, and the reference sentence length. . In the pose setting parameter selection unit 36, information about the use from the initial information setting unit 1, the field of the input sentence from the field determination unit 6, or the sentence length information from the reference sentence length determination unit 9, the “use”, A line in which the description of “field” or “standard sentence length” is equal to each other is searched, and the pose setting parameters (three parameters of the pose setting parameters 1 to 3 in the embodiment of FIG. 9) described in the line are searched. Is read.

That is, for example, the information about the use supplied from the initial information setting unit 1, the field of the input sentence supplied from the field determination unit 6, or the sentence length information supplied from the reference sentence length determination unit 9 is If each is “reading”, “politics”, or “long”, the pose setting parameters 1 to 3
As each, the first of the pose setting parameter dictionary 34
P1-1, P2-1 and P3-1 described in the line are read.

In this embodiment, for example, the number of mora is used as the pose setting parameters 1 to 3.
That is, in FIG. 9, the columns of “usage, field, or standard sentence length” are “general”, “general”, or “medium” (fifth line), and the actual pose setting parameters 1 to 3 are set. 11, 26, or 46 is described as the value of, which represents the number of mora.

The information about the application supplied from the initial information setting section 1 may be "unknown" as described above. In this case, the application is, for example, general in the pause setting parameter selecting section 36. It is designed to be treated as a "general" that represents a specific purpose.

Further, in the present embodiment, regarding the use of the input sentence, only the information about the use supplied from the initial information setting unit 1 is used, but the use of the input sentence is estimated by some method. You may do it.

On the other hand, the prosody parameter selection unit 37 selects a prosody parameter from the prosody parameter dictionary 35 based on the information about the use, the sentence length information, and the field of the input sentence. Here, in the prosody parameter dictionary 35, for example, as shown in FIG. 10A, prosody parameters, that is, in the present embodiment, as described above, the speech rate, the pause length (pause length), and the phrase command. (Phrase symbol) and a specific numerical value of the accent command magnitude (accent symbol) are stored in association with the intended use, field, and reference sentence length of the input sentence. . In the prosody parameter selection unit 37, information about the use from the initial information setting unit 1, the field of the input sentence from the field determination unit 6, or the sentence length information from the reference sentence length determination unit 9 and the “use” and “field” , Or “standard sentence length” is searched for a line that is equal to the description, and the prosody parameter described in that line is read.

That is, for example, the information about the application supplied from the initial information setting unit 1, the field of the input sentence supplied from the field determination unit 6, or the sentence length information supplied from the reference sentence length determination unit 9 is When the reading is “reading”, “politics”, or “long”, the speech speed M, the pause length (set of pause lengths) S1, S2, S3, S4, S5, the size of the phrase command (of the phrase command) Size set) P0,
P1, P2, P3, P4, size of accent command (set of size of accent command) A0, A1, A2, A
3 and A4 are read from the first line of the prosody parameter dictionary 35.

In the present embodiment, the pause length, the phrase command size, and the accent command size set for the synthesized voice are set as described above in the prosody parameter dictionary 35.
Set of pose lengths S1, S2, S3 read from
S4, S5, set of phrase command size P0, P
1, P2, P3, P4, or each of the accent command size sets A0, A1, A2, A3, A4 (this selection is performed in the sound processing unit 15 as described later). .

Here, FIG. 10B shows specific numerical values of the speech rate, the pose length set, the phrase command size set, and the accent command size set read from the prosody parameter dictionary 35. Is shown. FIG.
When the prosody parameter as shown in 0 (B) is read from the prosody parameter dictionary 35, the speech rate is set to 7 mora / sec. Also, the pose length is 1500, 80
It is selected from 0, 400, 200, and 80 ms (milliseconds). Furthermore, the size of the phrase command is 0.5,
0.35, 0.25, 0.15, 0.1, and the size of the accent command is 0.7, 0.5, 0.
It is selected from 35, 0.2 and 0.1.

As described above, the parameter selection unit 13
In FIG. 1, the pose setting parameters and prosody parameters read from the parameter dictionary 12 are supplied to the prosody information generation unit 14.

