JPH0229797A - Text voice converting device - Google Patents

Abstract

PURPOSE:To voice texts of respective languages by a simple system constitution so that a user easily understands by performing conversion into voicing and phoneme symbols for a 2nd predetermined language separately by a character string voicing and phoneme symbol converting means for a 1st language and then composing a voice by a voice composing means. CONSTITUTION:A character string from an input device 11 is inputted to the language processing means 3 of the voicing and phoneme symbol generating means 1 of the test voice converting device and processed by the language processing means 3 according to a program stored in a main control part 13. This processing means 3 converts the character string into a voicing and phoneme symbol sequence for the 1st language written in a document storage part 14 and a voicing and phoneme symbol converting means 4 converts the voicing and phoneme symbol sequence converted by the means 3 into voicing and phoneme symbols for the 2nd language (English or German). Then the acoustic synthesizing means 22 composes a voice and outputs texts of many languages by the simple system constitution with the voice which is easily understood by the user.

Description

[Detailed Description of the Invention] [) Already Required] An object of the present invention is to output multilingual speech that is easy for users to understand without increasing the size of the device, regarding a text-to-speech conversion device capable of converting text in multiple languages into speech. , and the string of characters written in the first language is translated into a second language whose pronunciation is different from that of the first language.
The present invention is configured to include a pronunciation/prosodic symbol generation means for converting into a pronunciation/prosodic symbol string of the language, and a speech synthesis means for synthesizing speech based on the pronunciation/prosodic symbol.

[Industrial application field]

The present invention relates to a text-to-speech conversion device that converts text to speech, and particularly relates to a text-to-speech conversion device that can convert text in multiple languages into speech.

In recent years, it has been used as a means of vocalization for disabled people with speech disorders,
2. Description of the Related Art Text-to-speech conversion devices are increasingly being used as reading devices for assisting the blind, text-to-speech devices for electronic mail, and even for educational purposes.

A text-to-speech conversion device is a device that automatically converts sentences used in daily life into speech and outputs it. For example, when a Japanese sentence is entered from a keyboard, the input text is output as speech. .

Sentences input from the keyboard are finally separated into synthesis units, which are units of pronunciation, based on a Japanese dictionary and speech synthesis rules, and acoustic parameters are created for each synthesis unit.

Then, a speech waveform is synthesized by a speech synthesizer based on the acoustic parameters.

The synthesis units mentioned above include the current phoneme (C1■), kana syllable (CV), VCV, CVC, and a combination of two phonemes (
ctyad, dipho with CV, VC) as the basic unit
ne etc. are being considered. C indicates a consonant, ■ indicates a vowel.

For example, for "Kamome", if the synthesis unit is the phoneme (C1■), /ni/, /a/, /m/, 10/, /m/, /e
If / is a kana syllable, / k a /, /mO/,
/m e /. Also, if it is VCV,
/ ka /, / a m o /, / o m e
/, if CVC / k a m /, / m
It is decomposed into o m /, / m e /.

[Conventional technology]

In recent years, with internationalization, international information exchange has become more active year by year, and text-to-speech conversion devices are also required to have the ability to support not only foreign languages but also multiple languages.

Conventionally, as a text-to-speech conversion device capable of converting speech in multiple languages, there has been one that uses phonemes (C, V) as synthesis units as shown in FIG.

In the figure, 61 is a character string input device such as a keyboard, and the text input from the input device 61 is converted into a predetermined character code such as a JIS code by an input control section 62, and then sent via the input control section 62. It is sent to the main control section 63 and stored in the text storage section 64. The sentences stored in the sentence storage section 64 are read out by the language processing section 65 and decomposed into phoneme units.

Furthermore, the language processing unit 65 inserts prosodic information between the phoneme strings to express the accent and intonation peculiar to the language according to the written language, and the chronological phoneme information,
The prosody information is output to the acoustic processing section 66. Sound processing section 66
creates acoustic parameters (control parameters for speech synthesis) while referring to the phoneme file 67 based on input phoneme information, prosody information, etc., and generates speech using a built-in speech synthesizer (not shown). The synthesized voice is output to the outside via the speaker 68. Note that the phoneme file 67 stores spectral parameters of each phoneme, and the acoustic processing unit 66 stores the phoneme file 67.
The spectral parameters corresponding to each phoneme are read out and the speech is synthesized. In this way, since the phoneme, which is the smallest unit of speech, is used as the unit of synthesis, it is possible to produce speech in almost any language.

