JPH0420998A - Voice synthesizing device - Google Patents

Abstract

PURPOSE:To properly read a word having plural meanings by providing a subject extracting part, which properly reads words having the same expression, and a word selecting part. CONSTITUTION:A character broadcast reception part receives and detects a TV radio wave to extract character broadcast data. picture element data in character broadcast data and character patterns generated by a character generating part 2 are temporarily stored in a pattern memory 3 and are outputted to a TV screen 4. The character generating part 2 receives a character code in the character broadcast to send it to the memory 3. A musical sound generating part 5 receives a musical sound code and generates musical sounds to output them from a speaker 6. A sentence segmenting part 8 analyzes the connection of sentences on a character page memory 7 and segments sentence in the read order and sends them to a language processing part 10, and a morpheme analyzing part 11 uses a word dictionary 14 to determine word by morpheme analysis. Frequencies in apperance of determined word classifications are stored in a word classification memory 15. Plural words remaining as the result of analysis are selected by a word selecting part 12 ad are subjected to rhythm conversion by a rhythm processing part 13, and a voice is synthesized by a voice synthesizing part 9 and is outputted from a speaker.

Description

[Detailed description of the invention] limb division The present invention relates to a speech synthesis device.

There are words in Jobei Yabuka's writings that have multiple readings. In order to distinguish these words, conventionally, the pronunciation of the word was determined by analyzing the syntax and connecting the meanings using the meanings of the words for each sentence. "Automatic reading of isomorphic words" (Masaru Miyazaki Journal of the Institute of Electronics and Communication Engineers '89/3 Vol.

J6g-D No, 3 pp, 392-399),
Hierarchical word attributes that extend semantic classification to distinguish between words with the same spelling but different pronunciations (homographs), analysis of kari-uke within compound nouns, and analysis of kari-uke between bunsetsu. We are making a distinction between the two. However, there are cases in which the pronunciation of a word cannot be determined just by analyzing a single sentence. In particular, in the case of abbreviations, for example, ``I like science fiction,'' it may be difficult to determine whether the word is San Francisco or science fiction because the abbreviation has multiple meanings.

The present invention was made in view of the above-mentioned circumstances, and
In particular, the purpose of this invention is to realize a speech synthesis device that can read words with multiple meanings.

Isamu---Sei.

In order to achieve the above object, the present invention (1) inputs a sentence containing kanji and kana, converts it into phonological information representing the reading and prosodic information such as accent, and selects a phoneme from a phoneme dictionary according to the information; A regular speech synthesis device that synthesizes arbitrary speech by sequentially combining based on a certain rule, which includes a topic extractor and a word selector, and the topic extractor is used for the word selector to select words. , and the word selector selects a reading of a word from a plurality of possible readings in the read-aloud text according to the information stored by the topic extractor, Furthermore,
In (1) above, (2) topic extraction is performed by a topic extractor by counting the frequency of each type of word that appears in the read-aloud text, and (3) it is performed using a thesaurus of words that appear in the read-out text. The present invention is further characterized in that (4) the means for inputting the text to be read out is a teletext receiver. Hereinafter, embodiments of the present invention will be explained using the base Is.

FIG. 1 is a block diagram for explaining one embodiment of a speech synthesis device according to the present invention, in which 1 is a teletext receiving section, 2 is a character generating section, 3 is a pattern memory, 4 is a television screen, 5
1 is a musical sound generation section, 6 is a speaker, 7 is a character page memory, 8 is a sentence cutting section, 9 is a speech synthesis section, 10 is a language processing section, and the language processing section 10 includes a morphological analysis section 11 and a word selection section 12. , a phoneme processing section 13, a word dictionary 14, a word classification memory 15, and the like. Text input means are various, such as character recognition devices and word processors, and are not limited to teletext; however, here we will discuss the case where code transmission ()) hybrid) system teletext broadcasting, which is implemented in Japan, is used as a text input means. An example will be explained.

Figures 1 to 3 are diagrams for explaining an embodiment in which a method of counting the frequency of each type of word is used for topic extraction. In the classification memory, the word selection section directly corresponds to the word selection section. Hereinafter, embodiments will be explained along the drawings.

The teletext receiving unit 1 receives and detects television waves,
Extract teletext data. The pattern memory 3 temporarily stores pixel data in teletext data and character patterns generated by the character generation section 2, and outputs them to the television screen 4. A character generating section 2 receives character codes during teletext broadcasting and sends them to a pattern memory 3. The musical tone generator 5 receives a musical tone code during teletext broadcasting and generates a musical tone corresponding to the chord.

Output from speaker 6. The above portions are the same as those of the conventional teletext receiving device, and detailed explanation will be omitted.

The character page memory 7 stores only the character codes in the teletext data on a page-by-page basis, and has an address corresponding to the display coordinate position when displaying characters on the television screen.

The sentence cutting unit 8 analyzes the connections between sentences on the character page memory 7, cuts the sentences in the order of reading, and sends the sentences to the language processing unit 10.