Next, FIG. 11 shows an example of the structure of the prosody information generator 14. The pose setting rule dictionary 38A stores a plurality of types of pose setting rules (prosodic information generation rules) for determining (generating) the position of a pose, for example, as prosody information of synthetic speech. The pose setting unit 38 selects one of a plurality of types of pose setting rules stored in the pose setting rule dictionary 38A, and uses the pose setting rule to determine the position of the pose to be set in the synthetic sound. It is done like this. The position of the determined pose is reflected in the structure tree supplied from the language analysis unit 2 and supplied to the prosody tree generation unit 39.

The prosody tree generation unit 39 generates a prosody tree in which the prosody information of the synthetic voice is represented by a tree structure based on the structure tree supplied from the pose setting unit 38, and the accent processing unit 40.
To be supplied. Accent processing unit 40
Is adapted to convert the prosody tree from the prosody tree generation unit 39 into a prosody tree in consideration of accent fusion and output it to the prosody information generation unit 41. The prosody information output generation unit 41 is configured to convert the prosody tree from the accent processing unit 40 into a format that can be processed by the acoustic processing unit 15, and output the prosody tree to the acoustic processing unit 15.

Prosody information generator 1 configured as described above
In 4, the language analysis unit 2 supplies the structure tree to the pose setting unit 38, and the parameter selecting unit 13 supplies the pose setting parameters. Further, in the pause setting unit 38, information about the use of the input sentence (hereinafter referred to as use information) from the initial information setting unit 1, the field determining unit 6, or the reference sentence length determining unit 9 and the field of the input sentence. , Or sentence length information is also supplied.

The pause setting unit 38 is the initial information setting unit 1,
From the field determination unit 6 or the standard sentence length determination unit 9 respectively,
When the usage information, the field of the input sentence, or the sentence length information is received, the pose setting rule dictionary 38 is based on the information.
Any one of a plurality of types of pose setting rules stored in A is selected. That is, the pose setting rule dictionary 38A
, Each of a plurality of types of pose setting rules is stored in association with appropriate usage information, fields of the input sentence, and sentence length information. The pose setting unit 38 receives the usage information of the received input information and the input sentence. The pose setting rule associated with the field and the sentence length information is read from the pose setting rule dictionary 38A.

After that, when the pose setting unit 38 receives the structure tree or the pose setting parameter from the language analyzing unit 2 or the parameter selecting unit 13, respectively, the language analyzing unit 2
By applying the selected pose setting rule to the structure tree from 1 to 3 in consideration of the pose setting parameters from the parameter selecting unit 13, the position of the pose to be inserted into the synthesized sound is determined.

Therefore, since the pose setting unit 38 sets a pose at a position that matches the characteristics of the input sentence, it is possible to provide the user with a synthetic sound that is easy to hear.

Here, in this embodiment, the pose setting unit 38
In, for example, "Pose derivation method based on morpheme modification relation", Kagami, Asano, Akabane, Proceedings of ASJ,
The position of the pose is determined by using the method described in 1-P-32, pp401, October-November 1994 and the like. In this method, it is assumed that a grammatical unit such as a compound word, a clause, a phrase, or a clause is characterized by a morpheme located at the beginning and a morpheme located at the end thereof, and a grammatical unit A and a grammatical unit B following the grammatical unit
The priority order (priority) for setting a pause between and is determined based on the relationship between the morpheme located at the end of the grammatical unit A and the morpheme located at the beginning of the grammar unit B. The position of the final pose is determined by narrowing down the positions where the poses are inserted in consideration of the number of mora between the poses.

In this method, the priority rule for determining the priority order for setting the pose and the narrowing down rule for narrowing down the position to insert the pose are used as the rules for determining the position of the pose. Therefore, the pose setting rules stored in the pose setting rule dictionary 38A are also composed of such priority rules and narrowing down rules.

FIG. 12 shows an example of the priority rule and the narrowing rule. Note that FIG. 12A shows the priority rule and FIG. 12B shows the narrowing rule.

When the pause setting rule shown in FIG. 12 is selected, the pause setting unit 38 first applies the priority rule shown in FIG. The highest priority (priority 1) is given, and the next highest priority (priority 2) is given between a specific independent word or particle and the reading point. Furthermore, between a specific particle (excluding particles to which priority 2 is applied) and the reading point, between a specific connecting particle, a case particle, a particle, a parallel particle and a non-verbal independent word, or a conjugation word sequence. Priorities 3 to 5 are given between the body shape, the noun, the case particle “no”, and the noun. In addition, a priority 6 is attached to a portion having a non-modifying relationship except a portion to which the priorities 1 to 5 are applied.

