JP2006349787A - Method and device for synthesizing voices - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To easily grasp an image concerning an actual synthesized sound output relating to voice synthesis, when editing synthesized voice data. <P>SOLUTION: The method for synthesizing voices including phonetic signs set to a text sentence and processing for editing a reading form when producing composite voices from the input text sentence, includes sentence separation processing for making a set of separated sentences (separated sentence data 1551) of a limited length by separating a text mode sentence on the basis of a predetermined sentence separation reference by a sentence separation processing part 11, and the editing processing is carried out in units of the separated sentences obtained by the sentence separation processing. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a speech synthesis technique for generating synthesized speech from a given text sentence.

  In speech synthesis processing, syntactic analysis is performed on a given text sentence mixed with kana and kana, a kana character string including phonetic symbols and prosodic symbols is generated, a desired reading format is set, and these synthesized sounds are set. A synthesized sound is generated based on the data. More specifically, standard settings (default) are provided for pronunciation, prosody, and reading format. By this standard setting, the above kana character string is first generated and the reading format is set, and then they are displayed on the editing screen. An operation of displaying and editing the standard set synthesized sound data to be changed manually is performed in the speech synthesis process.

  The editing work in such speech synthesis processing is performed while imagining the synthesized sound that is actually output. Therefore, whether the actual output synthesized sound image is easy to grasp on the editing screen greatly affects the work efficiency. That is, if it is difficult to grasp the output image, the work of re-editing after listening to the actually output synthesized sound is repeated many times, resulting in poor work efficiency.

  For this reason, various ideas have been made for editing the synthesized sound data. For example, in the speech synthesizer disclosed in Patent Document 1, it is shown that a character string representing the prosody for the sentence is displayed by the relative position relationship between the state display of each character itself and other characters in the synthesized sound data. Yes. Specifically, the sound quality is the display color, the voice pitch is the vertical position of the character display position, the utterance speed is the character spacing, the voice loudness is the character brightness, and the accent size is above the character. A character string representing the prosody and the like is displayed on the editing screen by the accented symbol so that the output image of the synthesized sound can be visually grasped.

JP-A-8-77152

  The method disclosed in Patent Document 1 that enables the output image of the synthesized sound to be visually grasped can be expected to be effective. However, the method disclosed in Patent Document 1 complicates the display of prosody and the like. Therefore, if the text sentence to be synthesized is longer than a certain level, it is highly likely that it is difficult to grasp the output image due to the complicated display of prosody and the like. In other words, it is considered that the difficulty of imaging the output synthesized sound in editing work is more affected by the length of the text sentence to be processed. If this is not taken into account, there is a limit to the ease of imaging the output synthesized sound. That is.

  The above is the problem of imaging the output synthesized sound when the synthesized sound data is edited in the speech synthesis process. However, the speech synthesis process also has a problem regarding the editing of the text sentence reading speed and pause length by the synthesized sound. In other words, in the editing of synthesized sound data, the reading speed by the synthesized sound is also subject to editing, but when there is a limit on the reading time of the given text sentence, the optimum reading speed etc. in relation to this time limit etc. It is a problem that it is not always easy to set.

  Furthermore, the voice synthesis process may require a more effective synthesized sound depending on the purpose of use of the synthesized sound obtained by the voice synthesis process. It has become a challenge.

  The present invention has been made on the basis of the above-described knowledge, and it is possible to more easily grasp an image of an actual output synthesized sound when editing synthesized sound data with respect to speech synthesis. The primary purpose is to increase the efficiency of the synthesis process, and it is easier to edit the synthesized speech data when there is a time limit for reading a given text sentence for speech synthesis. The second object is to do so, and the third object is to enable generation of synthesized sound that is more effective in accordance with the purpose of use.

  For the first purpose, in the present invention, in the speech synthesis method including the process of editing the phonetic symbols and the reading format set in the text sentence, the synthetic speech is generated from the input text sentence. Including a sentence delimiter process in which a sentence of a text sentence is delimited by a predetermined sentence delimiter criterion to make a set of delimited sentences of a finite length, and the editing process is performed in units of delimiter sentences obtained by the sentence delimiter process It is characterized by being made.

  In the present invention, in the speech synthesis method as described above, terminal characters such as a punctuation mark, a question mark, and an exclamation mark are used as the sentence delimiter.

  In the present invention, the speech synthesis method as described above includes a process for displaying the reading format information editing screen and editing the reading format information and a process for displaying the pronunciation information editing screen and editing the pronunciation information. It is supposed to be.

  Further, in the present invention, for the first object, in a speech synthesizer that generates synthesized speech from an input text sentence, the sentence of the text sentence is delimited by a predetermined sentence delimitation criterion, and is separated by a finite length. It is characterized by having a sentence delimiter processing unit as a set of sentences.

  In the present invention, in the speech synthesizer as described above, a terminal character such as a punctuation mark, a question mark, or an exclamation mark is used as the sentence delimiter.

  Further, in the present invention, for the speech synthesizer as described above, for the intermediate synthesized sound data generated by setting pronunciation information and reading format information in the text sentence, the reading format information is edited in units of the delimited sentences. A reading information processing unit is provided.

  According to the present invention, the speech synthesizer as described above further includes a pronunciation information processing unit for editing the pronunciation information for the intermediate synthesized sound data in units of the delimited sentences.

  Further, in the present invention, for the second object, in the speech synthesizer for generating synthesized sound from the input text sentence, intermediate synthesized sound data generated by setting pronunciation information and reading format information to the text sentence. A synthetic sound data display / editing unit for editing the sound, and the synthetic sound data display / editing unit obtains a time related to reproduction when the generated synthetic sound is reproduced and output, and is used for the editing. It is characterized by including a time calculation unit to be displayed on the screen.

  In the present invention, the speech synthesizer as described above is provided with a sentence delimiter processing unit that divides the sentence of the text sentence according to a predetermined sentence delimiter criterion to make a set of delimited sentences of a finite length, and the time calculation The section obtains a delimiter sentence reproduction time required for the synthesized sound reproduction for each delimiter sentence and a delimiter sentence reproduction start time from the start of the synthesized sound reproduction for the text sentence until the delimited sentence reproduction starts. I try to let them.

  Further, in the present invention, for the third object, in a speech synthesizer that generates a synthesized sound from an input text sentence, a sound effect / background sound storage unit that stores a sound effect and a background sound put on the synthesized sound And a reading information processing unit for performing an editing process for adding sound effects and background sounds to the synthesized sound.

  In the present invention, sentence separation processing is performed by dividing a text sentence according to a predetermined sentence separation criterion into a set of separation sentences having a finite length, and editing of phonetic symbols and reading format can be performed in units of separation sentences. ing. For this reason, according to the present invention, even when the text sentence is a long sentence, it is possible to more easily grasp an image of the actual output synthesized sound during editing, and the efficiency of the speech synthesis process can be increased. Become.

  Further, in the present invention, a time related to reproduction when the generated synthesized sound is reproduced and output is obtained, and can be displayed on a screen for editing. Therefore, according to the present invention, it is possible to more easily edit the synthesized sound data when there is a time limit in reading a given text sentence.

  In the present invention, there is provided a sound effect / background sound storage unit for storing a sound effect and a background sound to be put on a synthesized sound, and a reading information processing unit for performing an editing process for putting the sound effect / background sound on the synthesized sound I am trying to provide it. For this reason, according to the present invention, it becomes possible to put a sound effect and a background sound on the synthesized sound, and it is possible to generate a synthesized sound that is more effective according to the purpose of use.

  Hereinafter, modes for carrying out the present invention will be described. To implement the present invention, a computer program related to a speech synthesis method as a software element is mounted on a data processing device such as a personal computer as a hardware element to constitute a speech synthesis apparatus. FIG. 1 shows a system configuration of a speech synthesizer according to an embodiment. The speech synthesizer 1 according to this embodiment includes a sentence delimiter processing unit 11, a text analysis unit 12, a word dictionary 13, a display / operation unit 14, a synthesized sound data display / editing unit 15, a phoneme piece storage unit 16, and an acoustic processing unit 17. , An audio file creation unit 18, a sound effect / background sound storage unit 19, an audio file playback unit 20, and a speaker 21.

  The sentence delimiter processing unit 11 performs a process of making a set of delimited sentences having a finite length by delimiting the sentence of the input text sentence T according to a predetermined sentence delimiter criterion. Usually, terminators such as punctuation marks, punctuation marks, question marks, and exclamation marks are used as sentence break criteria. In the present embodiment, punctuation marks, question marks, and exclamation marks are used as sentence delimiters. The text sentence T divided into a plurality of delimiter sentences by the sentence delimiter processing section 11 is passed to the text analysis section 12, and the delimited sentence is used as a unit in the text storage section 155 described later in the synthesized sound data display / editing section 15. It is stored as a set of delimited sentence data 1551.

