JPH0229796A - Numeral reading adding means for rule voice synthesizing device - Google Patents

Abstract

PURPOSE:To give correct reading at the time of a morpheme and to give the reading which can cope with general phoneme variation by registering all combinations of the numeral having phoneme variation which can not be handled according to the rule and auxiliary numeral previously in a dictionary as exceptional reading. CONSTITUTION:The device is provided with a means which changes fluctuations of various numerical expressions such as decimals, fractions, and round numbers and expressions using arithmetic numbers and Chinese character numbers to a standard form based upon the Chinese character numbers, analyzes the format of a document after the conversion to the standard form and divides the document into words, and generates the part of speech, reading, etc., of each word. Further, a means which handles variation in phoneme by the combination of a numeral and an auxiliary numeral is provided and all combinations of numeral and auxiliary numeral which can not handle the phoneme variation are regarded as exceptional reading according to the rule based upon the array of phonemes at the junction part between the numeral and auxiliary numeral and registered in the dictionary as one-digit words. Then correct reading is given at the time of the morphemic analysis, and standard reading which is not registered in the dictionary as exceptional reading to give the reading which handles general phoneme variation.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to a method for assigning numeral pronunciations in a regular speech synthesizer that reads and outputs speech from arbitrary Japanese sentences containing numeric notations.

BACKGROUND OF THE INVENTION In recent years, natural language processing equipment has made remarkable progress, and one example is a regular speech synthesis device that outputs speech from arbitrary Japanese sentences. This device also handles number words, but Chinese numerals, arithmetic numerals, etc. are used in combination to represent numbers in Japanese sentences. Furthermore, depending on the combination of a number word and a particle, there are some that have idiomatic readings, such as ``two days'' (tashi), and ``nine books'' (kiyuuhon).
In some cases, the pronunciation of the number word or particle word changes, such as ``rlo book (jiyuppon)''.

Here, in order to cope with the change in reading due to the combination of a number word and a particle, the following method has conventionally been adopted. First, the phonological change patterns of number words are classified into several types, information about which pattern a certain number word takes is added to the number word, and the phoneme of the number word is changed by combining with the number word.

Next, the patterns of phonological change of the classifier itself are classified into several types, and information about which pattern a certain classifier takes is added to the classifier, and the phonology of the classifier can be changed depending on the combination with the number word. It is something.

Problems to be Solved by the Invention However, in such a method, it is necessary to register in advance which number word phoneme change pattern and the number word phoneme change pattern the number word takes. Therefore, if an unknown particle is included in the input sentence (in the case of Japanese sentences, there are various number notations), it is not possible to deal with phonological changes.

Means to solve the problem A means to convert various number notations such as decimals, fractions, round numbers, etc., as well as variations in the notation using arithmetic numerals and Chinese numerals, into a standard form based on Chinese numerals, and a standard form. It is equipped with a means to morphologically analyze the converted sentence, divide it into words, and give the part of speech and pronunciation of the word, and a means to deal with changes in phonology due to the combination of number words and particle words. The rules based on the phonetic arrangement of connected parts cannot handle the phonological changes caused by the combination of number words and classifiers as exception readings, and the one-digit number word and classifier are registered together as one word in the dictionary, and the correct reading is determined during morphological analysis. A standard reading is given to those that are not registered in the dictionary as exceptional readings, and finally, a rule based on the phonological arrangement of the connection part between a number word and a particle is used to deal with the phonological change between a number word and a particle.

At this time, we distinguish the part of speech of the words that are read in Japanese in the exceptional reading, and when we perform morphological analysis, we do not connect the words that are read in Japanese to normal number words, so that when the number is one digit, it is read in Japanese. , otherwise the standard reading is used.

Among the action phonological change patterns, there are those that can correspond to phonological changes using rules based on the phonological arrangement of the connecting part between number words and particle words.
There are both cases of phonological changes that cannot be handled depending on the rules. However, there are only a few combinations of number words and particle words that have phonological changes that cannot be handled by rules.
All of these are registered in advance in the dictionary as exceptional readings.

Then, during morphological analysis, the correct reading is assigned, and for most of the exception readings that are not registered in the dictionary, the reading of Shibeichi is given, and finally, the phonological arrangement of the connection part between the number word and the particle is given. The rules will correspond to phonological changes.

Furthermore, by using the connection check of adjacent words performed by morphological analysis, it is possible to distinguish between Japanese readings and non-Japanese readings.

In other words, the process of assigning number word readings is simplified, and even if there is an unknown particle in the input sentence, as long as the standard reading can be determined, it is possible to deal with changes in general readings.

Example An embodiment of the present invention will be described with reference to the drawings.

The number word reading assignment process of this embodiment is as shown in the drawing.
After reading the sentence, it is performed in three stages: standardization of number notation, morphological analysis, and assignment of number word reading.