In the language processing unit 10, the morphological analysis unit 11 first performs morphological analysis using the word dictionary 14 to determine words based on the sentences received from the sentence segmentation unit 8 (see the word dictionary in Figure 2). (showing the structure of @14).

The word dictionary 14 has 1 notation and pronunciation as shown in Figure 2.
In addition to accents, word classifications that indicate the rough meaning of words are registered.

The word classification memory 15 counts the number of occurrences of the word classification for the word determined after the morphological analysis.

Remember it. The stored contents are not deleted even after the analysis of one sentence is completed and are left as information for the analysis of the next sentence.

Furthermore, since the topic may change in a long sentence, for example, every time a paragraph is detected, the entire contents of the word classification memory are erased to prevent the effects of old topics from remaining.

If a plurality of words remain for one notation as a result of the morphological analysis, the word selection unit 12 selects the words.

The word selection unit 12 examines the word classification for each word, obtains the number of occurrences of the corresponding word classification from the word classification memory, and selects the word with the highest number of occurrences.

Figure 3 shows an example of analysis using this method. In the example sentence (I attended a convention in New York. I also attended an SF convention.), regarding the word SF, SF (place name) and SF (literature) are used.
There are two possible readings. Therefore, if we check the number of occurrences stored in the word classification memory for each word classification.

The word NY counts as one word category for place names. Therefore, the number of occurrences of place names increases, and SF
Select the place name San Francisco as the reading.

The phonological processing unit 13 performs phonological conversion for pronunciation, such as making it consonant, making it dull, or shifting the accent, on the determined word string, and sends the phonetic symbol string to the speech synthesis unit 9. The speech synthesis section 9 selects phonemes from the phoneme dictionary based on the phonetic symbol string, sequentially combines them based on rules, and outputs speech from the speaker 6.

4 to 6 are diagrams for explaining an example when a thesaurus is used for topic extraction. In FIG. 4, everything except the language processing unit 10 is the same as in FIG. ,
Portions similar to those in FIG. 1 are omitted from illustration. Furthermore, as shown in FIG. 5, each word has higher-ranking words and lower-ranking words, and the word dictionary 14 has a structure in which higher-ranking and lower-ranking words can be found from a certain word. ing.

The word memory 16 stores words that appeared in sentences before the sentence being processed. Results of morphological analysis.

If multiple words are selected as candidates, the word selection unit 12
calculates the strength of the relationship between each word and the words stored in the word memory 16, and selects the word. In order to examine the strength of the relationship, for example, in FIG. 5, it is conceivable to use the number of branches passing between words as the distance between words.

Figure 6 is an example of analysis (Example sentence: Winter is colder than summer. It was difficult for me to get out of bed during January.) There are two readings for January, viz. , the term ``Hitotoki'' and the name of the month ``Ichigatsu'' can be considered, but if we add up the distances for each word ``Summer'' and ``Winter'' in the previous sentence, we find that ``Ichigatsu'' is the name of the month. '' is found to be closer to the word that appears in the previous sentence, and the pronunciation of ``ichigatsu'' is selected.

As is clear from the above description, according to the present invention, a speech synthesis device is provided with a topic extracting section and a word selecting section for distinguishing between words written in the same manner, so that words can be selected according to the topic. Can be read apart. Also, when the frequency of each type of word appearing in the input text is used for topic extraction.

Since existing dictionaries can be used relatively easily, the device is easy to implement. Furthermore, when a word thesaurus is used, words that are related but have different meanings can be treated using the concept of distance between words. Therefore,
It is possible to examine relationships with more words than when using word types, and more precise processing is possible. Furthermore,
When using teletext as input text for a speech synthesizer,
Since there are many abbreviations in teletext broadcasts, it is necessary to distinguish between abbreviations that have the same notation, and this method using topic extraction is considered to be highly effective.

[Brief explanation of drawings]

Figures 1 to 3 are diagrams for explaining an embodiment of the present invention using a method of counting the frequency of each type of word for topic extraction, and Figures 4 to 6 are diagrams for explaining one topic. FIG. 2 is a diagram for explaining an embodiment of the present invention when a thesaurus is used for extraction. 1... Teletext receiving section, 2... Character generating section, 3...
- Pattern memory, 4... TV screen, 5... Musical sound generation section, 6... Speaker, 7... Character page memory, 8... Sentence cutting section, 9... Speech synthesis section, 10
...Language processing unit, 11...Morphological analysis unit, 12...
Word selection unit, 13... Phonological processing unit, 14... Word dictionary, 15... Word classification memory, 16... Word memory. Patent applicant Rico Co., Ltd. attended a convention in New York. I also attended SF conventions.

Claims (1)

[Claims] 1. Input a sentence containing kanji and kana, convert it into phonological information representing the reading and prosodic information such as accent, select phonemes from the phoneme dictionary according to the information, and combine them sequentially based on certain rules to create any arbitrary A regular speech synthesis device for synthesizing speech, comprising a topic extractor and a word selector, the topic extractor storing information for the word selector to select words,
The speech synthesis device is characterized in that the word selector selects the pronunciation of a word that can be read in a plurality of ways in the text to be read, depending on the information stored by the topic extractor.