However, in order to perform speech synthesis using phonemes as synthesis units, a lot of phonetic and acoustic knowledge is required, making it extremely difficult to achieve clear speech. This has caused problems such as difficulty for users to understand when the language is in another language.

An attempt has been made to solve these problems and obtain practical sound quality by using a speech conversion device using syllables as synthesis units, as shown in FIG. In the same figure, input value r1171,
The input control section 72, the main control section 73, the text storage section 74, and the speaker 78 are the same as those shown in FIG. 10 with the same names, so a detailed explanation will be omitted.

This voice conversion device is indicated by broken lines 80, 80, .
As shown in the box 80, a dedicated language processing unit 8 is provided for each language.
1. A sound processing section 82 and an audio file 83 are provided.

The input character example is converted by the language processing section 81 into a string of phonetic symbols and prosodic symbols corresponding to the written language, and the acoustic processing section 82 synthesizes waveforms using syllables as synthesis units. The audio file 83 stores spectral parameters for each syllable of each language, and the audio processing unit 82
performs waveform synthesis while referring to the audio file 83. In this way, by performing waveform synthesis using syllables as synthesis units, the sound becomes clearer and it is possible to obtain practical sound quality.

[Problem to be solved by the invention]

As mentioned above, when phonemes are used as synthesis units, the acoustic processing unit 66 can be shared by all languages, making it possible to be smaller and lower in cost. There was a practical problem.

Furthermore, if syllables are used as synthesis units, practical sound quality can be obtained, but it is necessary to provide multiple language processing units 81, acoustic processing units 82, and audio files 83 for each language.
The disadvantages are that the device becomes huge and requires a lot of installation space, and the cost is also high.

An object of the present invention is to output multilingual audio that is easy for users to understand without increasing the size of the device.

[Means to solve the problem]

FIG. 1 is a diagram showing the principle of the present invention.

The pronunciation/prosodic symbol generation means 1 converts a character string (text) in a first language written in a multilingual language such as English, German, or French into a second language for pronunciation different from the first language. Convert to pronunciation/prosodic symbol string. For example, if the second language is Japanese, a string of characters written in another language is converted into a string of pronunciation and prosodic symbols for Japanese that expresses Japanese pronunciation and intonation. .

The speech synthesis means 2 synthesizes speech in the optimal synthesis unit for the second language based on the pronunciation/prosodic symbol string for the second language generated by the pronunciation/prosodic symbol generation means 1.

The pronunciation/prosodic symbol generation means 1 includes, for example, a language processing means 3 that converts a character string into a pronunciation/prosodic symbol string for the first language in which it is written, and the pronunciation/prosodic symbol generated by the language processing means 3. It has pronunciation/prosodic symbol converting means 4 for converting the string into a string of pronunciation/prosodic symbols for the second language.

For example, if the character string is written in English and the second language is Japanese, the character string is once converted into a pronunciation/prosodic symbol string for English by the language processing means 3. Then, the pronunciation/prosodic symbol string for English is converted into a pronunciation/prosodic symbol string for Japanese by the pronunciation/prosodic symbol converting means 4.

[For production]

In the present invention, a character string of a first language written in multiple languages is converted into a pronunciation/prosodic symbol string for a second language determined by f by the pronunciation/prosodic symbol conversion means 1, and the speech synthesis means 2, speech synthesis is performed in appropriate synthesis units.

Therefore, for example, a sentence written in English is read out with Japanese-like pronunciation and intonation. For this reason,
If the predetermined language is the user's native language, the user will be able to easily understand the content of the text.

In addition, since only one voice synthesizing means 2 needs to be provided, it is possible to reduce the size and cost.

〔Example〕

Examples will be described below with reference to the drawings. FIG. 2 is a system configuration diagram of one embodiment.

The input device 11 is a device for inputting a character string of a sentence using a keyboard or the like, and the character string inputted from the input device 11 is converted into a character code such as a JTS code by the input control section 12. The data is then sent to the main control section 13, which consists of a microprocessor, etc., and is written into the text storage section 14, which consists of an external storage device such as a floppy disk or hard disk.

The main control section 13 is composed of a microprocessor, etc., and executes a program stored in a ROM (read only memory), not shown, to control the input control section 12, the language processing section 15, and the audio processing section 17. It's in control.

The language processing unit 15 converts character strings of Japanese sentences into a time series of pronunciation and prosodic symbols for Japanese. It has an English language processing unit 15b that converts it into a time series, and a German language processing unit 15C that converts the character string of a German sentence into a time series of pronunciation and prosodic symbols for German.
The time-series pronunciation and prosodic symbols output from the Japanese language processing unit 15a are directly output to the acoustic processing unit 17.

In addition, the time series phonetic symbols and prosodic symbols output from the English language processing section 15b and the German language processing section 15c are as follows:
The signals are sent to pronunciation/prosodic symbol converters 16b and 16c, respectively.

English pronunciation/prosodic symbol converter 16b, German pronunciation/
The prosodic symbol conversion unit 16c converts the English pronunciation/prosodic symbol string and the German pronunciation/prosodic symbol string into Japanese pronunciation/prosodic symbol strings for pronunciation and intonation in each language. Convert it into a string of symbols and send it to the acoustic processing unit 17
This is what is output to.

The acoustic processing unit 17 is a block that synthesizes a speech waveform based on input time-series Japanese phonetic symbols and prosodic symbols, and the details thereof will be described later.

The audio file 18 is a file that stores spectral parameters, which are information about the spectral envelope of a synthesis unit, and sound source parameters, which are information about sound intensity. The sequence is divided into synthesis units, and the spectral parameters and sound source parameters of the synthesis units are read out from the audio file 18 using the divided synthesis units as keys to perform speech synthesis.

The audio waveform synthesized by the audio processing section 17 is outputted via the speaker 19.

FIG. 3 is a block configuration diagram showing an example of the configuration of the language processing section 15. As shown in FIG. Japanese language processing unit 15. Both the English language processing section 15b and the German language processing section 15 have a block configuration as shown in FIG. 1
The case 5a will be explained as an example.

The preprocessing unit 21 reads out a text string in predetermined length units from the text storage unit 14 under the control of the main control unit 13, and separates the text using punctuation marks ",", ",".

The word identification unit 22 searches a word dictionary 23 that stores Japanese words, and divides the sentence into word units. The word dictionary 23 stores information indicating the part of speech, pronunciation symbol, accent position of a word, etc., and the part of speech information, pronunciation symbol, accent position information, etc. regarding each word are also read out. The syntactic analysis unit 24 divides the sentence into clauses based on word part-of-speech information, and creates syntactic information such as clause relationships.

The pronunciation analysis unit 25 considers the phonetic symbols of the preceding and following words, and
Correct phonetic symbols or set phonetic symbols for words such as proper nouns and numerals that are not registered in the word dictionary.

The prosody analysis unit 26 also calculates intonation and pause length based on the syntactic information created by the syntax analysis unit 24, and calculates the accent position of a word followed by a compound word or adjunct based on predetermined rules. search, correction of prosodic symbols,
Perform generation.

Through the series of operations described above, the Japanese language processing unit 15
A Japanese sentence is converted into a string of pronunciation and prosodic symbols for Japanese and output to the audio processing section 17.

English language processing section 15b, German language processing section 15c
, the word dictionary 23 stores words for English and German, and the words for each language are divided into cylinders based on English sentence structure rules and German sentence structure rules. A string of phonetic symbols and prosodic symbols is created.

FIG. 4 shows phoneme symbols for phonemes created by the English language processing unit 15b.

Linguistically, English phonetic symbols are expressed by the phoneme symbols 31 shown in the same figure, but in this embodiment, each phoneme symbol 31 is expressed as an alphabet (ARP 8B to T) Match symbol 32,
This alphabesoto symbol 32 is used to represent the English pronunciation symbol. For example, the phoneme symbol 31 of "e" is represented by the alphabetic symbol 32 rEYj, and the phoneme symbol 31 of raJ is represented by the alphabetic symbol 32 of rAAJ. The word dictionary 23' in the English language processing section 15b stores alphabet symbols 32 as phonetic symbols.

Next, FIG. 5 is a block diagram of an embodiment of the English pronunciation/prosodic symbol conversion section 15b.

The phoneme conversion table is a table that stores Japanese phonetic symbols 45 corresponding to English phonetic symbols (Alpabesoto symbol 32), as shown in FIG. ) etc. For example, the alphabeft symbol 32 of rIYJ is “■
-" corresponds to the Japanese phonetic symbol 45, and similarly r
The alphabetic symbol 32 of HHJ corresponds to the Japanese phonetic symbol 45 of rHJ. Note that "-" is a long sound symbol. When the Japanese phoneme conversion unit 4 receives a string consisting of English phonetic symbols (phoneme symbols) 31 and prosodic symbols shown in FIG. 4 from the English language processing unit 15b, , converts English phonetic symbols (phoneme symbols) 31 into Japanese phonetic symbols 45 with reference to the phoneme conversion table 42 .

