JP2023044436A - Synthetic voice generation data forming method, synthetic voice generation method, and synthetic voice generation device - Google Patents

Abstract

To provide a synthetic voice generation data forming method, a synthetic voice generation method, and a synthetic voice generation device for generating a synthetic voice of more natural utterance.SOLUTION: A pause length is allocated, as a pause position, to a position that satisfies predetermined conditions out of a symbol included in a text in text data and the text, and synthetic voice generation data in which pause length information indicating a pause length is inserted is formed at the pause position. In the synthetic voice generation data, in a case that the symbol included in the text is a comma, pause length information indicating a first pause length is inserted immediately after the comma as the pause length, and in a case where the symbol included in the text is a period, pause length information indicating a second pause length longer than the first pause length is inserted immediately after the period as the pause length. In addition, immediately after a comma and a period included in a text in parentheses in the text data, pause length information indicating pause lengths shorter than a comma and a period positioned outside the parentheses is each inserted.SELECTED DRAWING: Figure 2

Description

The present disclosure relates to a synthetic speech generation data formation method, a synthetic speech generation method, and a synthetic speech generation device.

2. Description of the Related Art In recent years, various services using a technique of converting text data into voice data and uttering the data have been provided. For this purpose, various techniques have been used to generate synthetic speech data from text data. (see Patent Document 1). In corpus-based speech synthesis, speech data uttered by a person is divided into predetermined units and stored in a database, and synthesized speech data is generated by concatenating the predetermined units of speech data extracted from the database during speech synthesis. (See Patent Document 1, for example).

Japanese Unexamined Patent Application Publication No. 2012-103668

However, in the speech synthesis method described above, although a predetermined unit of speech data uttered by a person is linked, it still does not lead to synthesized speech of natural utterances such as those read out by a person, and unnatural speech is produced. Synthetic speech may be generated.
An object of the present disclosure is to provide a synthetic speech generation data formation method, a synthetic speech generation method, and a synthetic speech generation apparatus that generate synthetic speech of more natural utterance.

In order to solve the above problems, a synthetic speech generation data formation method according to an aspect of the present disclosure sets a pause length as a pause position that satisfies a predetermined condition among symbols and text included in text in text data. Data for generating synthesized speech is formed by inserting pause length information indicating the length of the pause at the assigned pause position.

In order to solve the above problems, a synthesized speech generation method according to an aspect of the present disclosure provides a synthesized speech in which pause length information indicating the length of a pause is inserted at a pause position where a predetermined pause of text corresponding to text data is inserted. Acquire speech generation data, convert sentences corresponding to text data into phonetic notation, use phonetic notation to generate prosodic information of intonation and duration, and include speech data for each human-generated synthesis unit Select a synthesis unit corresponding to the phonetic notation from the speech database, connect the synthesis unit to the pause position included in the synthetic speech generation data via a pause of a length corresponding to the pause length information, and add prosody information to generate synthesized speech.

In order to solve the above problems, a synthesized speech generation apparatus according to an aspect of the present disclosure provides synthesis in which pause length information indicating the length of a pause is inserted at a pause position where a predetermined pause of text corresponding to text data is inserted. Synthetic speech generation data acquisition unit for acquiring data for speech generation, phonetic transcription conversion unit for converting sentences corresponding to text data into phonetic transcription, and sentence intonation using the phonetic transcription acquired from the phonetic transcription conversion unit and a prosody processing unit that generates prosody information of duration, and a synthesis unit corresponding to the phonetic notation acquired from the phonetic notation conversion unit from a speech database that includes human-generated speech data for each synthesis unit, and synthesized speech. a speech synthesizing unit that connects a synthesis unit to a pause position included in the generation data via a pause having a length corresponding to the pause length information, and adds prosody information to generate synthesized speech.

According to aspects of the present disclosure, it is possible to provide a synthetic speech generation data formation method, a synthetic speech generation method, and a synthetic speech generation device that generate synthetic speech of more natural utterance.

4 is a table showing an example of a conversion rule to pause length information in the synthetic speech generation data forming method according to the first embodiment of the present disclosure; 1 is a block diagram showing a configuration example of a synthetic speech generation data formation device that executes a synthetic speech generation data formation method according to a first embodiment of the present disclosure; FIG. FIG. 3 is a schematic diagram showing an example of synthetic speech generation data formed using the synthetic speech generation data formation method according to the first embodiment of the present disclosure; FIG. 10 is a block diagram showing a configuration example of a synthetic speech generation data formation device that executes a synthetic speech generation data formation method according to the second embodiment of the present disclosure; FIG. 5 is a schematic diagram showing an example of synthetic speech generation data formed using the synthetic speech generation data formation method according to the second embodiment of the present disclosure; FIG. 12 is a block diagram showing a configuration example of a synthetic speech generation device according to the third embodiment of the present disclosure; FIG. 13 is a block diagram showing another configuration example of the synthetic speech generation device according to the third embodiment of the present disclosure;

Hereinafter, the present disclosure will be described through embodiments, but the following embodiments do not limit the invention according to the claims. Also, not all combinations of features described in the embodiments are essential for the solution of the invention. Moreover, the drawings schematically show the invention according to the claims, and the configuration and function of each part are different from the actual method and apparatus.