When the setting of the priority is completed, the pose setting unit 38 finally determines the position where the pose is set according to the narrowing-down rule of FIG. That is, in the pose setting unit 38, according to the narrowing-down rule 1, first, a pose having a pose length corresponding to a node boundary is set in a portion having a priority of 1, and a portion having a priority of 2 is set in a portion having a priority of 2. A pose with a pose length corresponding to the phrase boundary is set.

Here, the pose length corresponding to the clause boundary or phrase boundary means the longest one or the second longest one of the set of pose lengths selected by the parameter selecting unit 13 described above, respectively. Therefore, FIG.
The pose length set (1500, 800,
400, 200, 80) is selected by the parameter selection unit 13, the part with priority 1 or 2 is
A 1500 or 800 ms pause will be inserted, respectively.

Further, according to the narrowing-down rule 2, the pose setting unit 38 assigns the priorities 3 and 4 so that the number of mora between the poses is equal to or more than <parameter 1> mora and less than <parameter 2> mora. In order of priority, the parts with higher priority (in this embodiment, the order of priority 3 and 4)
A pose is inserted in. In addition, according to the narrowing-down rules 3 and 4, the pose setting unit 38 assigns a priority to a section in which the condition of the narrowing-down rule 2 is not satisfied and the number of mora between pauses is <parameter 3> or more. Pauses are set in the areas in which the priority is high.

Here, in the pose setting section 38, the <parameter 1>, <parameter 2>, or <parameter 3> portion of the narrowing-down rule corresponds to the parameter selecting section 13
To the pose setting parameters 1 to 3 (FIG. 9). Therefore, from the parameter selection unit 13, as the pose setting parameters 1 to 3, 11 and 2 described in the fifth line of the pose setting parameter dictionary 34 of FIG.
When 6 and 46 are supplied to the pose setting unit 38, the <parameter 1>, <parameter 2>, or <parameter 3> part of the narrowing rule is 11, 26, or
Or 46 respectively.

The <parameter 1>, <parameter 2>, or <parameter 3> part of the narrowing rule disclosed in the above-mentioned document is mainly a numerical value obtained by statistically analyzing newspaper articles. Therefore, it is unclear whether the narrowing rule is appropriate for all kinds of input sentences. In contrast, FIG.
In the speech synthesizer of, in the parameter selection unit 13,
Since numerical values (pose setting parameters 1 to 3) corresponding to the characteristics of the input sentence (in the present embodiment, the field of the input sentence, the purpose, and the reference sentence length) are used, an appropriate narrowing rule can be set for each of the various input sentences. It becomes possible to apply.

In the same manner, the position of the pose is determined by applying the narrowing-down rule in the same manner, whereby the pose setting unit 38 causes the following structure tree to reflect the position of the pose, for example. Is output.

((Adverb phrase (Heisei Heise noun) (7 Shichi number) (Yearn number auxiliary number) ・ ・ ・ (, null reading point) PAU2) ((Conjunction phrase (Council-selection Sangiinsen noun) ( And particle))) (Consecutive phrase (Social Party Shakaito noun) (Haha particle)) ・ ・ ・ PAU1)

In the above structure tree, PAU #
i (PAU1 and PAU2 in the above case) indicates the position of the pose. The number #i following PAU is
Corresponds to the pose length. That is, the parameter selection unit 13
From, for example, when a set of pose lengths (1500, 800, 400, 200, 80) as shown in FIG. 10B is supplied, when the number #i is 1 to 5,
In the sound processing unit 15, the pose length is 150
It is set to 0, 800, 400, 200, or 80.

Further, the pose setting section 38 also divides the structure tree into prosodic words by using a known technique, if necessary.

Furthermore, the pose setting method is not limited to the method disclosed in the above-mentioned document.

The structure tree in which the position of the pose is reflected is supplied from the pose setting unit 38 to the prosody tree generating unit 39. In the prosody tree generation unit 39, the structure tree from the pose setting unit 38 is converted into a prosody tree.

