JPH02244095A - Device for reading sentence aloud - Google Patents

Abstract

PURPOSE:To allow the sentence including an assigned character or character array to be read aloud with natural voice by extracting the character or character array which is previously assigned in vocal sound information and rhythm information from a KANJI (Chinese character)-KANA (Japanese syllabary) mingled text and making identification as to whether the assigned character is one word or clause. CONSTITUTION:The character and character array which are previously assigned in the vocal sound information and the rhythm information are extracted from the KANJI-KANA mingled text inputted by an assigned character extracting means 11 and are subjected to the identification as to whether the character or character train is the word or clause in an assigned character identifying means 13. Reading aloud of the assigned character or character array as specified is, therefore, possible and the identification as to whether the character or character train is the word or clause is possible. The reading aloud by the natural voice is executed in this way without generating errors in setting of the vocal sound information and others as the sentence.

Description

[Detailed Description of the Invention] [Summary] Regarding a text-to-speech device that reads out sentences containing kanji and kana in a natural voice, the device has a configuration that allows phonological information and prosody information to be specified for a predetermined character or string of characters. The purpose of this system is to read out sentences containing letters or strings of letters in a natural voice.The system inputs sentences containing kanji and kana, and analyzes the readings and other phonological information obtained through linguistic analysis, as well as pauses between words, accent combinations, and other information. A text-to-speech device that synthesizes speech based on prosodic information accepts as input a sentence containing characters or character strings representing phonological and prosodic information, and extracts phonological and prosodic information from this kanji-kana mixture in advance. Specified character extraction means that extracts a specified character or string of characters, identifies whether the specified character is a single word or a clause containing multiple words, and analyzes sentences containing the specified character accordingly. and designated character identification means for specifying the specified character.

[Industrial application field]

The present invention relates to a text reading device that reads out Japanese sentences (text) containing Kanji and Kana in a natural voice.

In conventional speech synthesis, phonological information and prosodic information are manually provided, and the desired speech is generated using a "synthesis based on rules" technique.

By the way, with the development of word processors and OCR technology, it has become easier to input Japanese sentences, but the proofreading process still involves reading the documents together, with one person acting as the reader.

Therefore, in future speech synthesis systems, it is desired to have a device that can automatically convert these Japanese sentences (sentences containing kanji and kana that are not separated) directly into continuous speech and read them out loud.

[Conventional technology]

To read aloud sentences containing kanji and kana, the words in the sentence are correctly identified (identified), the words are automatically assigned the correct phoneme (reading) and accent, and natural accents such as bose and intonation are automatically assigned to the sentence. Language analysis technology is needed to do this.

Therefore, the structure of a text-to-speech device that reads out sentences containing kanji and kana input from an input device includes a language processing section that performs this language analysis, and a language processing section that synthesizes the corresponding speech based on the phonological and prosody information obtained here. It can be broadly divided into a speech synthesis section and a speech synthesis section.

In other words, the language processing unit decomposes the input sentence containing kanji and kana into predetermined units (words, word strings), and uses a word dictionary to perform word identification processing that adds pronunciation, accent information, and grammatical information to each unit. After that, processing is performed to reconstruct compound words, clauses, and inter-clause relationships (accent phrases, exhalation paragraphs) into higher-level linguistic units in the order of compound words, clauses, and inter-clause relationships (accent phrases, exhalation paragraphs), modify phonological information as compound words or sentences, and correct accents and inflections. This configuration determines prosodic information such as nation and passes each piece of information to the speech synthesis section.

By the way, when reading sentences using the results of normal language analysis, unnatural sounds may be produced due to deficiencies in the word dictionary or language processing errors. In such cases, for characters or character strings whose pronunciation seems unnatural in the kanji-kana-mixed sentences that are being read out, phonological and prosodic information can be specified separately, without using the results of linguistic analysis. A method of directly synthesizing speech from such information is used.