  The text analysis unit 12 receives the text sentence T delimited by the delimiter sentence and performs analysis using the word dictionary 13. The analysis by the text analysis unit 12 is a standard setting of syntactic analysis of the text sentence T and pronunciation information based thereon. In the standard settings of pronunciation information, settings for pronunciation (reading, accenting, intonation, pose setting, etc.) are generated based on syntax analysis to generate standard synthesized sound data, and kana characters with phonetic symbols and prosodic symbols Generate a column. These processes are well-known processes in the field of speech synthesis. The result of setting phonetic symbols as standard is stored as delimited sentence data 1551 in the text storage unit 155. Here, in addition to the standard setting of the pronunciation information by the text analysis unit 12, by setting a standard reading format, synthesized voice data of the default setting by the standard setting is generated. However, the final synthesized sound data requires an acoustic conversion process that assigns phoneme data (phoneme segments) based on the settings of pronunciation information and reading format information, so only the pronunciation information and reading format information are set. The data at the point in time is intermediate and is intermediate synthetic sound data. The standard reading format is set by the reading information processing unit 151 in the synthesized sound data display / editing unit 15 as described later.

  The word dictionary 13 is a dictionary used for converting the kanji in the text sentence T mixed with kanji into kana by the text analysis unit 12.

  The display / operation unit 14 includes a monitor that displays data necessary for speech synthesis processing and an input / output device such as a keyboard and a mouse that performs input / output of data necessary for speech synthesis processing.

  The synthesized sound data display / editing unit 15 mainly functions for editing processing of intermediate synthesized sound data, that is, editing processing for changing settings of pronunciation information and reading format information in the intermediate synthesized sound data by a user's manual operation. In the editing process, the synthesized sound data already generated by standard setting or the like is displayed on the monitor in the display / operation unit 14, and the user changes the reading format information and the pronunciation information in that state. The synthesized sound data display / editing unit 15 includes a reading information processing unit 151, a pronunciation information processing unit 152, a time calculation unit 153, a display processing unit 154, and a text storage unit 155.

  The reading information processing unit 151 has three functions. The first function is a function for editing processing that changes the setting of reading format information by a user's manual operation. A specific example of this function will be described later.

  The second function of the reading information processing unit 151 is a function relating to editing processing for placing a sound effect such as a bell sound or a background sound such as music on the synthesized sound. By applying sound effects and background sounds to the synthesized sound by this function, the synthesized sound can be made more effective according to the purpose of use. A specific example of the editing process in which sound effects and background sounds are added to the synthesized sound will be described later.

  A third function of the reading information processing unit 151 is a function for generating standard intermediate synthetic sound data by further setting a standard reading format for the delimited sentence data in which standard pronunciation information is set. is there. The processing by this function is performed by setting a standard reading format for each delimiter sentence data, and is performed in a format in which the setting information is displayed with a predetermined symbol at the beginning of each sentence. The setting information regarding the reading format includes “voice type: male / female”, “voice output format: 8 kHz 8 bit / 11 kHz 8 bit / 16 kHz 8 bit / 22 kHz 8 bit / 8 kHz 16 bit / 11 kHz 16 bit / 16 kHz 16 bit / 22 kHz 16 bit”, “reading method: normal reading / live reading”, “Emotion effect: calm / joy / anger / sadness”, “acoustic effect: none / echo / robot”, “end of sentence pause length: 0.1 to 2.5 seconds”, “reading speed: 10 steps”, “voice Height: 10 levels "," volume: 10 levels "," voice inflection: 10 levels "," silence: yes / no "," high frequency emphasis: yes / no "," nose turbidity: yes / no " "and so on. Here, items that overlap the items in the pronunciation information described later, such as “silenced” and “nasalization”, are also included in the reading format information, but these items in the reading format are for the whole delimited sentence. In the pronunciation information, they are for individual characters in the delimiter. Note that “voiceless” means omitting the utterance of “I” sound and “U” sound sandwiched between unvoiced consonants. For example, when “text” is uttered, the vowel part “c” of “su” is not clearly uttered. Such devoicing is important in generating more natural Japanese synthesized sounds.