Furthermore, the phoneme modification rule 44 is a storage area that stores information regarding Japanese phonetic rules.
), and when consonants (C) are consecutive, information on vowels (V) to be inserted between the consonants are stored.

For example, the vowel (V) added to the consonant (C) at the end of a word is generally rUJ, but in the case of rTJ, rOJ is added, and the vowel (V) is different for each consonant.

In addition, a voiceless plosive consonant (
It stores rules such as adding a consonant (``tsu'') when a word ends with a continuation of P, T, K).

The phoneme correction unit 43 converts the string of pronunciation and prosodic symbols for Japanese input from the phoneme conversion unit 41 for Japanese into phoneme correction rules 4.
4, add a predetermined vowel (V) to the final consonant (C).
), a consonant (``tsu''), etc. are added to convert it into the correct Japanese pronunciation symbol, and output to the sound processing section 17.

The German pronunciation/prosodic symbol converter 16c has almost the same configuration as the English pronunciation/prosodic symbol converter 16b, so a detailed explanation will be omitted.

Next, FIG. 7 is a diagram showing an example of the configuration of the sound source processing section 17.

The time length setting unit 51 has a table in which the length of utterance of each Japanese phonetic symbol is stored, and when a string of pronunciation/prosodic symbols is input, each synthesis unit of speech is set while referring to this table. Set the vocalization time length.

The audio file 52 is a file that stores spectral parameters consisting of PARCOR coefficients and sound source parameters indicating sound intensity for each synthesis unit, and is stored in, for example, a ROM (read only memory), floppy disk, or hard disk. It is created in an external storage device such as .

The spectral parameter generation unit 53 generates the audio file 52
Set the spectral parameters for each synthesis unit by referring to . The sound source parameter generation unit 54 also refers to the audio file 52 and sets sound source parameters for each synthesis unit.

Furthermore, the pitch parameter generation unit 55 stores basic intonation patterns, and uses pitch parameters (indicating the fundamental frequency) for each synthesis unit to express natural innation based on the accent position within the clause. Set.

The above-mentioned spectrum parameters, pinch parameters, and sound source parameters are chronologically processed by a PARCOR type waveform synthesis unit 56.
The waveform synthesis unit 56 synthesizes the audio.

Next, the operation of this embodiment configured as described above will be explained with reference to the flowcharts of FIGS. 8 and 9.

First, an example will be explained in which a Japanese character string "I met him" is input from the input device 11.

When the language processing section 15 reads out the character string "I met him" from the sentence storage section 13, it first determines whether it is a Japanese character string (SAI).

The above-described determination SAI is performed, for example, as follows. In general, Japanese is a sentence containing kanji and kana, and contains a 2-byte code, while English contains only a 1-byte code, so discrimination is made based on the difference in the number of bytes of the code representing this character.

In addition, the code representing "kana" and "kanji" (Japanese written code) and the code representing alpha bent (English written code) have different code value ranges, and are distinguished by the code value. It is also possible.

Therefore, it is possible to determine whether a sentence is in Japanese by the first character code of the sentence, but if a character code that is common to both, such as a number or symbol, is located at the beginning, the character code at the end must be searched. You can tell by looking at it.

When it is determined that the character string is Japanese through the processing described above, the character string is sent to the Japanese language processing section 15a.

The Japanese language processing unit 15a performs syntactic analysis via a preprocessing unit 21, a word identification unit 22, and a syntactic analysis unit 23 while referring to a Japanese word dictionary 23, and then performs syntactic analysis using a pronunciation analysis unit 25,
The Japanese character string "I met him" is read through the prosody analysis unit 26.
A'QTΔ,'' (SA2). Here, "'" and "," are prosodic symbols, "" is a symbol indicating an accent position, and "," is a symbol indicating a pause length.

rWATAsHIWA KA' RENI A'
The character strings of Japanese pronunciation and prosodic symbols QTA and J are output from the Japanese language processing unit 15a to the audio processing unit 17, and converted into speech by the audio processing unit 17 (SA3).

On the other hand, when the English character string rl MET HIMJ is input from the input device 11, it is converted into a JIS code by the input control unit 12, and the input data of the character string is sent to the main control unit 14 via the input control unit 12. The text is sent and written into the text storage section 13.

When the language processing unit 14 inputs the character string rl METHIMJ from the text storage unit 13 according to a control command sent from the main control unit 14, the language processing unit 14 determines that it is an English character string because rIJ is an alpha bent (5AI), and inputs the character string rl METHIMJ from the text storage unit 13. The character string is output to the English language processing section 15b.

