JPH02251998A - Voice synthesizing device - Google Patents

Abstract

PURPOSE:To understand the contents of a voice which is reproduced reversely by performing reverse reproduction in meaningful units and fast reading operation in the reverse reproduction. CONSTITUTION:The voice synthesizing device which synthesizes a voice in accordance with an input character symbol string is provided with a reading unit division part 33 and a reverse reading control part 38 and the reading unit division part 33 divides the input character symbol string into meaningful units in accordance with the reverse reproduction indication signal from the reverse reading control part 38 and a specification signal specifying the kind of the meaningful units. In the divided character symbol strings, the character strings are outputted in the forward direction along the time base to synthesize a voice and groups of the divided character symbol strings are outputted in the opposite array on the time base to synthesize a voice. Consequently, the reverse reproduction can be performed in the meaningful units and the contents of the reversely reproduced voice can easily be understood.

Description

DETAILED DESCRIPTION OF THE INVENTION <Industrial Application Field> The present invention relates to a speech synthesizer that generates rule-based synthesized speech.

<Prior Art> Conventionally, when a speech synthesizer reproduces audio in reverse, audio data recorded in a storage unit is reproduced in the reverse direction with respect to the time axis.

Furthermore, when a text-to-speech synthesizer that performs speech synthesis from an arbitrary string of characters and symbols is used to reproduce speech in reverse, it is performed as follows.

That is, in FIG. 3, a character and symbol string (for example, a sentence containing Japanese, kanji, and kana) is input to a character and symbol string input section 1 and sent to a character and symbol string analysis section 2. Then, the character/symbol string analysis unit 2 analyzes the morphemes (words) of the input character/symbol string as will be described in detail later.

Performs syntactic analysis, semantic analysis, etc. In order to control prosody, the synthesized speech parameter generation unit 3 generates accents and pauses in clauses and exhalation paragraphs when words are chained based on the accent and syntactic structure of each word identified by the morphological analysis in the character symbol string analysis unit 2. Configure settings. Further, parameter time series of duration, pitch pattern, power pattern, and phonetic feature parameters (partial autocorrelation coefficient, line spectrum pair, formant, etc.) for the synthesis unit corresponding to the uttered voice are obtained. Then, the speech synthesis section 4 generates an actual synthesized speech waveform based on the parameter time series for speech synthesis, and outputs it from the synthesized speech output section 5.

FIG. 4 is a more detailed block diagram of the character symbol string analysis section 2. The character symbol string analysis section 2 includes a morphological analysis section 21, a syntactic analysis section 22. It is composed of a meaning analysis section 23 and a dictionary (a kanji dictionary, a symbol dictionary, a meaning dictionary, a word dictionary, a connection prohibition dictionary, etc.) 24. The morphological analysis section 21 is a character symbol string input section! Morphological analysis is performed on the input character string using a dictionary (Kanji dictionary, symbol dictionary, word dictionary, etc.) 24 to identify words and obtain grammatical information such as parts of speech and accents of the identified words. The syntactic analysis unit 22 determines the syntax of the word identified by the morphological analysis unit 2, using a dictionary (such as a connection prohibition dictionary) 24.

Further, the meaning analysis section 23 determines the meaning of the input string of characters and symbols using a dictionary (semantic dictionary, etc.) 24. However, the character symbol string analysis section 2 is not necessarily the morphological analysis section 21. Syntax analysis unit 22. It is not necessary to configure the semantic analysis unit 23 and the dictionary 24. - That is, the morphological analysis unit 21. It may be configured by only the syntactic analysis unit 22 and the dictionary 24, or the morphological analysis unit 21 and the dictionary 24.

When the above-mentioned text-to-speech synthesizer performs fast reading in forward reading, the synthesized speech parameter generation unit 3 should reduce the number of pauses set, shorten the duration of the pauses, or This is done by shortening the duration of the synthesis unit.

<Problems to be Solved by the Invention> However, in the above-mentioned conventional speech synthesis device, when reproducing speech in reverse, recorded speech data is read out and reproduced in reverse with respect to the time axis, so phonemes, syllables, etc. is not generated correctly. Therefore, there is a problem in that it is extremely difficult to understand the content of the audio that is played back in reverse.