1. First Embodiment Hereinafter, a synthetic speech generating data forming method according to the first embodiment will be described. In the first embodiment, a synthetic speech generation data formation program and a synthetic speech generation data formation device for causing a computer to execute a synthetic speech generation data formation method will be described.

(1.1) Synthetic Speech Generation Data Formation Method A synthetic speech generation data formation method according to the first embodiment will be described.
The synthetic speech generation data forming method according to the first embodiment is executed at least by the following method.
(A) Symbols included in text in text data and positions satisfying predetermined conditions in the text are assigned pause lengths as pause positions. (B) Pause length information indicating the pause length is inserted into the pause positions of the text. Formation of Data for Synthetic Speech Generation As described above, data for synthetic speech generation in which pause length information indicating a pause of an appropriate length is inserted at a predetermined position of the text is formed. When speech synthesis is performed using such data for generating synthetic speech, it is possible to generate synthesized speech that is more natural and uttered.
The pause length information has, for example, the following pause lengths.

(reading point)
In the data forming method for generating synthetic speech according to the present embodiment, when the symbol included in the text is a comma (,), pause length information indicating the first pause length is inserted immediately after the comma. is preferred.
The first pause length is, for example, preferably 300 msec to 500 msec, more preferably 350 msec to 450 msec, for example 400 msec.

In the method of forming data for generating synthetic speech according to the present embodiment, when the symbol included in the text is a period (.), a second pause length longer than the first pause length is set immediately after the period. It is preferable to insert pause length information indicating
The second pause length is, for example, preferably 900 msec or more and 1500 msec or less, more preferably 900 msec or more and 1100 msec or less, for example 1000 msec.
By forming synthetic speech generation data in which pause length information having a longer pause length than the reading point is inserted at the position of the period, synthetic speech of more natural utterance can be generated.

In addition, in the method of forming data for generating synthetic speech according to the present embodiment, if the symbol included in the text located between the symbols indicated by the brackets (“”) in the text data is a comma (,), puts pause length information indicating a third pause length, which is shorter than the first pause length (the pause length of the full stop located outside the brackets), immediately after the comma located between the brackets. preferably inserted.
The third pause length is, for example, preferably 150 msec or more and 300 msec or less, more preferably 150 msec or more and 250 msec or less, and is, for example, 200 msec.

Further, in the method of forming synthetic speech generation data according to the present embodiment, when the symbol included in the text located between the square brackets ("") is a full stop (.), Immediately after the positioned period, it is preferable to insert pause length information indicating a fourth pause length shorter than the second pause length (the pause length of the comma positioned outside the brackets) as the pause length.
The fourth pause length is, for example, preferably 450 msec to 900 msec, more preferably 650 msec to 750 msec, for example 700 msec.
The text located between the symbols indicated by the brackets is often the text indicating serifs or the like. For this reason, by making the pause length at the position of the period or comma inside the brackets shorter than the pause length at the position of the period or comma outside the brackets, the synthesized speech corresponds to the text inside the brackets. By shortening the pauses in the speech, it is possible to obtain synthesized speech with a more natural utterance.

(parentheses, etc.)
In the method of forming data for generating synthesized speech according to the present embodiment, when the symbols included in the text are parentheses, at least a pause indicating the fifth pause length as the pause length immediately before the front parenthesis (") among the parentheses It is preferred to insert the length information, preferably the pause length information is inserted just before the front bracket, or just before the front bracket and after the back bracket ("). For example, when parentheses are consecutive (for example, "", etc.), the pause length information is inserted only immediately before the front parenthesis (") and the back parenthesis (""). can be prevented from being duplicated.
Here, "parentheses" refer to brackets such as brackets (including double brackets), round brackets, corner brackets, square brackets, and curly brackets.

The fifth pause length is, for example, preferably 500 msec to 1000 msec, more preferably 500 msec to 600 msec, for example 500 msec.
As mentioned above, the text located between the bracketed symbols is often, for example, a dialogue or a phrase explaining an important matter. For this reason, by inserting pause length information at least just before the parentheses, a pause is inserted between the speech corresponding to the text in the parentheses and the speech before and after the synthesized speech, so that the text in the parentheses Synthetic speech of natural utterance can be obtained by making it easier for the listener to concentrate on the corresponding speech.