First, in the standardization of number notation, as shown in Table 1,
It converts numbers into signs based on number notation and its standard form. With this, it is sufficient to register the standard form of kanji numerals in the dictionary.

On the other hand, in the dictionary, ordinary number words are listed in Table 2 as shown in Table 2.
In addition to the standard reading of number words, various number notations such as decimals, fractions, and round numbers, as well as variations in notation using arithmetic numerals and Chinese numerals, are changed to the standard reading of kanji numerals.

In addition, the rules based on the phonological arrangement of the connection part between number words and particle words cannot accommodate the phonological changes caused by the combination of number words and particle words, and as an exception reading, one-digit number words and particle words are grouped together as one.
Examples of words that should be registered in the dictionary are those shown in Table 3. As can be seen from Table 3, the words with exceptional readings in this example are divided into two groups. Although the changes in pronunciation of the words belonging to any group cannot correspond to the rules, the words belonging to group 2 in particular have a Japanese reading (ordinary reading). In other words, Japanese pronunciations like Group 2 are used only for one-digit numbers (1 to 9), so for example, Example 1 [September 3rd J - [September (Kugatsu) ” “Mitsuka” Example 2 “I-Mitsu” -N- (Jiyu-) J r三(san) “Nichi”
Naru form 1 The Japanese pronunciation for elemental analysis is shown in the example, and in cases like Example 1, read according to Group 2! j
, but when it is no longer a single digit, such as "13" in Example 2, the reading is given as "10" as a normal numeral. In other words, for lit words belonging to group 2, by not connecting them to normal number words, it is possible to distinguish between cases in which they are read in Japanese, as in example 1, and cases in which they are not, as in example 2. . Furthermore, if the word is not registered in the dictionary as an exceptional reading, the number word and the number word are analyzed as separate words, and each is given a standard reading.

Finally, among the number words and classifiers that have been given standard readings, the readings are changed using phoneme change rules for those that undergo phonetic changes such as consonantization, dusky consonance, and semi-voiced consonance.

The rules used in this case are as shown in Tables 4 and 5, which are determined by the phonetic arrangement of the connecting parts between number words and particle words.

Table 4 Phonological change rules for number words Table 5 Phonological change rules for classifiers As described above, according to the number word reading assignment process of this embodiment, the number word reading assignment process can be simplified without reducing the accuracy of number word reading assignment. Even if an unknown classifier is included in the input sentence, as long as the standard pronunciation of the classifier is given, the reading can be assigned in response to general phonological changes.

Effects of the Invention As described above, the present invention has two types of phonological change patterns: those that can be accommodated by the phonological and related rules at the connection between a number word and a particle, and those that cannot be accommodated by the rules. There are two types, but since there are a small number of combinations of number words and classifiers that have phonetic changes that cannot be handled by rules, all of these are registered in the dictionary as exceptional readings in advance, and the correct reading is assigned during morphological analysis. Then, standard readings are given to words that are not registered in the dictionary as exceptional readings, and finally, phonological changes are dealt with using rules based on the phonological arrangement of the connection between number words and particle words.Furthermore, during morphological analysis, By using the connection check of adjacent words to distinguish between Japanese readings and non-Japanese readings, the process of adding number readings can be simplified without reducing the accuracy of adding number readings. Moreover, even if an unknown classifier is included in the input sentence, as long as the standard pronunciation of the classifier is given, it is possible to assign a reading that can accommodate general phonological changes.

[Brief explanation of the drawing]

The drawing is a flowchart showing the processing procedure of the present invention.

Claims (1)

[Scope of Claims] 1. A method for assigning number readings in a regular speech synthesizer that outputs speech of arbitrary Japanese sentences, using various number notations such as decimals, fractions, and round numbers, as well as arithmetic numerals and Chinese numerals. A means of converting the fluctuations in writing into a standard form based on Chinese numerals, a morphological analysis of the sentence after converting it to the standard form, and dividing it into words.
It is equipped with a means for giving the part of speech and pronunciation of a word, and a means for responding to phonological changes due to the combination of a number word and a classifier, and a rule based on the phonological order of the connection part between a number word and a classifier, which corresponds to a phonological change due to a combination of a number word and a classifier. 1-digit number words and classifiers are collectively read as 1, with all words that cannot correspond to
It is registered as a word in the dictionary, the correct reading is given during morphological analysis, the standard reading is given to words that are not registered in the dictionary as exceptional readings, and finally, the phonological order of the connection part between number word and particle is given. A method for assigning number readings in a regular speech synthesis device, characterized in that the rules correspond to phonological changes between number words and particle words. 2. For exception readings, we distinguish the parts of speech for Japanese readings, and when performing morphological analysis, we do not connect Japanese readings to normal number words, so when it is a one-digit number, we use Japanese readings. 2. The method for assigning numeral pronunciations in a regular speech synthesizer according to claim 1, wherein the standard pronunciation is used in other cases.