Furthermore, in the above-mentioned conventional text-to-speech synthesis device, when reproducing speech in reverse, reverse reproduction is performed in units of syllables, so it is possible to generate synthesized speech that is correctly reproduced in reverse in units of syllables. However, since the unit of reproduction is a syllable, it cannot be said that synthesized speech is generated in units with color. Therefore, there is a problem in that it is extremely difficult to understand the content of the audio that is played back in reverse. Therefore, if you perform fast reading with reverse playback (hereinafter referred to as fast reading during reverse playback),
This makes it even more difficult to understand the content of the output audio.

SUMMARY OF THE INVENTION Accordingly, an object of the present invention is to provide a speech synthesis device that can understand the contents of reversely reproduced speech by performing reverse reproduction and fast reading during reverse reproduction in meaningful units.

<Means for Solving the Problems> In order to achieve the above object, a first invention provides a speech synthesis device that generates speech synthesis parameters from input character strings and performs speech synthesis based on the speech synthesis parameters. When performing reverse playback, the input character string is divided into meaningful units such as sentences or phrases, and within this divided character symbol string, the character symbol string is played back in the forward direction with respect to the time axis. and a reading unit division section that outputs the set of character symbol strings in reverse order with respect to the time axis, an instruction signal that instructs the reading unit division section to perform reverse playback, and the above-mentioned meaning. If the input character string is to be played back in reverse by a reverse reading control unit that outputs a designation signal specifying the type of unit in which the character string is divided into meaningful units by the reading unit division unit. As a set of character symbol strings, speech is synthesized in the reverse direction with respect to the time axis, but within the divided character symbol strings, speech is synthesized in the forward direction with respect to the time axis. There is.

Further, the second invention provides a speech synthesis device that generates speech synthesis parameters from an input character symbol string and performs speech synthesis based on the speech synthesis parameters, and when performing reverse playback, The symbol string is divided into meaningful units such as sentences or clauses, and within this divided character symbol string, the character symbol string is output in the forward direction with respect to the time axis, and the time axis of this character symbol string set is output. a reading-out unit division section that outputs the arrangement order of the unit in reverse order; and a reverse unit that outputs an instruction signal instructing execution of reverse playback and a specification signal specifying the type of the meaningful unit to the reading-out unit division section. a reading control section, a character symbol string extraction section that extracts a character symbol string required by the operator from the input character string when performing fast reading, and a character symbol string extraction section for the character symbol string extraction section. , a fast reading control unit that outputs an instruction signal for instructing the execution of fast reading and a specification signal for specifying the type of character symbol string to be extracted, and when performing reverse reproduction of the input character symbol string, the above As a set of character symbol strings divided into meaningful units by the reading unit division section, speech is synthesized in the reverse direction with respect to the time axis, but within the above divided character symbol strings, voice is synthesized in the order of the time axis. The present invention is characterized in that, while the speech is synthesized in the direction, when performing fast reading, the speech is synthesized based only on the character and symbol strings extracted by the character and symbol string extraction section.

<Operation> In the first invention, when the instruction signal instructing the reading unit dividing unit to perform reverse reproduction by the reverse reading control unit h) and the designation signal specifying the type of meaningful unit are input,
The readout unit division section divides the input character symbol string into meaningful units of the type specified by the designation signal, such as sentences or clauses. Within this divided character and symbol string, the character and symbol strings are output in the forward direction with respect to the time axis, and the arrangement order of the set of character and symbol strings with respect to the time axis is reversed and output. Therefore, when performing reverse playback of the input character string, the voice is synthesized in the reverse direction with respect to the time axis as a set of character symbol strings divided into meaningful units by the reading unit division section. However, within the divided character and symbol strings, speech is synthesized in the forward direction with respect to the time axis.

In the second invention, when an instruction signal instructing reverse playback and a designation signal specifying a type of meaningful unit are input from the reverse reading control unit to the reading unit dividing unit, the reading unit dividing unit , the input character symbol string is divided into meaningful units of the type specified by the above-mentioned designation signal, such as sentences or clauses. Within this divided character and symbol string, the character and symbol strings are output in the forward direction with respect to the time axis, and the arrangement order of the set of character and symbol strings with respect to the time axis is reversed and output.

Therefore, when performing reverse playback of the input character string, the voice is synthesized in the reverse direction with respect to the time axis as a set of character symbol strings divided into meaningful units by the reading unit division section. However, within the divided character and symbol strings, speech is synthesized in the forward direction with respect to the time axis.

On the other hand, when an instruction signal instructing quick reading from the fast reading control unit and a designation signal specifying the type of character and symbol string to be extracted are input to the character and symbol string extraction unit, the character and symbol string extraction unit A character and symbol string of the type designated by the designation signal is extracted from the input character and symbol strings. Therefore, when performing quick reading of the input character string, speech is synthesized based only on the character string extracted by the character string extraction section.

<Example> Hereinafter, the present invention will be explained in detail with reference to illustrated embodiments.

FIG. 1 is a block diagram of an embodiment according to the first invention.

In FIG. 1, the character symbol string (
For example, a sentence containing Japanese, kanji, and kana is input and sent to the character/symbol string analysis section 32. Then, the character symbol string analysis unit 32 performs morphological analysis, syntactic analysis, semantic analysis, etc. of the input character symbol string. Although omitted in FIG. 1, the character/symbol string analysis section 32 includes the morpheme analysis section 21.1 shown in FIG. It is composed of a syntactic analysis section 22, a semantic analysis section 23, and a dictionary 24.

On the other hand, a control signal for instructing reverse reproduction is input to the control signal input section 37 and sent to the reverse reading control section 38. Then, the reverse reading control unit 38 analyzes the input control signal and inputs various reverse reproduction commands to the reading unit division unit 33. Here, the various reverse playback commands include a control command for performing/not performing reverse playback, a control command for specifying the type of unit for performing reverse playback when performing reverse playback, and the like.

The reading unit dividing section 33 uses the analysis result of the character symbol string previously analyzed by the character symbol string analyzing section 32 (i.e., the accent and syntactic structure of each word) and the above input from the reverse reading control section 38. According to various reverse play commands,
When performing reverse playback, the analysis result of the character/symbol string is divided into meaningful units required by the operator. and,
The divided character-symbol string analysis result sets are arranged in reverse order with respect to the time axis and sent to the synthesized speech parameter generation section 34. On the other hand, within the above-mentioned divided character symbol string analysis results, the character symbol and string analysis results are sent to the synthesized speech parameter generation section 34 in the forward direction with respect to the time axis. By doing this, it is possible to perform reverse playback in meaningful units. On the other hand, when reverse playback is not performed (that is, when a control command for not performing reverse playback is input from the reverse reading control section 38), the reading unit division section 33 does nothing.

In order to control the prosody, the synthesized speech parameter generation section 34 uses the accents and syntactic structures of each word identified by the character symbol string analysis section 32 to control the accents and pauses of clauses and exhalation paragraphs when words are chained. Configure settings. Further, parameter time series of duration, pitch pattern, power pattern, and phonetic feature parameters (partial autocorrelation coefficient, line spectrum pair, formant, etc.) for the synthesis unit corresponding to the uttered voice are obtained. Then, the speech synthesis section 35 generates an actual synthesized speech waveform based on the parameter time series for speech synthesis, and generates the synthesized speech output section 3.
It is output from 6.

In this case, the synthesized speech parameter generation unit 34 sets the accents and pauses of phrases and exhalation paragraphs in the order of the accent and lateral sentence structure of each word sent from the reading unit division unit 33, and sets parameters for the synthesis unit. Since a time series is generated, the synthesized speech outputted from the synthesized speech output section 36 is played back in meaningful units, and synthesized speech whose contents can be understood is output.

The following units are meaningful units that are divided by the reading unit dividing section 33 during reverse playback.

(A) Sentence unit (Punctuate unit) This division by sentence unit is a division in which the sentence analyzed by the character symbol string analysis unit 32 is used as a unit.

At that time, the structure of the character symbol string analysis section 32 is the same as that of the morphological analysis section 21. Syntax analysis unit 22. It is not necessary to provide both the semantic analysis unit 23 and the dictionary 24, and the morphological analysis unit 2
! and the dictionary 24 alone. Also,
The input character string may be divided by extracting the period by paying attention to the period indicating the break between sentence units, and regarding the character string between the extracted periods as a sentence. In this case, the character/symbol string analysis section 32 is not required for sentence-by-sentence division.

([() Clause unit This clause unit division is a division based on the clause analyzed by the character symbol string analysis unit 32.

