JPS62119591A - Sentence reciting apparatus - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed Description of the Invention] [Summary of the Invention] The present invention identifies words by matching written sentences with words in a dictionary section, converts them into pronunciations, and reads out the sentences in a speech synthesis section. Regarding equipment.

Conventionally, in this type of text reading device, certain katakana or hiragana characters in a text cannot be matched with words in a dictionary, so word identification may fail and the correct text may not be read aloud. In the text reading device of the present invention, if a katakana or plain name character string that does not exist in the dictionary part exists in the text, the IW analysis is temporarily scrolled and the text is read backwards. Then, the beginning of the word existing in the dictionary section is detected, the M° analysis is restarted from there, and the ignored unknown words in Chikaratakari and Hiragana are converted into readings one by one and input into the speech synthesis section. The company provides a device that reads the correct text by reading the text aloud.

[Industrial application field]

The present invention relates to the configuration of a Japanese language processing device, and in particular, the present invention identifies words, etc. by matching each word, etc. of a sentence composed of words, phrases, etc. with words, etc. in a dictionary section, and identifies the words, etc. by a speech synthesis section. The present invention relates to the configuration of a text reading device that reads texts aloud.

[Traditional technique]

With the development of integration technology, the performance of computer systems has improved as processors, memory elements, printers, disk devices, etc. have become more efficient and cheaper, and the form of use has expanded to include not only individual use but also multipurpose use of systems via networks. It has progressed. In terms of input/output devices, advances in pattern recognition and voice recognition technology have led to improvements in the interface with humans, such as kanji input/output or voice input/output, and processing devices dedicated to Japanese processing have also been used as OA equipment. has been considered important. In particular, word processors are equipped with sentence structure processing functions and have become popular as devices that automatically edit, store, or create sentences, and Japanese sentences can now be easily stored in external storage devices of computer systems. Recently, so-called discrete word devices have been put into practical use, which recognize individual syllables and words by dividing human speech into units such as syllables and words. Now you can enter it into the system. When checking whether Japanese sentences entered using such a word processor or voice input device have been input correctly, or simply listening to newspaper articles etc. that have already been stored in the system, the machine is The importance of text reading devices that read texts aloud has increased. This text reading device needs a function to identify words of a text or words such as clauses or phrases by matching them with words in a dictionary section, and read out each sentence via a speech synthesis section. However, as seen in newspaper articles, there are many Japanese texts that use proper nouns or new words in katakana strings, katakana for emphasis, or plain names instead of kanji. . When analyzing such sentences with a reading device, it is necessary to analyze words as character strings (basically words, but clauses and phrases may also be used, and these will be collectively referred to as words). Identification very often fails because the word etc. does not exist in the dictionary section. Conventionally, this type of Bundo-yomi device has the disadvantage that if it fails to identify a word or the like, it will not be able to read out the correct sentence.

[Problem that the invention seeks to solve]

The present invention eliminates the shortcomings of the conventional text reading device, and when the words in the dictionary section cannot be matched with the input words, that is, parts of unknown words are ignored and identified. The present invention provides a reading device that can correctly read out the entire sentence by analyzing the possible parts and converting the ignored unknown words into readings one by one.

[Means for solving problems]

In order to achieve the above object, the present invention provides a text file section that stores texts, a text input/output section that controls text input or output from the text file section, and a text input/output section that controls the text input/output from the text file section. a sentence analysis section that divides the text read out through the section into small character string units and matches each unit with the contents of the dictionary section for identification; and a sentence analysis section that converts the reading output from the sentence analysis section into speech. a speech synthesis section, an unknown word detection means connected to the sentence analysis section and detecting an unknown word in a sentence that was not present in the dictionary section, and a position detection section detecting the start and end positions of the unknown word; It consists of an unknown word reading conversion means that creates the reading of the unknown word from the notation, and when an unknown word that does not exist in the dictionary section exists in a sentence, sentence g analysis is paused at the beginning of the unknown word and the sentence is read backwards. The present invention is characterized in that the sentence analysis is restarted from the part following the unknown word, and the reading of the unknown word is created from the notation.

[For production]

In the present invention, when a katakana character string or a plain name character string that does not exist in the dictionary is found in a sentence, identification analysis is temporarily stopped, the sentence is searched backwards, and the analysis is started from a word that is found to be the beginning of a newly identified word. The above-mentioned katakana and hiragana, which were ignored for analysis, are converted into pronunciations one by one and transferred to the speech synthesis section to read out the text.

〔Example〕

Next, an embodiment of the text reading device of the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram of the structure of a text reading device according to the present invention.

The main control unit 1 is a part that sequentially retrieves programs stored in advance in a storage device and performs sequential control according to them, similar to a general-purpose computer, and is composed of a central processing unit, a main storage device, an input/output control device, etc. This is the part where The display device 1o and the keyboard input section 11 are connected to an input/output control device in the main control section 1, and are parts for exchanging information between a human user and the text reading device. The document file 2 is a file in which a document to be read out is stored, and is generally stored in an external storage device such as a Fronbi disk device or a disk device. Each document is read from the document file 2 by the document input/output unit 3. The document input/output section 3 is a section that controls the transfer of each document from the document file 2 to the text analysis section 4 under the control of the main control section 1. Each sentence is constructed as a notation of a character string, and the character string is basically a word, but may also be a clause or a phrase, and will hereinafter be referred to as a "word, etc.".

Each sentence read from the document file 2 by the text input/output section 3 is separated by symbols such as ",", ".", etc. in the text analysis section 4, and is matched with words stored in the dictionary section 5 to find words, etc. Identification is performed. The dictionary section 5 is, for example, a kanji-kana conversion dictionary that inputs and outputs kanji and converts them into kana. If a word such as a kanji character is stored in the dictionary section 5 in each sentence outputted from the text input/output section 3, the sentence analysis section 4 can immediately convert the reading, and send the word to the rule synthesis section 6. The converted reading is given. The rule synthesis unit 6 outputs the information content corresponding to the reading in the form of voice to the audio output unit 7.
This is the part that processes it so that it can be output from. The speech synthesis section, which consists of a rule synthesis section 6 and a speech output section 7, analyzes and records the data for speech output in advance, and then reads out character strings such as words aloud in accordance with the pronunciation. This is the part that

Various analysis techniques such as waveform encoding and linear predictive analysis have already been used for data analysis for speech synthesis, and it is also possible to use a waveform processing system that encodes the speech waveform itself. Processing of a generation source processing system based on a generation model and using its control signals as data may be used. The former has good sound quality because it retains the characteristics of long sound combinations between phonemes, accents, and intonation contained in natural speech such as words and phrases, but it is not suitable for speech output other than combinations of words and phrases that have been analyzed in advance. has the disadvantage that it cannot be output. On the other hand, according to the latter method, the word or sentence to be output is input in the form of a series of phonemes or syllable symbols, so it is possible to synthesize any text. However, there is a problem in that various features of speech must be added artificially. The rule synthesis unit 6 is particularly a method for processing generation sources using phonemes or syllables, and follows a format in which speech analysis data in units of phonemes or syllables is used as a control signal for a speech production model. Therefore, the rule synthesis section 6 has no limit to the number of synthesized sounds, and regularizes the way the sounds are transformed in terms of acoustic properties based on the long combination and prosodic characteristics found in natural speech. By regularizing the transformation of this acoustic property and modifying the control signal based on this, the quality of the synthesized sound is improved so that it closely resembles natural speech. Then, in this rule synthesis section 6, t;Il'v elements, syllable symbols, or character sequences are converted into sentences t', ? fan! If the reading is inputted from the analysis section 4, an arbitrary synthesized sound corresponding to it is outputted to the speech output section 7.
It is possible to output from When using the rule synthesis section 6 to synthesize speech from each sentence in the sentence file 2 under the analysis of the text IW analysis section 4, it is ideal that a human being can perform the process of reading a book aloud. It detects and divides the boundaries of words, etc. from the text string, goes through the process of converting the phoneme symbol strings of the detected words, etc., and finally adds the accent of the word or the innation of the sentence. must not. In addition, although this rule synthesis section 6 generally has a smaller storage capacity than the waveform processing type, the circuit configuration becomes more complicated, so it is possible to reduce the complexity of the circuit configuration by executing digital calculations on a one-chip LSI. is necessary. Furthermore, it is necessary to generate control signals necessary for the operation of the speech synthesizer based on pre-stored speech data and predetermined rules. There are two types of LSIs for this purpose: formant type and PARCOR type. For example, PARCOR
The system speech synthesis LSI synthesizes speech by connecting lattice-type digital filters, each consisting of two dividers and two adders, in a cascade in the middle stage. The rule synthesis unit 6, which is configured using such an LSI, converts an input character string into a phoneme sequence according to rules and divides it into a so-called C-V-V syllable system, for example, a consonant → -vowel deca vowel system. . As for prosodic features, the fundamental frequency, amplitude, phoneme duration, etc. for the synthesized speech are determined using stored data of acoustic data stored in advance. It is known that the sound quality of synthesized speech according to such rules has an intelligibility of 90% or more for 100 Japanese syllables.