(heading)
In the method of forming synthetic speech generation data according to the present embodiment, when the text is a headline, pause length information indicating the sixth pause length is inserted immediately after the headline (an example of a position that satisfies a predetermined condition). . The sixth pause length is preferably longer than the other pause lengths inserted immediately before or after the non-heading text (ie, the first through fifth pause lengths).
The sixth pause length is, for example, preferably 1500 msec to 4500 msec, more preferably 2000 msec to 3000 msec. If the text contains multiple types of headings (e.g. main headings (e.g. headings at the beginning of each chapter) and sub-headings), the 6th pause length immediately after the main headings should be longer than the 6th pause length immediately after the sub-headings. is preferred. For example, the length of the sixth pause immediately after the main headline is assumed to be 3000 msec, and the length of the sixth pause immediately after the small headline is assumed to be 2000 msec.
In this way, by inserting a relatively long pause immediately after the headline, it is possible to easily concentrate the attention of the listener on the speech corresponding to the text of the headline, and to produce natural synthesized speech.

(Summary of sentences)
In the method of forming data for generating synthetic speech according to the present embodiment, when the text has a semantic unity, it is preferable to insert the pause length information immediately after the unity of the sentence (an example of a position that satisfies a predetermined condition). . Here, a "group of sentences" means, for example, a plurality of sentences described in one heading and explaining related contents. At this time, immediately after the unity of sentences, a pause indicating a seventh pause length longer than the other pause lengths inserted immediately before or after the text other than the headline (that is, the first to fifth pause lengths) Long information is preferably inserted.

The seventh pose length is preferably about the same as the headline, and when multiple types of headlines such as a large headline and a subheadline are used as headlines, a longer pose length than the subheadline, which has a relatively short pose length, is used. It is preferable to have The seventh pause length is, for example, preferably 2500 msec to 4500 msec, more preferably 3000 msec to 4000 msec. For example, if the sixth pause length immediately after the main heading is 3000 msec and the sixth pause length immediately after the subheading is 2000 msec, the seventh pause length is preferably 3000 msec.
In this way, by inserting a relatively long pause immediately after the unity of sentences, it becomes easier for the listener to understand the discontinuity of the content of the speech corresponding to the text, and natural-sounding synthesized speech can be obtained.

(others)
Further, as shown in FIG. 1, in the synthetic speech generating data formation method according to the present embodiment, pause length information can be inserted at a position that satisfies a predetermined condition according to each symbol or condition.
For example, pause length information can be represented by a leader such as a two-point leader (...) or three-point leader (...), a question mark (?), an exclamation point (!), a vertical bar (|), a dash (-), a round number, or a square. It is inserted immediately after symbols such as enclosed alphanumeric characters such as enclosed numbers. FIG. 1 shows a condition for inserting pause length information, an example of pause length (preferred range of pause length), and a specific example of pause length information. The leader indicates silence (pause) in conversation, lingering at the end of a sentence, omission in a sentence, and dashes also indicate pauses. Immediately after the leader, a pause length of 1000 msec to 1500 msec is preferably assigned, for example, 1000 msec. Since vertical lines often indicate breaks in sentences, it is preferable to assign longer pause lengths than, for example, full stops or commas. A pause length of 1000 msec to 1500 msec is preferably assigned immediately after the vertical line, for example, 1000 msec. In addition, since enclosed alphanumeric characters often indicate the beginning of a line of itemized sentences, for example, it is preferable to assign a pause length that is approximately the same length as a comma. Immediately after the enclosed alphanumeric characters, a pause length of 300 msec to 500 msec is preferably assigned, for example 300 msec.

Here, when two or more symbols such as punctuation marks and parentheses (so-called punctuation) are consecutive, no pause is assigned between consecutive symbols, and a pause is assigned only to the end of the consecutive symbols. can be If the symbols are consecutive, for example, it is preferable to insert the longer pause length of each symbol immediately after the following symbol.
For example, if text containing a question mark (?) at the end is written between front brackets (") and back brackets ("), no pause is assigned between the consecutive question marks and back brackets, You only need to assign a pause immediately after the trailing curly brace. At this time, at the pause position assigned immediately after the trailing square bracket, the pause length of the longer question mark (900-1500msec) or the pause length of the trailing square bracket (500-1000msec) (eg 1200 msec).

Pause lengths may also be assigned to other symbols not shown in FIG.
Further, when generating synthetic speech in a conventional synthetic speech generation apparatus, if a pause of minute length is included at a position other than a position that satisfies a predetermined condition among symbols and text included in the text, the pause may be generated. An eighth pause length information may be inserted at the position. In this case, the eighth pause length is, for example, preferably 130 msec or more and 200 msec or less, more preferably 140 msec or more and 170 msec or less, for example 150 msec.