At that time, the structure of the character symbol string analysis section 32 is the same as that of the morphological analysis section 21. Syntax analysis unit 22. It is not necessary to provide both the semantic analysis unit 23 and the dictionary 24, and the morphological analysis unit 2
1. It may be configured by only the syntax analysis unit 22 and the dictionary 24.

(C) Punctuation mark unit Punctuation mark unit division is a division based on punctuation marks analyzed by the character symbol string analysis section 32.

In this case as well, the punctuation marks in the character symbol string may be extracted in the same manner as the period unit described above, and the character symbol string may be divided by regarding the character symbol string between the extracted punctuation marks as one unit. In this case as well, the character/symbol string analysis unit 32 is not required for dividing into punctuation mark units.

In this way, in the first invention, the control signal input section 3
According to various reverse reproduction commands outputted from the reverse reading control section 38 based on control signals from 7, the input character symbol string is divided into meaningful units by the reading word division section 33, and the divided characters are As a set of symbol string analysis results, speech is synthesized in the opposite direction to the time axis and output. On the other hand, within the above-mentioned divided character/symbol string analysis results, speech is synthesized and output in the forward direction with respect to the time axis. Therefore, reverse playback can be performed in meaningful units, and synthesized speech whose content can be understood during reverse playback can be output.

FIG. 2 is a block diagram of an embodiment according to the second invention.

In FIG. 2, the character symbol string (
For example, a sentence containing Japanese, kanji, and kana is input and sent to the character/symbol string analysis section 42. Then, the character symbol string analysis unit 42 performs morphological analysis, syntactic analysis, semantic analysis, etc. of the input character symbol string. Here, the character symbol string analysis section 42 is the morphological analysis section 2! in FIG. 4, similar to the first invention. .

Syntax analysis unit 22. It is composed of a semantic analysis section 23 and a dictionary 24 (also in this case, it is not necessary to include all of the morphological analysis section 21, syntactic analysis section 22, semantic analysis section 23, and dictionary 24, and they may be omitted as necessary. good).

On the other hand, a control signal for instructing reverse reproduction or fast reading is input to the control signal input section 48, and is separated into a control signal for instructing reverse playback and a control signal for instructing fast reading. A control signal instructing reverse playback is input to the reverse reading control section 49, while a control signal instructing fast reading is input to the fast reading control section 50.

Then, the reverse reading control section 49 analyzes the input control signal and reads out various reverse reproduction commands to the unit dividing section 4.
Output to 3. Here, the various reverse playback commands include a control command for performing/not performing reverse playback, a control command for specifying the type of unit for performing reverse playback when performing reverse playback, and the like.

Further, the fast reading control unit 50 analyzes the input control signal and outputs various fast reading commands to the character symbol string extraction unit 44 and the synthesized speech parameter generation unit 45. Here, the various fast reading commands output to the character symbol string extracting section 44 include a control command to perform/not perform fast reading, a control command to specify a fast reading method when performing fast reading, etc. . In addition, the various quick reading commands output to the speech synthesis parameter generation section 45 are
5, the control command for performing/not performing fast reading according to the above-mentioned prior art, and the pause setting method, pause duration, and uttered voice when performing fast reading according to the prior art using the speech synthesis parameter generation unit 45. This is a command to set the duration for the corresponding synthesis unit, etc.

The reading unit division section 43 executes reverse reproduction according to the analysis result of the character symbol string previously analyzed by the character symbol string analysis section 42 and the various reverse reproduction commands inputted from the reverse reading control section 49. At this time, the analysis result of the character symbol string is divided into meaningful units required by the operator. and,
The divided character and symbol string analysis result sets are arranged in reverse order with respect to the time axis and sent to the character and symbol string extraction unit 44.

On the other hand, within the divided character and symbol string analysis results, the character and symbol string analysis results are sent to the character and symbol string extraction section 44 in the forward direction with respect to the time axis. By doing this, it is possible to perform reverse playback in meaningful units. On the other hand, when reverse playback is not performed, the reading unit division section 43 does nothing. In this case, the units for performing reverse playback are (A) sentence units (full stop units), (B) phrase units, and (
C) Use punctuation marks, etc.