Each sentence read out from the document file 2 via the text input/output unit 3 is divided into units of words or phrases, and is matched with words in the dictionary unit 5 and identified. In this case, unknown words such as hiragana or katakana that do not exist in the dictionary section 5 may exist in the text. Especially in newspaper articles, there are many sentences in which hiragana is used instead of katakana character strings, proper nouns, new words, or katakana or kanji for emphasis, and when such sentences are analyzed, it is quite difficult to identify the words. Identification may fail due to probability. As a result, it becomes impossible to read correct sentences if left as is. Therefore, in the present invention, an unknown word detection section 40, a restart position detection section 41, and an unknown word reading conversion section 42 are connected to the sentence analysis section 4, and by operating these circuits, words etc. in the dictionary section 5 and words in the sentence are connected. If a word cannot be matched, the unknown word is ignored and analyzed first, and the pronunciation of that part is supplemented from the notation, so that the correct reading can be used. That is, if an unknown word such as hiragana or katakana that does not exist in the dictionary section 5 exists in the text, the word cannot be identified, so the unknown dh detection section 40 determines the position P+ of the unknown word. Then, the text is searched backward as it is, and the restart position detecting unit 41 finds the kanji or katakana notation position 81]2 which is the beginning of a new single dΔ. First, analysis is performed up to the position P+ of the unknown word, and then analysis is performed again from P2.

Then, in the unknown word reading conversion section 42, the reading of the unknown word part from the unknown word position WPI to before the word position P2 where the word existing in the dictionary section 5 is located is found, converted into a reading, and sent to the rule synthesis section 6. The text is synthesized and the text is read aloud by the speech output unit 7.

For example, when a sentence such as "There is a pause" is read out by the text reading device of the present invention, the analysis starts from the first character string "So". In this case, the hiragana ``there'' is in the dictionary section 5.
, and the identification of this word shall be carried out. However, let us assume that the following katakana word ``puturi'' does not exist as a word in the dictionary. Therefore, since this word cannot be identified, the unknown word detection unit 40 stores the position P+ of the unknown word, that is, the position of the character "pu" as a pointer, and searches the sentence backward. Then, the position P2 of the kanji character that is known to be the beginning of a new word, that is, the word "interruption", that is, the position of the character "chu" is found by the restart position detection unit-14=41. Therefore, after performing the analysis up to Pl, no analysis is performed on the unknown word ``Putsurito'', and the analysis is restarted on the next part ``Interrupted''. Part 1 & P1 to before P2 “Putsuri and -1
The unknown word part 31ε is changed character by character into a reading in the unknown edict reading Zl conversion unit 42. Zunawara ``Putsurito'' is pronounced as ``Butsurito'' in katakana. According to the above analysis results, all of the above sentences are converted into katakana, which is converted into katakana "Sokodeptsuritochudanshitail", which is given as a reading to the rule synthesis section 6, where it is synthesized into speech and outputted as speech from the speech output section 7. Ru.