Furthermore, the text for generating synthesized speech may include numbers indicating annotations. For this reason, speech prohibition information may be inserted as a tag before and after the annotation to prevent the text from being uttered, and synthetic speech generation data may be generated that does not include the number or the like indicating the annotation.
As a result, in the synthesized speech generated from the synthetic speech generation data, it is possible to prevent the annotation from being uttered, which is unrelated to the context of the text and hinders the natural utterance of the synthesized speech.

(1.2) Basic Configuration of Synthetic Speech Generation Data Formation Program A synthetic speech generation data formation program according to the present embodiment will be described. A synthetic speech generation data forming apparatus 10, which will be described later, forms synthetic speech generation data in accordance with a program that causes a computer to execute at least the following operations (a) and (b). The following programs are non-temporarily recorded on a recording medium such as a hard disk drive or memory, or an optical disc such as a DVD disc or Blu-ray (registered trademark). The following programs may be distributed via the Internet. Furthermore, the following programs may be recorded on a cloud server and executed via the Internet.

(a) assigning a pause length to a symbol included in the text in the text data and a position that satisfies a predetermined condition in the text as a pause position; (b) inserting pause length information indicating the pause length to the pause position of the text. forming data for generating synthesized speech

(1.3) Basic Configuration of Synthetic Speech Generation Data Formation Device Hereinafter, a synthetic speech generation data formation device 10 that executes the synthetic speech generation data formation method according to the first embodiment will be described with reference to FIG. do. FIG. 2 is a functional block diagram for explaining the basic configuration of the synthetic speech generation data forming apparatus 10 and the functions of each section.

As shown in FIG. 2, the synthetic speech generation data forming apparatus 10 includes a text data processing section 11 and a pause setting section 12 . Text data representing the content of a book, for example, is input to the synthetic speech generation data forming apparatus 10, and the text data processing section 11 and the pause setting section 12 process each section to generate an appropriate length of text at a predetermined position of the text. Synthetic speech generation data into which information indicating a pause is inserted is formed and output.

Here, as the text data input to the synthesized speech generation data forming apparatus 10, for example, typesetting data, which is data in which characters and the like are arranged according to layout designation, is used. The typesetting data includes, for example, a tag indicating at least one of information regarding a headline, sentence arrangement, line feed position, and blank line width in the case of a book. Therefore, in the synthetic speech generation data forming apparatus 10, it is possible to easily determine the part corresponding to the headline in the text and the last part of the group of sentences (for example, immediately before the headline), which is preferable.
The text data input to the synthesized speech generating data forming apparatus 10 may be, for example, manuscript data containing only character information (not containing tags for typesetting data, etc.).

The synthetic speech generation data forming apparatus 10 includes an input/output unit for inputting text data and outputting synthetic speech generation data, and a storage for storing a program for causing a computer to execute the above-described synthetic speech generation data forming method. and a controller (not shown in FIG. 2) for controlling the operation in the device. The text data processing section 11 and the pause setting section 12 are implemented by a computer executing a synthetic speech generation data forming program.
Each part of the synthetic speech generating data forming apparatus 10 will be described below.

<Text data processing unit>
When the input text data is typesetting data, the text data processing unit 11 extracts text from the text data and analyzes the text. The text data has a typesetting tag inserted therein. Also, in text data, when a plurality of sentences are described in one heading to explain related contents, these multiple sentences are a "unit of sentences". The text data processing unit 11 detects a heading (an example of a predetermined condition) according to a tag indicating a heading, a tag indicating a line feed, or the like.
In this case, it is not necessary to use the language dictionary 51, the part-of-speech dictionary 52, etc. shown in FIG. 2 for analyzing the text.

Further, when the input text data is manuscript data, the text data processing unit 11 analyzes the text in the text data to detect headlines and sentences. The text data processing unit 11 may analyze the text using, for example, the language dictionary 51 and the part-of-speech dictionary 52 to detect headlines and sentences.
Note that text analysis may be performed using a trained model generated by machine learning. For example, a trained model is generated by machine learning using text data in which tags indicating headlines, tags indicating the end of sentences, etc. are inserted as data for learning. By inserting text data in which tags as described above are not inserted into such a trained model and analyzing the text data, it is possible to extract headings or the ending part of a group of sentences from the text data.

<Pose setting part>
The pause setting unit 12 assigns a pause length to the position of the symbol included in the text in the text data as a pause position, and inserts pause length information indicating the pause length into the pause position. Also, the pause setting unit 12 inserts pause length information at a predetermined position in the text that is detected by the text data processing unit 11 and satisfies a predetermined condition. The pause length information is inserted immediately before or after the symbol according to an example of the rule table shown in FIG. The pause setting unit 12 inserts pause length information indicating a pause length corresponding to the detected condition.
As a result, the pause setting unit 12 forms synthetic speech generation data. The pause setting unit 12 outputs the generated synthetic speech generation data. Moreover, the pause setting unit 12 may store the synthetic speech generation data in a storage unit (not shown). The synthetic speech generation data stored in the storage unit can be output via the input/output unit.

FIG. 3 shows a specific example of a text representation of synthetic speech generation data into which pause length information has been inserted. In FIG. 3, for the sake of explanation, pause length information is shown only for symbols and part of text that satisfies a predetermined condition.
As shown in FIG. 3, the pause length of 3000 msec is indicated immediately after the text P1 of "Chapter 1: What is a lower-class elderly person?" Pause length information "<vtml_pause time="3000"/>" is inserted.
Similarly, pause length information “<vtml_pause time=“3000”/>” indicating that the pause length is 3000 msec is inserted immediately after the text P4 at the end of the set of sentences.

Among the texts in the synthetic speech generation data, text P2 of "What is a lower-ranked elderly person?" Pause length information “<vtml_pause time=“2000”/>” indicating that the pause length is 2000 msec is inserted immediately after the text P7 of “not”.
Among the texts in the synthetic speech generation data, the pause length information "< vtml_pause time=”1000”/>” is inserted.

Pause length information “<vtml_pause time=“500”/>” indicating that the pause length is 500 msec is inserted immediately before the parenthesis P3 in the text in the synthetic speech generation data.
Pause length information “<vtml_pause time=“300”/>” indicating that the pause length is 300 msec is inserted immediately after the circled number P6 in the text in the synthetic speech generation data.
In this manner, the pause setting unit 12 inserts pause length information indicating a pause length suitable for each of the symbols included in the text in the text data and the position of the text that satisfies a predetermined condition.

Further, the pause setting unit 12 may form synthetic speech generation data including the tags of the typesetting data together with the pause length information.
The above-described synthetic speech generation data forming apparatus can generate synthetic speech generation data for synthesizing natural speech that is similar to a person reading out characters.

2. Second Embodiment A method of forming synthetic speech generation data according to a second embodiment will be described below. Further, in the second embodiment, a synthetic speech generation data formation program and a synthetic speech generation data formation device for causing a computer to execute a synthetic speech generation data formation method will be described.

(2.1) Synthetic Speech Generation Data Formation Method The synthetic speech generation data formation method according to the second embodiment includes (A) and (B) of the synthetic speech generation data formation method according to the first embodiment, It is executed by the following methods.
(C) Inserting audio information for adding sound effects to audio data at a position in the text data that satisfies a predetermined condition. For example, if the text is a headline, the audio information immediately follows the headline. insert In addition, the acoustic information includes, for example, link destination information indicating a link destination of the acoustic data, that is, a storage destination of the acoustic data.

As described above, synthesized speech generation data is formed in which pause length information and acoustic information for adding acoustic effects to speech data are inserted at predetermined positions of the text. When speech synthesis is performed using such synthetic speech generation data, it is possible to generate synthetic speech that is more natural than synthetic speech, and to make it easy for listeners to understand scene changes even with synthesized speech alone. be able to. In addition, even without a voice editing device, audio information can be edited by simply editing the text in the tag that indicates the audio information, and it is possible to edit the link destination and set the sound such as reverb and echo. Generation and editing of data for generation becomes ready.

(2.2) Basic Configuration of Synthetic Speech Generation Data Formation Program A synthetic speech generation data formation program according to the present embodiment will be described. Synthetic speech generation data forming apparatus 20, which will be described later, generates synthetic speech generation data in accordance with a program that causes a computer to execute the operations (a) and (b) described in the first embodiment and the following (c). Form.
(c) Inserting acoustic information for adding acoustic effects to voice data at positions satisfying predetermined conditions in text in text data.

(2.3) Basic Configuration of Synthetic Speech Generation Data Formation Apparatus Hereinafter, a synthetic speech generation data formation apparatus 20 that executes a synthetic speech generation data formation method according to the second embodiment will be described with reference to FIG. do. FIG. 4 is a functional block diagram for explaining the basic configuration of the synthetic speech generation data formation device 20 and the function of each section.

As shown in FIG. 4, the synthetic speech generation data forming apparatus 20 includes a text data processing section 11 and a pause setting section 12 as well as an audio setting section 23 . That is, the synthetic speech generation data formation device 20 differs from the synthetic speech generation data formation device 10 in that the sound setting unit 23 is provided. The synthetic speech generation data formation device 20 inserts acoustic information for adding acoustic effects to the speech data at a position that satisfies a predetermined condition. Acoustic effects such as echo (reverberation) can be applied.
The sound setting unit 23 will be described below. Note that the text data processing unit 11 and the pause setting unit 12 have the same configurations as the units described in the first embodiment, so description thereof will be omitted.