That is, when the structure tree as described above is supplied from the pose setting section 38, for example, the prosody tree generation section 39.
Then, for example, the following prosodic tree is generated.

[0112]

Note that, in principle, the leading part in the parentheses represents the reading (phonological information) of the morpheme in capitalized Roman letters, and the ending part in the parentheses represents the part of speech of the morpheme. In addition, phonological information @ SHI is "shi"
(SHI) represents the unvoiced sound "Sui", and QI is
It represents the nasal dull sound "Nigi" of "Gi" (GI) (however, "N" is written in small size). Furthermore, -represents articulation and $ represents the consonant "tsu". Also, X
Indicates the sound repellency "n", and "represents the position of the accent.

The prosody tree is supplied from the prosody tree generator 39 to the accent processor 40. In addition to the prosody tree, the accent processing unit 40 includes a word and a part of speech (words of a part of speech to be emphasized) to be emphasized from the emphasized word selecting unit 11 (hereinafter,
Including both, it is said that the emphasis word) is also supplied. In the accent processing unit 40, for the prosodic tree,
Information on accents that change due to accent fusion is reflected. Further, in the accent processing unit 40, the same morpheme as the emphasized word from the emphasized word selecting unit 11 is searched from the prosodic tree, and an accent is set for the morpheme.

Since the emphasized word is a word that is preferably emphasized in the field of the input sentence as described above, the user is provided with a synthetic sound emphasizing such a word. On the other hand, it becomes possible to accurately convey information by the synthesized voice.

The accent setting method includes, for example, “Prosodic Rule for Synthesis of Japanese Text Speech”, Kawai, Hirose, Fujisaki, Acoustical Society Journal, Vol. Fifty,
No. 6, it is possible to use the method disclosed in pp433-442 and the like. However, the accent setting method is not limited to this.

For example, the above-described prosodic tree is processed by the accent processing unit 40 and converted as follows.

[0118]

The prosody tree processed by the accent processor 40 is supplied to the prosody information output generator 41. In the prosody information output generation unit 41, the prosody tree from the accent processing unit 40 is converted into a format that can be processed by the acoustic processing unit 15. That is, for example, the above prosodic tree is converted as follows.

"(HE-SE-) (@ SHICHIN
E'X) ... PAU2 (SAXQIIXSEXDE)
(Shakaito-wa) ... (KI $ SHI) P
AU3 ... PAU1 "

Here, the information converted as described above is
Since it includes both phonological information and prosody information, it is hereinafter referred to as phonological prosody information.

In addition to the prosody tree, the prosody information output generation unit 41 is also supplied with prosody parameters from the parameter selection unit 13. The prosody information output generation unit 41
After converting the prosodic tree into phonological prosodic information, the prosodic tree is output to the acoustic processing unit 15 together with the prosodic parameters.

The acoustic processing unit 15 stores a prosody control model for controlling prosodic features such as a fundamental frequency (pitch frequency) of a synthetic voice, a phoneme duration, and power (eg, phoneme-based power). Therefore, the prosody control model is driven using the prosody parameters from the prosody information output generation unit 41. That is, when a so-called Fujisaki model, for example, is stored as the prosody control model, the acoustic processing unit 1
In FIG. 5, for example, the size of a phrase command or an accent command for driving the Fujisaki model is a prosody parameter from the prosody information output generation unit 41 (a set of the size of the phrase command and a set of the size of the accent command). The Fujisaki model is driven by using one of the following. As a result, prosody data such as the fundamental frequency of the synthetic sound, the power and duration of the phoneme, is generated.

Specifically, as the prosody data, for example, the basic frequency value for each phrase of the synthetic sound, the number of frames indicating the duration of each phoneme, the number of frames indicating the power value of each syllable, and the like are generated.