Next, the operation of the text reading device of the present invention shown in FIG. 1 will be explained according to flowchart 1 in FIG. 2. According to the flowchart of FIG. 2, each sentence is input to the sentence analysis section 4, each word of the sentence is matched and identified with the words of the dictionary section 5, and a word lattice is created. Then, it is determined whether an unknown word in katakana or hiragana exists in each sentence. If there is no unknown word, WLIS analyzes the sentence from SP, which is the pointer at the beginning of the sentence, to EP, which is the pointer to the end of the sentence, and stores the analysis results.
The analysis results are added to T and transferred to the rule synthesis section 6. However, if there is an unknown word in katakana or hiragana, pointer P1 of the first character of the unknown word is sent to the unknown word detection unit 40.
Find by.

Then search the text backwards. As a result of the search, the user looks at the pointer P2 of the first character of the word that is found in the dictionary again. That is, the restart position detection unit 41 detects a position in the sentence that is known to be the beginning of a new word after the unknown word. After that, the sentence is analyzed from SP, which is the pointer at the beginning of the sentence, to Pl-1, which is one point before PI, which is the pointer to the first character of the unknown word, and WL is an area where the analysis results are stored.
Add to IST. Next, the pronunciation from Pl, which is the first character of the unknown word, to P2-1, which is the end pointer of the unknown word, is created character by character from the notation and added to the WLIST.

That is, the unknown word reading converter 42 adds the reading of the unknown part to the WLIST. Then, P2 is set as SP as the beginning of the sentence, and analysis is restarted from P2. After that, check again to see if there are any unknown words. By repeating this process, it is possible to analyze or add readings of unknown words to the WLIST until the end of one sentence, and then transfer all the readings of the sentence to the rule synthesis unit 6, convert it into speech according to speech synthesis, and read the sentence. Enables you to perform second-order operations.

〔Effect of the invention〕

In this way, the text reading device of the present invention ignores text analysis for unknown words for which words in the text cannot be matched with words in the dictionary, and analyzes the other parts. By supplementing the pronunciation from the notation, it is possible to read out the correct reading, and the result is that it can be synthesized for any sentence sound.

[Brief explanation of drawings]

FIG. 1 is a block diagram of the structure of the text reading device of the present invention. FIG. 2 shows a flowchart of its operation. ■・・・Main control unit, 2...Document file, 3...Text input/output section, 4...Text analysis section 1 , 5...Dictionary department, 6... Rule synthesis section, 7...Audio output section, 10...display section, 11...input section, 40...Unknown word detection unit, 41... restart position detection unit, 42...Unknown word reading conversion unit.

Claims (1)

[Scope of Claims] A text file section that stores texts; a text input/output section that controls inputting or outputting texts from the text file section; and a text input/output section that controls text input/output from the text file section; a sentence analysis unit that divides the issued sentence into small character string units and matches each unit with the contents of the dictionary unit for identification; and a speech synthesis unit that converts the reading output from the sentence analysis unit into speech. , an unknown word detection unit connected to the sentence analysis unit and configured to find an unknown word in a sentence that is not present in the dictionary unit; a position detection unit configured to detect the beginning and end positions of the unknown word; and an unknown word reading conversion means that creates a reading from the notation, and when an unknown word that does not exist in the dictionary part exists in a sentence, the sentence analysis is paused at the beginning of the unknown word and the sentence is searched backwards. A text reading device characterized in that text analysis is restarted from the part following the unknown word, and the reading of the unknown word is created from the notation.