When performing quick reading, the character symbol string extraction section 44 extracts the operation sound or the required character symbol string analysis results from among the analysis results of the input character symbol strings from the quick reading system gjJ section 50. The reading method is selected by a method based on a control command specifying the reading method and outputted to the synthesized speech parameter generation unit 45. In that case, the following methods are available as the fast reading method specified by the control command specifying the fast reading method. That is, a simple method is to select the above-mentioned reverse playback unit according to the speaking speed. For example, if the reverse playback unit is a sentence unit and you want to double the reading speed,
Every other sentence, which is a unit of reverse reproduction, is output. Further, as a method for obtaining speech synthesis whose content is easier to understand, there is a method of outputting synthesized speech consisting only of a subject and a predicate based on the syntactic analysis results and semantic analysis results obtained by the character symbol string analysis section 42.

Here, when a control signal instructing reverse playback and a control signal instructing fast reading are input to the control signal input section 48 to instruct fast reading during reverse playback, the reading unit dividing section 43 extracts character symbol strings. section 44 is driven. First, the reading unit dividing unit 43 converts the input character symbol string analysis result into the above (A
), (B) or (C), and the arrangement of the divided character/symbol string analysis result set with respect to the time axis is reversed and sent. Next, the character-symbol string extraction section 44 extracts the character-symbol string analysis results required by the operator from the set of character-symbol string analysis results sent in reverse order from the reading unit dividing section 43, and extracts the result of the character-symbol string analysis required by the operator, and generates synthesized speech parameters. It is output to the generation section 45.

In order to control the prosody, the synthesized speech parameter generation section 45 generates accents of clauses and exhalation paragraphs when words are chained based on the accent and syntactic structure of each word identified by the character symbol string analysis section 32. Set the pose. Further, parameter time series of duration, pitch pattern, power pattern, and phonetic feature parameters (partial autocorrelation coefficient, i-spectrum pair, formant, etc.) for the synthesis unit corresponding to the uttered voice are obtained. Then, the speech synthesis section 46 generates an actual synthesized speech waveform based on the parameter time series for speech synthesis, and the synthesized speech output section 46 generates an actual synthesized speech waveform.
It is output from 7.

Here, when the fast reading control unit 5o outputs a fast reading command according to the prior art to the synthesized speech parameter generation unit 45, the pause setting method, pause duration, synthesis unit duration, etc. are changed. be done. In that case, by using this conventional fast reading process together with the character/symbol string selection fast reading process using the character/symbol string extraction f$44 described above, it is possible to understand the uttered content of the reverse playback in a shorter time. You will be able to do this.

In this way, in the second invention, the control signal input section 4
8, the analysis result of the input character symbol string is divided into meaningful units by the reading word division section 43, and the divided As a set of character symbol string analysis results, speech is synthesized and output in the reverse direction with respect to the time axis. On the other hand, within the above-mentioned divided character/symbol string analysis results, speech is synthesized and output in the forward direction with respect to the time axis.

Therefore, reverse playback can be performed in meaningful units, and synthesized speech whose content can be understood during reverse playback can be output.

Further, in accordance with various fast reading commands output from the fast reading control unit 50 based on the control signal from the control signal input unit 37, the character symbol string extraction unit 44 extracts the operator from the input character symbol string analysis results. Select the required character/symbol string analysis results, synthesize and output speech. Therefore, rapid reading can be performed in meaningful units, and synthesized speech that can be understood during rapid reading can be output.

Furthermore, when performing fast reading during reverse playback, the character and symbol string extraction section 44 selects a set of character and symbol string analysis results that are divided into meaningful units by the reading unit division section 43 and output in reverse order. , selects the character/symbol string analysis results required by the operator, synthesizes speech, and outputs the result. therefore,
Fast reading during reverse playback can be performed in meaningful units, and synthesized speech that can be understood when fast reading during reverse playback can be output.

The above embodiment describes a text-to-speech synthesis device. However, the present invention is not limited to this, and includes a device that synthesizes speech based on stored speech data (for example, an automatic translation device for speech output).
May be used for.

<Effects of the Invention> As is clear from the above, the first invention provides the speech synthesis device that performs speech synthesis based on input character symbol strings with a reading unit division section and a read-through control section. The reading unit dividing unit divides the input character symbol string into meaningful units according to a reverse playback instruction signal from the reading control unit and a designation signal specifying the type of meaningful unit,
Within this divided character symbol string, the character symbol strings are output in the forward direction with respect to the time axis to synthesize speech, and the arrangement of the divided character symbol string sets with respect to the time axis is reversed and output. Since the voices are synthesized, it is possible to perform reverse playback in meaningful units, and it is possible to easily understand the content of the voice reversely reproduced by the voice synthesizer of the present invention.

Further, the second invention provides a speech synthesis device that performs speech synthesis based on an input character symbol string, and includes a reading unit division section, a reading control section, a single character symbol extraction section, and a fast reading control section. In accordance with the instruction signal for reverse playback from the reading control section and the specification signal specifying the type of meaningful unit, the reading unit division section performs the reading unit dividing section in the order of the time axis within the character symbol string divided into meaningful units. In addition to synthesizing speech by outputting strings of characters and symbols in the same direction, the arrangement of the divided strings of characters and symbols with respect to the time axis is reversed and output to synthesize speech. Reverse reproduction can be performed, and the content of the voice reproduced in reverse by the speech synthesizer of the present invention can be easily understood.

Furthermore, in accordance with the quick reading instruction signal from the quick reading control unit and the specified signal specifying the type of character symbol string to be extracted, the character symbol string required by the operator is extracted from the input character symbol string. Since the speech is synthesized based on one string of extracted character symbols, when performing fast reading during reverse playback, the meanings sent from the reading unit division part in the reverse arrangement with respect to the time axis can be used. It is possible to extract a necessary character symbol string from a character symbol string divided into certain units by the character symbol extracting unit and synthesize speech, and the speech synthesis device of the present invention can quickly read the character symbol string during reverse playback. Can easily understand the contents of spoken voices.

[Brief explanation of drawings]

FIG. 1 is a block diagram of a speech synthesis device according to the first invention,
FIG. 2 is a block diagram of a speech synthesis device according to a second invention,
FIG. 3 is a block diagram of a conventional text-to-speech synthesis device, and FIG. 4 is a more detailed block diagram of the character symbol string analysis section in FIGS. 1, 2, and 3. 31.41...Character symbol string input section, 32.42...Character symbol string analysis section, 33.43...Reading unit division section, 44...Character symbol string extraction section, 34.45... Synthetic speech parameter generation unit, 35.4
6... Speech synthesis section, 36.47... Synthesized speech output section, 37.48... Control signal input section, 38.49... Reverse reading control section, 50... Fast reading control section.

Claims (2)

[Claims] (1) In a speech synthesis device that generates speech synthesis parameters from an input character symbol string and performs speech synthesis based on these speech synthesis parameters, when performing reverse playback, the input character symbol string is It divides into meaningful units such as clauses, and within these divided character and symbol strings, the character and symbol strings are output in the forward direction with respect to the time axis.
A reading unit division section that outputs the set of character symbol strings in reverse order with respect to the time axis, an instruction signal that instructs the reading unit division section to perform reverse playback, and the type of unit with the above-mentioned meaning. If the above-mentioned input character string is to be reproduced in reverse by a reverse reading control unit that outputs a designation signal specifying the A speech synthesis device characterized in that, although the speech is synthesized in the reverse direction with respect to the time axis, the speech is synthesized in the forward direction with respect to the time axis within the divided character symbol string. (2) When performing reverse playback in a speech synthesis device that generates speech synthesis parameters from an input character symbol string and performs speech synthesis based on these speech synthesis parameters, the input character symbol string can be converted into a sentence or It divides into meaningful units such as clauses, and within these divided character and symbol strings, the character and symbol strings are output in the forward direction with respect to the time axis.
A reading unit division section that outputs the set of character symbol strings in reverse order with respect to the time axis, an instruction signal that instructs the reading unit division section to perform reverse playback, and the type of unit with the above-mentioned meaning. a reverse reading control unit that outputs a designation signal that specifies a character symbol string; and a character symbol string extraction unit that extracts a character symbol string required by the operator from the input character symbol string when performing fast reading; The input character string is provided with a fast reading control unit that outputs an instruction signal instructing the execution of fast reading and a designation signal specifying the type of the character symbol string to be extracted to the character symbol string extraction unit. When performing reverse playback, the voice is synthesized in the reverse direction with respect to the time axis as a set of character symbol strings divided into meaningful units by the above-mentioned reading unit division section, but the divided character symbol strings When reading quickly, the voice is synthesized in the forward direction of the time axis, while in the case of fast reading, the voice is synthesized based only on the character symbol string extracted by the character symbol string extraction section. A speech synthesis device characterized by the following.