The English language processing unit 15b receives the input “■MET
The character string HIMl is converted into a string of English pronunciation/prosodic symbols rAY MEH' T Hl)I' Ml while referring to the English word dictionary 23', and outputted to the English pronunciation/prosodic symbol converter 16b. (SA4).

The English pronunciation/prosodic symbol converter 16 is rAYMEH'
T HIH'MJ's row of English pronunciation and prosodic symbols,
Japanese pronunciation of rAI ME' QTOH BiM
It is converted into a string of prosodic symbols and output to the acoustic processing unit 17 (S
A5). The details of this process SA5 will be explained with reference to the flowchart of FIG. 9.

When the Japanese phoneme conversion unit 41 of the English pronunciation/prosodic symbol conversion unit 16 inputs the string of English pronunciation/prosodic symbols rAY MEH' T HIH' Ml, it refers to the phoneme conversion table (see Figure 6). rAYJ-rAIJ, rM
J-rMJ, rEHj-rEJ, rHJ→rHJ, rl
Perform H-HJ, rMJ→rMJ conversion (SBI). As a result, rAY MEH' T HIH' Ml is r
AI ME'T H1'M'' and output to the phoneme modification unit 43.

The phoneme modification unit 43 converts rME' TJ to rME' QTJ, which is a consonant expression for Japanese, based on the rules stored in the phoneme modification rules 44. Furthermore, rM
E' QTJ ends with the consonant rTJ, so
Add rOJ after rTJ at the end of the word. As a result, rM
E'TJ is converted to rME''QTOJ.

Furthermore, since the word rH1'MJ also ends with the consonant rMJ, a vowel rUJ is added after the word ending rMJO.

Therefore, finally rAI ME'QTOHI'
The MJ is converted into a pronunciation/prosodic symbol string for Japanese and output to the audio processing section 17.

Then, the sound processing unit 17 emits the English sentence rl MET HIMJ with Japanese pronunciation as "I met him."

It is also possible to prepare a conversion table and conversion rules that can directly convert English phonemes into Japanese phonemes, and convert English sentences directly into pronunciation/prosodic symbol strings for Japanese. In this way, audio can be output faster.

In addition, in the above embodiment, English and German sentences are synthesized with Japanese pronunciation, but in addition to English and German, various languages such as French, Spanish, Italian, and Russian are also synthesized. It is also possible to synthesize with Japanese pronunciation.

Conversely, it is also possible to synthesize Japanese with English or German pronunciation, for example.

Furthermore, spectral parameters are not limited to PARCOR coefficients, but also formants, mel cepstrum, LSP (
Line spectrum pair: 1ine spectrum
ir) etc. may be used.

〔Effect of the invention〕

As described above, according to the present invention, multilingual text can be output in clear speech that is easy for users to understand with a simple system configuration without increasing the size of the device. Furthermore, since the audio is output in a native language pronunciation that is easy for the user to hear, it has the advantage that even people who are unfamiliar with foreign languages can easily understand the content.

[Brief explanation of the drawing]

Fig. 1 is a diagram showing the principle of the present invention, Fig. 2 is a system configuration diagram of an embodiment, Fig. 3 is a diagram showing an example of the configuration of the language processing section, Fig. 4 is a diagram showing phonetic symbols for English, FIG. 5 is a block diagram showing an example of the configuration of the English pronunciation/prosodic symbol converter; FIG. 6 is a diagram showing the contents of the phoneme conversion table; FIG. 7 is a block diagram showing an example of the configuration of the sound source processing unit; Figure 8 is a flowchart explaining the process of converting an input character string into a string of pronunciation and prosodic symbols for Japanese, and Figure 9 shows a sequence of pronunciation and prosodic symbols for English to a string of pronunciation and prosodic symbols for Japanese. FIG. 10 is a diagram showing an example of a conventional text-to-speech conversion device; FIG. 11 is a diagram showing another example of a conventional text-to-speech conversion device. 1... Pronunciation/prosodic symbol generation means, 2... Speech synthesis means, 3... Language processing means, 4... Pronunciation/prosodic symbol conversion means.

Claims (1)

[Scope of Claims] 1) Pronunciation/prosodic symbol generation means ( 1); and a speech synthesis means (2) for synthesizing speech based on the pronunciation/prosodic symbols. 2) The pronunciation/prosodic symbol generation means (1) includes a language processing means (3) for converting the character string into a pronunciation/prosodic symbol string for the written first language; Pronunciation/prosodic symbol converting means (
4) The text-to-speech conversion device according to claim 1, characterized in that it has the following.