<Sound setting section>
The sound setting unit 23 adds sound effects to the voice data obtained by reading the text, that is, applies sounds such as reverb and echo, or inserts sound effects at positions satisfying predetermined conditions in the text in the text data. inserts acoustic information for
When the text is a headline, the sound setting unit 23 inserts sound information at least one of before and after the headline. The sound information includes, for example, link destination information indicating the destination of the sound effect data, that is, the destination of the sound effect data.
Further, when the text is a headline, the sound setting unit 23 may insert sound information indicating the start point or end point of sound effects such as reverb and echo before and after the headline. In this case, a tag indicating the start of the sound effect is inserted as sound information before the headline, and a tag indicating the end of the sound effect is inserted as sound information after the headline.

FIG. 5 shows a specific example of a text representation of synthesized speech generation data in which acoustic information is inserted together with pause length information. In FIG. 5, for the sake of explanation, pause length information and sound information are shown only for symbols and part of text that satisfies a predetermined condition.
As shown in FIG. 5, among the texts in the synthetic speech generation data, just before the text P1 of "Chapter 1 What is a lower-class elderly person?" Acoustic information "<vtml_mark name="reverb_start"/>" and acoustic information "<vtml_mark name="sound: sound effect file.wav"/> indicating the link destination (URL of the save destination) of the sound effect to be played and are inserted. Further, among the texts in the synthetic speech generation data, audio information “<vtml_mark name=“reverb_end”/>” indicating the end of reverb is inserted immediately after the text P1, which is the headline.
The above-described synthetic speech generation data forming apparatus can generate synthetic speech generation data for synthesizing natural speech that is similar to a person reading out characters.

The above-described synthetic speech generation data forming apparatus generates synthetic speech generation data that includes acoustic information that is natural utterance as if a person were reading out characters and that can easily indicate a scene change to a listener. can be done.

3. Third Embodiment Hereinafter, a synthetic speech generation apparatus and a synthetic speech generation method according to a third embodiment will be described.
(3.1) Basic Configuration of Synthetic Speech Generation Method Hereinafter, a synthetic speech generation method according to the third embodiment will be described.
The synthetic speech generation method according to the third embodiment is executed at least by the following method.
(P) Acquire synthesized speech generation data in which pause length information indicating the length of the pause is inserted at the pause position where a predetermined pause of the text corresponding to the text data is inserted (Q) Pronounce the sentence corresponding to the text data (R) Generate prosodic information of intonation and duration using the phonetic transcription (S) Synthetic units corresponding to the phonetic transcriptions from a speech database containing speech data for each human-generated synthesis unit Select (T) Connect synthesis units to pause positions included in data for generating synthesized speech through pauses of length corresponding to pause length information, and add prosody information to generate synthesized speech.

In addition, the method for generating synthetic speech according to the third embodiment may be implemented by the following method before acquiring the data for generating synthetic speech.
(O) Generate synthetic speech generation data by inserting pause length information indicating the length of the pause at the pause position where a predetermined pause of the text corresponding to the text data is inserted. That is, the synthetic speech generation method according to the third embodiment. Then, data for generating synthesized speech may be formed separately.

(3.2) Basic Configuration of Synthetic Speech Generating Apparatus Hereinafter, a synthetic speech generating apparatus 100 that executes the data forming method for generating synthetic speech according to the third embodiment will be described with reference to FIG. FIG. 6 is a functional block diagram illustrating the basic configuration of the synthetic speech generating apparatus 100 and the functions of each section.

As shown in FIG. 6, the synthetic speech generation device 100 includes a language processing section 110, a prosody processing section 120, and a speech synthesis section . Synthesized speech generating apparatus 100 receives text data indicating the content of a book, for example, and generates synthesized speech of natural utterance through processing in language processing unit 110, prosody processing unit 120, and speech synthesis unit 130. For example, text data in which pause length information is not inserted is input to the synthesized speech generation apparatus 100 .

<Language processing unit>
The language processing unit 110 includes the functions of each unit of the synthetic speech generation data formation device 10 described in the first embodiment and the synthetic speech generation data formation device 20 described in the second embodiment.
The language processing unit 110 includes, for example, a text data processing unit 111 , a pause setting unit 112 , and a phonetic transcription conversion unit 114 . The text data processing unit 111 and the pause setting unit 112 form a synthetic speech generation data formation unit 115 having the same function as the synthetic speech generation data formation device 10 shown in FIG. 2 in the language processing unit 110 . The synthetic speech generation data forming unit forms synthetic speech generation data in which pause length information is inserted at positions (pause positions) satisfying predetermined conditions among symbols and text included in text in the text data. Here, the synthetic speech generation data formation unit may have the same configuration as the synthetic speech generation data formation device 20 shown in FIG.

The text data processing unit 111 and pause setting unit 112 of the language processing unit 110 have the same functions as the text data processing unit 11 and the pause setting unit 12 of the synthetic speech generation data forming apparatus 10 . That is, the language processing unit 110 differs from the synthetic speech generating data forming apparatus 10 described in the first embodiment in that it includes a phonetic transcription conversion unit 114 .
The phonetic notation conversion unit 114, the prosody processing unit 120, and the speech synthesis unit 130 of the language processing unit 110 will be described below. Also, descriptions of the text data processing unit 111 and the pause setting unit 112 are omitted.

<Language processing unit>
(Pronunciation converter)
The phonetic notation conversion unit 114 converts sentences corresponding to the input text data into phonetic notation. The text data input to the phonetic notation conversion unit 114 is synthetic speech generation data in which pause length information is inserted at predetermined pause positions in the text corresponding to the text data in the pause setting unit 112 .
The phonetic transcription converting unit 114 can communicate with, for example, the pronunciation dictionary 53, and converts the text corresponding to the input text data (data for generating synthesized speech) into phonetic transcription.
Note that the synthetic speech generation device 100 further includes a text data storage unit that stores text data (or synthetic speech generation data), and the language processing unit 110 acquires text data from the text data storage unit. good.

<Prosody processing part>
The prosody processing unit 120 includes a prosody information generation unit 121 that generates prosody information of sentence intonation and duration using the phonetic transcription acquired from the phonetic transcription conversion unit 114 of the language processing unit 110 . The prosody processing unit 120 outputs the generated prosody information to the speech synthesis unit 130 .

<Speech synthesizer>
The speech synthesizing unit 130 generates synthetic speech based on text data (synthetic speech generating data) and a speech database 54 containing speech data generated by humans for each synthesis unit. The speech synthesis unit 130 includes a synthesis unit selection unit 131 and a synthesis unit connection unit 132 .
Each unit of the speech synthesizing unit 130 will be described below.

(Synthetic unit selection part)
The synthesis unit selection unit 131 selects and extracts a synthesis unit corresponding to the phonetic transcription acquired from the phonetic transcription conversion unit 114 from the speech database 54 . The synthesis unit selection unit 131 transmits the extracted synthesis unit to the synthesis unit linking unit 132 .

(Synthetic unit connecting part)
The synthesis unit connection unit 132 connects the synthesis units extracted by the synthesis unit selection unit 131 and adds prosodic information to generate synthesized speech. At this time, the speech synthesizing unit 130 connects the synthesis unit to the pause position included in the synthetic speech generation data via a pause of a length corresponding to the pause length information, thereby generating synthetic speech of natural utterance. . In addition, if the data for generating synthetic speech is data for generating synthetic speech that includes information on the headline, sentence arrangement, position of line feed, and width of blank lines in the case of a book, together with the pause length information, the headline, Synthetic speech of natural speech may be generated by linking synthesis units via pauses corresponding to line breaks and blank line widths.

(3.3) Modification 1
In the third embodiment, the synthetic speech generating apparatus 100 in which text data for which speech synthesis is desired is inserted and speech synthesis is performed based on the text data has been described, but the configuration is not limited to this.
For example, the synthetic speech generation device 100A of Modification 1 may be a device that receives synthetic speech generation data separately generated and performs speech synthesis. In this case, as shown in FIG. 7, the language processing unit 110A of the synthetic speech generation device 100A does not have the text data processing unit 111 and the pause setting unit 112, and at least has the phonetic transcription converting unit 114. good. Synthetic speech generation data generated by the synthetic speech generation data forming unit 115 of the third embodiment is input to the synthetic speech generation device 100A. Therefore, if the phonetic transcription conversion unit 114 is provided, it can function as the language processing unit 110A of the synthetic speech generation device 100A.

(3.4) Modification 2
The language processing unit 110 may include, for example, a text data processing unit 111, a pause setting unit 112, a phonetic transcription conversion unit 114, and a sound setting unit (not shown). In this case, the sound setting section has the same function as the sound setting section 23 of the synthetic speech generation data forming apparatus 20 .
When the language processing unit 110 includes a sound setting unit, the speech synthesis unit 130 superimposes the sound effect obtained from the link destination of the sound information at a predetermined position (for example, the position of the headline) when connecting the synthesis units. For example, it is possible to apply sound effects such as reverb to headlines.

(3.5) Modification 3
In the third embodiment, the synthetic speech generation device 100 has the language processing unit 110 having the same function as the synthetic speech generation data forming device 10, but the configuration is not limited to this.
For example, the language processing unit 110 of the synthetic speech generation device 100 may have the same function as the synthetic speech generation data forming device 20 according to the second embodiment.

Although the embodiments of the present disclosure have been described above, the technical scope of the present disclosure is not limited to the technical scope described in the above-described embodiments. Various changes or improvements can be made to the above-described embodiments, and forms with such changes or improvements can also be included in the technical scope of the present disclosure. it is obvious.

10, 20 Synthetic speech generation data forming device 11, 111 Text data processing unit 12, 112 Pose setting unit 30 Machine learning device 31 Text data conversion unit 32 Storage unit 33 Learning data extraction unit 34 Learning unit 51 Language dictionary 52 Part of speech dictionary 53 Pronunciation dictionary 54 Speech database 100 Synthetic speech generation device 110 Language processing unit 114 Phonetic transcription conversion unit 120 Prosody processing unit 121 Prosody information generation unit 130 Speech synthesis unit 131 Synthesis unit selection unit 132 Synthesis unit connection unit

Claims (14)

assigning a pause length to a position satisfying a predetermined condition among symbols included in text in the text data and the text as a pause position;
A synthetic speech generation data forming method for forming synthetic speech generation data in which pause length information indicating the pause length is inserted at the pause position. if the symbol included in the text is a reading point, inserting the pause length information indicating a first pause length as the pause length immediately after the reading point;
2. When the symbol included in the text is a period, the pause length information indicating a second pause length longer than the first pause length is inserted immediately after the period as the pause length. 3. The method of forming data for generating synthetic speech according to . When the symbol included in the text located between the symbols indicated by the square brackets in the text data is a comma, the pause length is immediately after the comma located between the square brackets. inserting the pause length information indicating a third pause length shorter than the first pause length as
When the symbol included in the text located between the square brackets is a period, the pause length immediately after the period located between the square brackets is shorter than the second pause length. 3. The method of forming synthetic speech generation data according to claim 2, wherein said pause length information indicating a fourth pause length is inserted. 4. The pause length information indicating a fifth pause length as the pause length is inserted at least immediately before the preceding parenthesis among the parentheses when the symbol included in the text is a parenthesis. 2. The data forming method for generating synthetic speech according to 1 or 2 above. When the text is a headline, inserting the pause length information immediately after the headline indicating a sixth pause length longer than other pause lengths inserted immediately before or after the text other than the headline. 5. The method of forming data for generating synthetic speech according to any one of claims 1 to 4. Synthetic speech generation according to any one of claims 1 to 5, wherein when the text has a semantic unity, the pause length information indicating a seventh pause length is inserted immediately after the text unity. data formation method. 7. The synthetic speech generation use according to any one of claims 1 to 6, wherein acoustic information for adding a sound effect to the speech data is inserted at a position of the text in the text data that satisfies a predetermined condition. Data formation method. 8. The method of forming data for generating synthetic speech according to claim 7, wherein when the text is a headline, the acoustic information is inserted immediately after the headline. 9. The method of forming synthetic speech generation data according to claim 8, wherein said acoustic information includes link destination information indicating a link destination of said acoustic data. Acquiring synthesized speech generation data in which pause length information indicating the length of the pause is inserted at a pause position where a predetermined pause of the text corresponding to the text data is inserted;
converting the text corresponding to the text data into a phonetic transcription;
generating prosodic information of intonation and duration using the phonetic transcription;
selecting a synthesis unit corresponding to the phonetic notation from a speech database containing human-generated speech data for each synthesis unit;
Synthetic speech for generating synthetic speech by connecting the synthesis unit to the pause position included in the data for generating synthetic speech via the pause having a length corresponding to the pause length information and adding the prosody information. generation method. a synthetic speech generation data acquisition unit for acquiring synthetic speech generation data in which pause length information indicating the length of the pause is inserted at a pause position where a predetermined pause of the text corresponding to the text data is inserted;
a phonetic notation conversion unit that converts the text corresponding to the text data into a phonetic notation;
a prosody processing unit that generates prosody information of intonation and duration of the text using the phonetic transcription obtained from the phonetic transcription conversion unit;
A synthesis unit corresponding to the phonetic notation acquired from the phonetic notation conversion unit is selected from a speech database containing speech data generated by a person for each synthesis unit, and placed at the pose position included in the data for generating synthesized speech. a speech synthesizing unit that connects the synthesis units through the pauses of a length corresponding to the pause length information and adds the prosody information to generate synthetic speech;
Synthetic speech generator. A symbol included in the text and a position that satisfies a predetermined condition in the text are assigned a pause length as the pause position, and the data for generating synthesized speech in which the pause length information indicating the pause length is inserted into the pause position. further comprising a synthetic speech generation data formation unit that forms
12. The synthetic speech generation apparatus according to claim 11, wherein the synthetic speech generation data acquisition unit acquires the synthetic speech generation data from the synthetic speech generation data formation unit. 13. The synthetic speech generating apparatus according to claim 11, wherein said text data is data including information on the arrangement of said text, line feed position and blank line width in the case of a book. 14. The synthetic speech generating apparatus according to claim 13, further comprising a text data storage unit storing said text data.

Images