Further, the sound processing section 15 has a built-in voice synthesizer, for example, by a waveform segment connection, as a synthesizer for performing voice synthesis by a method similar to the conventional one, and CV, CVC / VCV, etc. The voice unit data (eg, digitized waveform data) necessary for the regular voice synthesis is stored in units of, and the prosody information output generation unit 41
Based on the phonological information included in the phonological prosody information from the above, and the above-mentioned prosody data (specific numerical values of the fundamental frequency, duration, and power time, etc.), the necessary speech segment data (speech segment data) is continuous. Connect so that the sound waveform is as specified. The acoustic processing unit 15 also refers to the set of pause lengths in the prosody parameters from the prosody information output generation unit 41 based on the phonological prosody information from the prosody information output generation unit 41, and poses the voice waveform. insert. Then, the sound processing unit 15 performs necessary processing such as D / A conversion processing on the sound waveform obtained as described above and supplies the sound waveform to the speaker 15A for output. As a result, a synthetic sound corresponding to the input sentence is output.

As described above, the prosody parameter and the pause setting rule used for the voice synthesis are changed based on the initial information input by operating the initial information setting section 1, that is, the initial information corresponding to the initial information. Since the voice synthesis is performed according to the prosody parameter and the pause setting rule, it is possible to provide the user with an appropriate synthesized voice. Furthermore, since the prosody parameters and pause setting rules used for voice synthesis are changed in accordance with the characteristics of the input sentence, it is possible to accurately convey the information to the user by the synthesized voice.

In this embodiment, the parameter selection unit 1
In 3, the set of pose lengths is made up of 5 pose lengths. In addition, the set of pose lengths may be made up of less than 5, or more than 5 pose lengths. It is possible. However, when the set of pause lengths is configured by one pause length, the pause length inserted in the synthetic voice is always a fixed length, and it becomes difficult to obtain a natural synthetic voice. Therefore,
The set of pose lengths and the like preferably comprises a plurality of pose lengths.

Further, in the present embodiment, a plurality of types of pose setting rules for setting a pose are prepared, and among them,
We chose the one to be used based on the initial information and the characteristics of the input sentence, but we also prepared multiple types of rules for generating other prosodic information, and select the one to be used from them. It is possible to

[0129]

According to the speech synthesizing apparatus of the first aspect, the rule storage means stores a plurality of types of prosody information generation rules for generating prosody information of synthetic speech.
Then, when the selection information inputting means inputs the selection information for selecting any one of the plurality of types of prosody information generation rules stored in the rule storage means, the selection information is inputted in the prosody information generation means. On the basis of the above, any one of a plurality of types of prosody information generation rules stored in the rule storage means is selected, and the prosody information generation rule is used to generate prosody information from the analysis result of the analysis means. . Therefore, it becomes possible to provide a synthesized sound having a prosody desired by the user.

According to the speech synthesizer of claim 7,
The rule storage means stores a plurality of types of prosody information generation rules for generating prosody information of synthetic sounds. On the other hand, the feature detection means detects the features of the input sentence, and the prosody information generation means generates a plurality of types of prosody information stored in the rule storage means based on the features of the input sentence detected by the feature detection means. One of the rules is selected. Further, the prosody information generating means uses the prosody information generation rule to generate prosody information from the analysis result of the analyzing means. Therefore, it is possible to provide a synthetic sound having a prosody corresponding to the characteristics of the input sentence.

According to the speech synthesis method of the thirteenth aspect, the input sentence is analyzed, and one of the plural types of prosody information generation rules stored in the rule storage means is selected. Then, using the prosody information generation rule, prosody information is generated from the analysis result of the input sentence, and a synthetic sound is generated based on the prosody information. Therefore, it is possible to provide the user with a synthetic sound having an appropriate prosody.

According to the voice synthesizing apparatus and the voice synthesizing method of the nineteenth aspect, the selection information for selecting one of the plurality of prosody parameters stored in the parameter storage means. Is input, any one of the plurality of prosody parameters stored in the parameter storage means is selected based on the selection information, and a synthetic sound having a prosody corresponding to the prosody parameter is generated. Therefore, it is possible to provide a synthesized sound having a prosody desired by the user.

According to the speech synthesizing apparatus of claim 20 and the speech synthesizing method of claim 24, the characteristics of the input sentence are detected, and a plurality of parameters stored in the parameter storage means are detected based on the characteristics. Any of the prosody parameters are selected. Then, a synthetic sound having a prosody corresponding to the prosody parameter is generated. Therefore, it is possible to provide a synthetic sound having a prosody that matches the characteristics of the input sentence.

[Brief description of drawings]

FIG. 1 is a block diagram showing the configuration of an embodiment of a speech synthesizer to which the present invention is applied.

FIG. 2 is a diagram showing a list of initial information (application of synthesized speech, field of input sentence, and reference sentence length of input sentence) that can be set by initial information setting unit 1 in FIG.

3 is a block diagram showing a configuration example of a language analysis unit 2 in FIG.

FIG. 4 is a diagram showing a configuration example of a keyword dictionary 4 of FIG.

5 is a diagram showing a configuration example of a keyword history table 5 in FIG.

FIG. 6 is a diagram showing a configuration example of an emphasized word dictionary 10 in FIG.

7 is a diagram showing a configuration example of a sentence length history table 8 in FIG.

8 is a diagram showing a configuration example of a parameter selection unit 13 of FIG.

9 is a diagram showing a configuration example of a pose setting parameter dictionary 34 of FIG.

10 is a diagram showing a configuration example of a prosody parameter dictionary 35 of FIG.

11 is a block diagram showing a configuration example of a prosody information generation unit 14 in FIG.

12 is a diagram showing an example of pose setting rules stored in a pose setting rule dictionary 38A of FIG. 11. FIG.

[Description of symbols] 1 initial information setting unit 2 language analysis unit 3 keyword degree extraction unit 4 keyword dictionary 5 keyword history table 6 field determination unit 7 sentence length acquisition unit 8 sentence length history table 9 standard sentence length determination unit 10 emphasized word dictionary 11 Enhanced Word Selection Section 12 Parameter Dictionary 13 Parameter Selection Section 14 Prosody Information Generation Section 15 Acoustic Processing Section 34 Pose Setting Parameter Dictionary 35 Prosody Parameter Dictionary 36 Pose Setting Parameter Selection Section 37 Prosody Parameter Selection Section 38 Pose Setting Section 39 Pose Setting Rule Dictionary 40 Prosody Tree Generation Unit 41 Accent Processing Unit 42 Prosody Information Output Generation Unit

 ─────────────────────────────────────────────────── ─── Continuation of front page (72) Inventor Koji Asano 6-735 Kitashinagawa, Shinagawa-ku, Tokyo Sony Corporation

Claims (24)