[Problem to be solved by the invention]

However, Japanese sentences containing kanji and kana do not include spaces or other symbols at the boundaries between words (they are not separated), so it is essential to identify these boundaries in language analysis. For characters or character strings for which phonological information and prosodic information are separately accounted for, it is clarified whether they are at the word level or at the phrase level, and the characters or character strings before and after are clearly identified. It is necessary to establish the relationship between

For example, in the sentence "I like Mikata.", if you specify that "Mikata" be pronounced as "Satsuki", the pronunciation as a sentence will be natural unless "Mikata" is treated as a single word. It doesn't become something. In other words, in normal pronunciation, the accents of r 三方ノ and the following ``ga 1 are combined to form one accented phrase, but if the pronunciation is analyzed while ignoring the word ``三方 j,'' which is specified separately. In this case, there was a possibility that the particle `` in V was not recognized as 1 (for example, the noun `` became 1''), and the accent combination could not be performed, resulting in unnatural pronunciation.

Also, in the sentence "I played today."
``Even if you specify the correct pronunciation using the accent phrase boundary ``/'' like ``Assonde/Shimatsuta J,'' if it is treated as a word, it will be accented with the word ``Today 1.'' occurs, and the pronunciation is also incorrect.

In addition, in the sentence ``A beautiful person who can rise up,'' even if you specify the pronunciation of ``Beautiful person who can rise up'' like ``r sakaagari/dekiru,'' if it is treated as a word and a noun, then the following ``r beautiful person'' Due to the principle that adjectives are not connected to nouns even though 1 is an adjective, the word identification process for r-beautiful J may not be performed correctly, resulting in incorrect pronunciation.

In this way, for characters or character strings for which phonological information and prosody information are separately specified, it is impossible to pronounce them correctly as a sentence unless it is recognized whether they are at the word level or the phrase level. It was difficult.

The present invention provides a configuration in which phonological information and prosody information can be specified for a predetermined character or character string.
To provide a text reading device capable of reading out a text containing specified characters or character strings in a natural voice.

[Means to solve the problem]

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

In the figure, the text reading device is configured to synthesize speech based on the readings and other phonological information obtained through linguistic analysis of sentences mixed with kanji and kana, as well as the pauses between words, accent combinations, and other prosodic information. be.

The designated character extracting means 11 manually extracts a kanji-kana mixed sentence containing characters or character strings representing phonological information and prosodic information, and extracts characters or character strings for which phonological information and prosody information are specified in advance from the kanji-kana mixed sentence. Extract.

The specified character identification means 13 identifies whether the specified character is one word or a phrase including a plurality of words, and accordingly instructs the analysis of the sentence including the specified character.

[For production]

In the present invention, the designated character extraction means 11 extracts characters and character strings for which phonological information and prosody information are designated in advance from an input sentence containing kanji and kana, and the designated character identification means 13 determines whether the characters are words. Identifies whether it is a clause.

Therefore, it is possible to read a specified character or character string as specified, and also to recognize whether it is a word or a phrase, so it is possible to avoid errors in the prosodic information and other settings of the text. It is possible to read out aloud in a natural voice without any noise.

〔Example〕

Hereinafter, embodiments of the present invention will be described in detail based on the drawings.

FIG. 2 is a block diagram showing the configuration of an embodiment of the present invention.

In the figure, an input device 23 , a preprocessing section 25 , a word identification section 27 , a phoneme/prosody processing section 29 , a speech synthesis section 31 , and a speaker 33 are connected to the main control section 21 . Further, a word dictionary 35 is connected to the word identification section 27, and an audio file 37 is connected to the speech synthesis section 31. Note that the preprocessing section 25, word identification section 27, and phoneme/prosody processing section 29 are functionally divided language processing sections.

Sentences containing kanji and kana input from the input device 23 are stored and divided into predetermined units (for example, punctuation marks) in the preprocessing unit 25 .

The word identification unit 27 identifies words or word strings while referring to the word dictionary 35, and extracts the pronunciation, accent information, and grammatical information of each word. Note that the word dictionary 35 includes notations for word identification, grammatical attributes representing connection relationships between words, and other word attributes, as well as phonemes (readings) used for processing in the next-stage phoneme/prosody processing unit 29. Accent and accent combination attributes are each registered.

The phoneme/prosody processing unit 29 performs processing to reconstruct into higher-level linguistic units in the order of compound words, clauses, and relationships between clauses. That is, based on the pronunciation of the word obtained by the word identification section 27,
Speech synthesis is performed according to phonological rules that obtain rendaku or phoneticization information generated by word concatenation, and metrical rules that determine accent phrases, exhalation paragraphs, and other tonal control information based on grammatical information of word and clause boundaries. Create various necessary control information (music information and prosody information).