  The pronunciation information processing unit 152 is a function for editing processing that changes the setting of pronunciation information by a user's manual operation. A specific example of this function will be described later. As pronunciation information to be edited, “pause insertion / deletion and pause length: 0.1 to 2.5 seconds”, “accent break insertion / deletion and accent rise adjustment”, “nasal muddy sound: yes / No "," Devoiced: Yes / No "," Accent: Yes / No ", etc.

  The time calculation unit 153 functions to obtain a reproduction-related time when the synthesized sound is output from the synthesized sound data edited by the reading information processing unit 151 or the pronunciation information processing unit 152 and to display the time on the editing screen. . The time calculated by the time calculation unit 153 is in units of delimiter sentences. The time required for the synthesized sound reproduction for each delimiter sentence (delimiter sentence reproduction time) and the start time of the synthesized sound reproduction for the text sentence T This is the time until playback starts (separated sentence playback start time). By obtaining such time data and displaying it on the editing screen, for example, when there is a limit on the reading time of a given text sentence T, the optimum reading speed and end pause are related to this time limit. Editing work to set the length can be performed more efficiently. Calculation of time data by the time calculation unit 153 is performed using an audio file created by the audio file creation unit 18.

  The display processing unit 154 has a function of displaying the intermediate synthesized sound data on the monitor of the display / operation unit 14 for editing so that the user can easily capture the output image of the synthesized sound.

  The text storage unit 155 sets the text sentence T divided into delimiter sentences or intermediate synthesized sound data generated by setting pronunciation information and reading format information to the text sentence T as a set of delimiter sentence data 1551 in units of delimiter sentences. Store.

  The phoneme piece storage unit 16 is a functional unit that stores a phoneme piece (phoneme data) from which a synthesized sound is generated. In general, several tens to hundreds of phoneme pieces are prepared for one character of kana. The sound processing unit 17 extracts the corresponding phoneme pieces from the phoneme piece storage unit 16 based on the pronunciation information and the reading format information set in the intermediate synthesized sound data, assigns them to each kana character in the intermediate synthesized sound data, and finally Synthetic sound data is generated. The audio file creation unit 18 creates an audio file such as a WAVE file from the synthesized sound data generated by assigning phonemes in the acoustic processing unit 17. The functions of the phoneme piece storage unit 16, the acoustic processing unit 17, and the audio file creation unit 18 are well-known functions in the field of speech synthesis.

  The sound effect / background sound storage unit 19 stores the sound effect and the background sound that are put on the synthesized sound by editing by the reading information processing unit 151.

  The audio file reproduction unit 20 is a functional part that reproduces the audio file created by the audio file creation unit 18 as a synthesized sound, and the synthesized sound reproduced thereby is output through the speaker 21.

  Hereinafter, the intermediate synthesized sound data and editing thereof will be described with specific examples. FIG. 2 shows an example of intermediate synthesized sound data. The intermediate synthesized sound data in this example is described in an XML document (Extensible Markup Language). The XML document includes “XML declaration (can be omitted)”, “XML type document declaration (can be omitted)”, and “XML document body”. In FIG. 2, “XML type document declaration” is omitted. The first line shows the XML declaration, and the second and subsequent lines are the XML document body. The description enclosed in <clause> to </ clause> in the second and subsequent lines indicates each piece of sentence data (separated sentence data 1551 in FIG. 1). The delimited sentence data is configured to include reading format information, pronunciation information (kana character string with pronunciation symbols and prosodic symbols), and time information related to reproduction. Here, in the case where a sound effect or background sound is put on a synthesized sound, delimiter text data is generated. In the example of FIG. 2, delimiter text data for sound effects is generated. Note that “sentence” in FIG. 2 corresponds to “separated sentence”, which is different from that generally used in the speech synthesis field.

  FIG. 3 shows an example of a reading format information editing screen displayed by the reading information processing unit 151 for editing the intermediate synthesized sound data. A drawing 51 is a sound effect / background sound setting symbol. Sound effects and background sounds can be set in units of delimiter sentences. When a sound effect or background sound is set in a certain delimiter sentence, a sound effect / background sound setting symbol is displayed at the beginning of the delimiter sentence. The setting of sound effects and background sounds can be canceled by an operation such as clicking the sound effect / background sound setting symbol with a mouse. The drawing 52 is a reading format information symbol with a delimited sentence as a unit, and is arranged at the head of the delimited sentence. In the example shown in the figure, the reading format information includes “emotional effect”, “nasalization”, “silence”, “volume”, “reading speed”, “voice pitch”, “voice inflection”, “end of sentence pause length” "Is displayed as a symbol. The display items in this display can be changed. Normally, the items whose settings are often changed in the editing operation are displayed, and the display of the items that often use the standard settings as they are is omitted. To change the reading format, a change operation screen is displayed by, for example, clicking the displayed reading format information symbol with a mouse, and the setting contents are changed on the screen. The drawing 53 shows the text content of the delimiter sentence. The drawing 54 is a sentence delimiter indicating a delimiter sentence delimiter. The process of dividing the text sentence T into delimited sentences is basically performed automatically by the sentence delimiter processing unit 11. However, there are cases where the delimiter becomes too long when delimited by sentence delimiters such as punctuation marks, question marks, and exclamation marks used in the automatic processing. In order to cope with such a case, it is also possible to adjust the length of the delimiter sentence by inserting a sentence delimiter by manual operation on this editing screen. The drawing 55 is a display of the time data obtained by the time calculation unit 153, and displays the delimited sentence reproduction time and the delimited sentence reproduction start time. This display is performed for the delimiter sentence when an operation for designating a delimiter sentence for which time data is to be displayed is performed with a mouse or the like.

  FIG. 4 shows an example of a pronunciation information editing screen displayed by the pronunciation information processing unit 152 for editing intermediate synthesized sound data. On the pronunciation information editing screen, pronunciation information, that is, phonetic symbols and kana character strings with prosodic symbols are displayed in units of delimited sentences. The drawing 61 is an accent symbol indicating that the character immediately below has an accent. The drawing 62 is an accent delimiter indicating an accent delimiter. A drawing 63 is a pause delimiter symbol indicating a pause delimiter. The drawing 64 is an unvoiced sound symbol indicating that the character immediately above is an unvoiced sound. The drawing 65 is a nasal cloud symbol indicating that the character directly above is a nasal cloud. To change the pronunciation settings, delete these symbols or add new ones.

  As described above, the intermediate synthesized sound data can be edited in units of delimiter sentences, so that even if the input text sentence is a long sentence, the image of the actual output synthesized sound when editing the intermediate synthesized sound data Can be grasped more easily, and the efficiency of speech synthesis processing can be improved.

  Hereinafter, the speech synthesis process executed by the speech synthesizer of FIG. 1 will be described. FIG. 5 shows the flow of the processing. In the first step 40 after the start of processing, the sentence delimiter processing unit 11 performs sentence delimiter processing. Specifically, a process is performed in which the sentence of the input text sentence T is separated by a punctuation mark, a question mark, and an exclamation mark sentence to form a set of delimiter sentences. The text sentence T delimited by the delimiter sentence is transferred to the text analysis unit 12 and is stored in the text storage unit 155 as a set of delimiter sentence data 1551. In step 41, the text analysis by the text analysis unit 12 is performed. As described above, this text analysis is syntactic analysis and setting of pronunciation information based on it (generating a kana character string with pronunciation symbols and prosodic symbols). In step 42, the reading format is initially set by the reading information processing unit 151. As described above, this processing is performed by setting a standard reading format for each delimited sentence data, and is performed in a format in which the setting information is displayed as a symbol at the beginning of each delimited sentence. In step 43, the intermediate synthesized sound data obtained through step 42 is displayed on the monitor. In step 44, it is determined whether or not the intermediate synthetic sound data displayed in step 43 needs to be edited. When the final synthesized sound data is created with the pronunciation setting and the reading format setting in the intermediate synthesized sound data being displayed as they are, that is, when editing is not necessary (determination is “No”), the process proceeds to step 46. If editing is required (determination is “Yes”), the process proceeds to step 45.

  In step 45, the reading information and the pronunciation information are edited for the intermediate synthesized sound data. When the editing in step 45 is completed, the process returns to step 43 again to confirm the editing result. In step 46, it is determined whether or not the synthesized sound is reproduced. That is, it is determined whether to continue the acoustic conversion process or to stop the process at the intermediate synthetic sound data stage. When the process is stopped at the stage of the intermediate synthesized sound data and the synthesized sound is not reproduced, the determination is “No”, and the process ends. In the case where the intermediate sound conversion data is subjected to the acoustic conversion process and the sound reproduction is performed, the determination is “Yes”, and the process proceeds to step 47. In step 47, acoustic conversion processing by the acoustic processing unit 17 is performed. That is, based on the pronunciation information and the reading format information set in the intermediate synthesized sound data, the corresponding phoneme piece is extracted from the phoneme piece storage unit 16 and assigned to each kana character in the intermediate synthesized sound data to obtain the final synthesized sound data. Is generated. In step 48, a sound file is generated by the sound file creation unit 18 from the final synthesized sound data generated in step 47. In step 49, the audio file reproduction unit 20 outputs a synthesized sound based on the audio file created by the audio file creation unit 18. The user listens to this output synthesized sound, and if it does not match the image, the process returns to step 44, and the processing from step 44 is repeated until the output synthesized sound matches the image.

  The present invention makes it possible to more easily grasp an image of an actual output synthesized sound when editing synthesized sound data for speech synthesis, thereby improving the efficiency of speech synthesis processing, etc. It has an effect and can be widely used in the field of speech synthesis.

It is a figure which shows the structure of the speech synthesizer by one Embodiment. It is a figure which shows the example of intermediate synthetic sound data. It is a figure which shows the example of a reading format information edit screen. It is a figure which shows the example of a pronunciation information edit screen. It is a figure which shows the flow of the process in the speech synthesis process performed with the speech synthesizer of FIG.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Speech synthesizer 11 Sentence delimiter processing part 12 Text analysis part 15 Synthetic sound data display / editing part 19 Sound effect / background sound storage part 151 Reading information processing part 152 Pronunciation information processing part 153 Time calculation part 1551 Separation sentence data

Claims (10)

In the speech synthesis method including the process of editing the phonetic symbols and the reading format set in the text sentence for generating the synthesized sound from the input text sentence,
Including a sentence delimiter process in which the sentence of the text sentence is delimited by a predetermined sentence delimiter criterion to form a set of delimited sentences having a finite length, and the editing process is performed in units of delimiter sentences obtained by the sentence delimiter process A speech synthesis method characterized by being configured as described above.   The speech synthesis method according to claim 1, wherein a terminal character such as a punctuation mark, a question mark, or an exclamation mark is used as the sentence delimiter.   3. The speech synthesis method according to claim 1, further comprising: a process for displaying a reading format information editing screen and editing the reading format information; and a process for displaying a pronunciation information editing screen and editing the pronunciation information. . In a speech synthesizer that generates synthesized speech from input text sentences,
A speech synthesizer comprising: a sentence delimiter processing unit configured to delimit a sentence of the text sentence according to a predetermined sentence delimiter criterion to obtain a set of delimited sentences having a finite length.   5. The speech synthesizer according to claim 4, wherein a terminal character such as a punctuation mark, a question mark, or an exclamation mark is used as the sentence break reference.   5. A reading information processing unit for editing the reading format information in units of the delimited sentences for intermediate synthesized sound data generated by setting pronunciation information and reading format information in the text sentence. Item 6. The speech synthesizer according to Item 5.   The speech synthesis apparatus according to claim 6, further comprising a pronunciation information processing unit for editing the pronunciation information for the intermediate synthesized sound data in units of the delimiter sentences. In a speech synthesizer that generates synthesized speech from input text sentences,
A synthetic sound data display / editing unit for editing intermediate synthetic sound data generated by setting pronunciation information and reading format information to the text sentence, and the synthetic sound data displaying / editing unit A speech synthesizer, comprising: a time calculation unit that obtains a time related to reproduction when the image is reproduced and output and displays the time on the editing screen.   A sentence delimiter processing unit configured to divide the sentence of the text sentence according to a predetermined sentence delimiter criterion to form a set of delimited sentences of a finite length, and the time calculation unit is a delimiter required for reproducing the synthesized sound for each delimited sentence 9. The speech synthesizer according to claim 8, wherein a sentence reproduction time and a delimiter sentence reproduction start time from a start time of synthetic sound reproduction for the text sentence to a start of reproduction of the delimiter sentence are obtained. In a speech synthesizer that generates synthesized speech from input text sentences,
A sound effect / background sound storage unit for storing the effect sound and background sound to be put on the synthesized sound, and a reading information processing unit for performing an editing process for putting the effect sound / background sound on the synthesized sound A speech synthesizer characterized by the above.

Images