[Claims] 1. A voice synthesizing device for generating a synthetic voice corresponding to an input sentence, comprising rule storage means for storing a plurality of types of prosody information generation rules for generating prosody information of the synthetic voice. A selection information input unit for inputting selection information for selecting one of a plurality of types of prosody information generation rules stored in the rule storage unit; an analysis unit for analyzing the input sentence; Based on the selection information input from the selection information input unit, any one of a plurality of types of prosody information generation rules stored in the rule storage unit is selected, and the prosody information generation rule is used, From the analysis result of the analysis means,
A speech synthesis apparatus comprising: a prosody information generation unit that generates the prosody information; and a synthetic sound generation unit that generates the synthetic sound based on the prosody information generated by the prosody information generation unit. 2. The voice synthesizing apparatus according to claim 1, wherein the selection information is information about an application of the synthetic sound. 3. The speech synthesizing apparatus according to claim 1, wherein the selection information is information about a field to which the input sentence belongs. 4. An emphasis word determining means for determining a word to be emphasized based on the selection information, which is information about a field to which the input sentence belongs, wherein the prosody information generating means includes the emphasis word determining means. The speech synthesis apparatus according to claim 3, wherein the prosody information is generated so that the word determined by is emphasized. 5. The speech synthesizer according to claim 1, wherein the selection information is information about a sentence length of the input sentence. 6. The speech synthesizer according to claim 1, wherein the prosody information is information about a position or a length of a pose. 7. A voice synthesizing device for generating a synthetic voice corresponding to an input sentence, comprising rule storage means for storing a plurality of types of prosody information generation rules for generating prosody information of the synthetic voice. A feature detecting unit that detects a feature of the input sentence, an analyzing unit that analyzes the input sentence, and a feature storing unit that stores the feature in the rule storage unit based on the feature of the input sentence detected by the feature detecting unit. A prosody information generation unit that selects any one of a plurality of types of prosody information generation rules and uses the prosody information generation rule to generate the prosody information from the analysis result of the analysis unit; and the prosody information generation unit. And a synthesized sound generating means for generating the synthesized sound based on the prosody information generated by the speech synthesis device. 8. The speech synthesizer according to claim 7, wherein the feature detecting means detects a field to which the input sentence belongs. 9. The method further comprises: an emphasized word determination unit that determines a word to be emphasized based on a field to which the input sentence detected by the feature detection unit belongs, wherein the prosody information generation unit includes the emphasized word determination unit. The speech synthesis apparatus according to claim 8, wherein the prosody information is generated so that the word determined by is emphasized. 10. A keyword detecting means for detecting a predetermined keyword from the input sentence is further provided, wherein the feature detecting means determines a field to which the input sentence belongs based on the keyword detected by the keyword detecting means. The speech synthesis apparatus according to claim 8, wherein the speech synthesis apparatus detects the speech. 11. The speech synthesis apparatus according to claim 7, wherein the feature detecting means detects a sentence length of the input sentence. 12. The speech synthesizer according to claim 7, wherein the prosody information is information about a position or a length of a pose. 13. A voice synthesizing method for a voice synthesizing device, comprising rule storing means for storing a plurality of types of prosody information generation rules for generating prosodic information of synthesized voice corresponding to an input sentence, said method comprising: The input sentence is analyzed, one of a plurality of types of prosody information generation rules stored in the rule storage means is selected, and the prosody information generation rule is used to determine the prosody from the analysis result of the input sentence. A voice synthesizing method characterized by generating information and generating the synthesized voice based on the prosody information. 14. A voice synthesizing device for generating a synthetic voice corresponding to an input sentence, comprising: parameter storage means for storing a plurality of prosody parameters for controlling the prosody of the synthetic voice; and the parameter storage. Selection information input means for inputting selection information for selecting one of a plurality of prosody parameters stored in the means, analysis means for analyzing the input sentence, input from the selection information input means Selection means for selecting any one of the plurality of prosody parameters stored in the parameter storage means on the basis of the selected information selected, and selection by the selection means based on the analysis result of the analysis means. And a synthesized sound generation unit that generates the synthesized sound having a prosody corresponding to the prosody parameter. 15. The voice synthesizing apparatus according to claim 14, wherein the selection information is information about a use of the synthetic sound. 16. The selection information is information about a field to which the input sentence belongs.
The speech synthesizer according to. 17. The speech synthesizer according to claim 14, wherein the selection information is information about a sentence length of the input sentence. 18. The speech synthesis apparatus according to claim 14, wherein the prosody parameter represents a speech rate, a phrase size, or an accent size. 19. A voice synthesizing method for a voice synthesizing device, comprising: a parameter storing means for storing a plurality of prosodic parameters for controlling a prosody of a synthetic voice corresponding to an input sentence, wherein the parameter storing means stores When selection information for selecting one of the plurality of stored prosody parameters is input, any one of the plurality of prosody parameters stored in the parameter storage means based on the selection information. A voice synthesizing method, characterized in that the selected voice is selected and the synthesized voice of the prosody corresponding to the prosody parameter is generated. 20. A voice synthesis device for generating a synthetic voice corresponding to an input sentence, comprising: parameter storage means for storing a plurality of prosody parameters for controlling the prosody of the synthetic voice; and the input sentence. Of the prosody parameters stored in the parameter storage unit based on the features of the input sentence detected by the feature detection unit Selection means for selecting any one of the above, and synthetic sound generation means for generating the synthetic sound of the prosody corresponding to the prosody parameter selected by the selection means based on the analysis result of the analysis means. A speech synthesizer characterized by the above. 21. The selection information is information about a field to which the input sentence belongs.
The speech synthesizer according to. 22. The speech synthesizer according to claim 20, wherein the selection information is information about a sentence length of the input sentence. 23. The speech synthesis apparatus according to claim 20, wherein the prosody parameter represents a speech rate, a phrase size, or an accent size. 24. A voice synthesizing method of a voice synthesizing device, comprising a parameter storage unit for storing a plurality of prosodic parameters for controlling a prosody of a synthetic voice corresponding to an input sentence, the feature of the input sentence. Detecting any one of the plurality of prosody parameters stored in the parameter storage means based on the characteristics of the input sentence, and generating the synthesized voice of the prosody corresponding to the prosody parameter. A method for synthesizing speech.