The audio synthesis unit 31 refers to the audio file 37, synthesizes audio signals according to each piece of control information, and outputs the synthesized audio signal from the speaker 33 as continuous audio.

Here, the feature of the present invention is that the preprocessing unit 25 includes a specified character extraction unit 41 that extracts characters or character strings for which phonetic information and prosody information are specified in advance. It is configured to include a designated character identification section 43 that identifies whether a is a word or a phrase and notifies the word identification section 27 and the phoneme/prosody processing section 29.

It is assumed that the input sentence containing kanji and kana includes characters or character strings representing phonological information and prosody information.

FIG. 3 is a flowchart showing features of the embodiment of the present invention.

The operation of the embodiment of the present invention will be described below with reference to FIGS. 2 and 3.

The preprocessing section 25 of the language processing section takes in the input sentence containing kanji and kana, and the designated character extraction section 41 extracts characters or character strings whose phonological information and prosodic information are specified in advance.

The designated character is further identified as a word or a phrase by the designated character identification unit 43, and if it is a phrase, phrase boundary information is added and the character is subjected to the following language processing.

Note that the identification can be performed by determining the type of symbol representing the designated phonetic information and prosody information.

That is, if the type of symbol is only phonetic and only indicates one accent, the character or character string corresponding to the specified symbol can be treated as a word. Furthermore, if a symbol has information other than phoneme, such as an accent phrase boundary, the character or character string corresponding to the specified symbol is treated as a phrase.

Such identification information is sent to the word identification unit 27 and the phoneme/prosody processing unit 2 along with the specified phoneme information and prosody information.
9, and predetermined processing is similarly performed on each word or phrase.

In other words, in the example sentence above, if the symbol representing the specified phonetic information and prosodic information is r sacchi 1, which represents only the phonological information, it is treated as a word, and the following particle ``ga 1'' and the accent combination are treated as words. In addition, if the specified symbol is r assonde/shimatsuta 1 that includes an accent phrase boundary '/J, it is treated as a phrase, and language processing is performed assuming that there is a phrase boundary before and after it. This prevents errors from occurring in the setting of prosody information in the phoneme/prosody processing section 29.

〔Effect of the invention〕

As described above, according to the present invention, a character or a character string whose phonetic information and prosodic information are specified by the user is
Since it is automatically determined whether it is a word or a phrase,
There is no effect on the word identification process for sentences containing specified characters and the subsequent setting of phonological and prosodic information.

Therefore, the text-to-speech device is capable of reading aloud based on specified phonological information and prosody information, and can also read out a text that includes such information in a natural voice, making it extremely useful in practice. .

[Brief explanation of drawings]

Figure 1 is a block diagram of the principle of the present invention, Figure 2 is a block diagram showing the configuration of an embodiment of the present invention, and Figure 3 is a block diagram of the principle of the present invention.
The figure is a flowchart showing features of the embodiment of the present invention. In the figure, 11 is a specified character extraction means, 13 is a specified character identification means, 21 is a main control unit, 23 is an input device, 25 is a preprocessing unit, 27 is a word identification unit, 29 is a phonological/prosody processing unit, 31 is a 33 is a speaker, 35 is a word dictionary, 37 is an audio file, 41 is a specified character extraction unit, and 43 is a specified character identification unit. Figure 1 is a block diagram of the principles of the present invention. Figure 3 is a flowchart showing the characteristic parts of the embodiments of the present invention.

Claims (1)

[Claims] (1) In a text-to-speech device that receives a sentence containing kanji and kana and synthesizes speech based on the pronunciation and other phonological information obtained through linguistic analysis, as well as pauses between words, accent combinations, and other prosodic information, the phonological information is Specified character extraction means (11) that accepts as input a kanji-kana mixed sentence containing characters or character strings representing prosodic information, and extracts characters or character strings for which phonological information and prosodic information are specified in advance from this kanji-kana mixed sentence (11). ), and designated character identification means (13) for identifying whether the designated character is a single word or a phrase including multiple words, and instructing the analysis of a sentence including the designated character accordingly. A text reading device that is characterized by: