JP3002202B2 - Numeral reading device in rule speech synthesizer - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a number reading device in a rule-based speech synthesizer that reads out an arbitrary Japanese sentence including a numerical notation and outputs it as a voice.

2. Description of the Related Art In recent years, the development of natural language processing devices has been remarkable, and one of them is a rule speech synthesizer that outputs an arbitrary Japanese sentence. According to this device, numeric characters are also handled, but the number notation of Japanese sentences includes Chinese numerals,
Arithmetic numbers are mixed. Furthermore, the combination of a numeral and a classifier is used to read a conventional phrase such as "Futatsuka", or "10" for "9" (Kyuhon). In some cases, the reading of a number or a classifier changes, such as.

Here, in order to cope with a change in reading due to a combination of a numeral and a classifier, the following method has conventionally been used.
First, the pattern of the phonetic change of a numeral is classified into several types, information indicating which pattern a certain classifier takes is added to the classifier, and the phoneme of the numeral is changed by a combination with the numeral. Next, the pattern of the phonetic change of the classifier itself is classified into several types, and information on which pattern a certain classifier takes is added to the classifier, and the phoneme of the classifier is changed by the combination with the numeral. It is to make it.

Problems to be Solved by the Invention However, in such a method, it is necessary to determine which of the numeral classifier and the classifier phoneme change pattern the classifier takes.
You must register before. Therefore, when an unknown classifier is included in an input sentence (in the case of a Japanese sentence, there are various numerical notations), it is not possible to cope with a phonological change.

Means for solving the problem A word dictionary, comprising a rule table, a morphological analysis process, a numeral part in a speech synthesis device that performs a numeral part addition process, a word dictionary, while storing the part of speech of the word, reading, The sequence of numbers and classifiers is stored as a one-word Japanese word reading, the rule table stores the phoneme changes for the phoneme sequence at the connecting part of the numbers and classifiers, and the morphological analysis process converts the input Japanese sentence into a word dictionary. When the part of speech and the reading are given by the morphological analysis processing, when the part of speech and the reading given by the morphological analysis process match the phoneme in the rule table and the corresponding phoneme change, Was changed.

Further, in the morphological analysis process, when the arrangement of the divided words is a numeral and a Japanese word reading, the Japanese word read word is divided into a number word and a classifier and a reading is given.

Action There are two types of phonological change patterns, one that can respond to phonological changes by a rule based on the phonological arrangement of the connecting part of a number and a classifier, and one that makes phonological changes that cannot be handled by the rules. However, since there are only a few combinations of numbers and classifiers that change phonemes that cannot be handled by the rules, these are all registered in the dictionary in advance as exceptional readings. Then, correct readings are given at the time of morphological analysis, standard readings are given for most exceptions that are not registered in the dictionary, and finally rules based on the phoneme arrangement of the connection part between the numeral and the classifier are given. Corresponds to the phoneme change.

In addition, by using the connection check of adjacent words performed by morphological analysis, it is possible to distinguish between reading Japanese words and reading other words.

In other words, the number reading process is simplified, and even if there are unknown classifiers in the input sentence, it is possible to cope with general reading changes as long as the standard reading is given.

Embodiment An embodiment of the present invention will be described with reference to the drawings. As shown in the drawing, the numeral reading addition processing according to the present embodiment is performed by three-stage processing of standardizing numerical notation, morphological analysis, and adding numeral reading after reading one sentence.

First, in standardization of numerical notation, as shown in Table 1 In addition, various numerical notations such as decimal numbers, fractions, and approximate numbers, and fluctuations in notations using arithmetic numerals and Chinese numerals are converted into a standard form based on Chinese numerals. In this way, it is sufficient to register the notation of the standard form of Chinese numerals in the dictionary.

On the other hand, in a dictionary, ordinary numerals are as shown in Table 2. Keep the standard reading like that.

In addition, the rule based on the phoneme arrangement of the connecting part of a numeral and a classifier cannot cope with a phonological change caused by a combination of a numeral and a classifier, and a one-digit numeral and a classifier are collectively registered as one word in the dictionary as an exceptional reading. For example, as shown in Table 3 There is As can be seen from Table 3, the words of the exceptional reading of this embodiment are divided into two groups. Although words belonging to any of the groups do not change in reading in accordance with the rules, in particular, words belonging to group 2 perform Japanese reading (conventional reading). That is, Japanese reading like Group 2 is used only in the case of a one-digit number (1 to 9). For example, Example 1 “September 3rd” → “September (Kugatsu)” Example 2 "Thirteen days"->"Thirteendays" → "Ten (san)""San(san)""Day(Nichi)" In the case of Example 1, the reading according to Group 2 is given, but when the number is no longer one digit as in "Thirteen" in Example 2, the reading is given as "10" as a normal numeral. In other words, the words belonging to the group 2 are not connected to ordinary numerals, so that the first example can be used.
And the case where it is not read as in Example 2. If the word is not registered in the dictionary as an exceptional reading, the numeral and the classifier are analyzed as different words, and a standard reading is given to each word.

Finally, among the numbers and classifiers to which standard readings are attached, the readings are changed according to the phonological change rules for those that change phonologically such as vocalization, muddying, and semi-dulling. The rules used at this time are as shown in Tables 4 and 5, which are defined by the phonemes of the connection between the numeral and the classifier.

As described above, according to the numeral reading addition processing of the present embodiment,
It is possible to simplify the process of giving a number reading without lowering the accuracy of the addition of the number reading, and even if the input sentence contains an unknown classifier, it is only necessary to provide a standard reading of the classifier. A reading can be given in response to a great phoneme change.

Advantageous Effects of the Invention As described above, the present invention provides a phonological change pattern in which a phonological change can be handled by a rule based on a phonological arrangement of a connecting part of a number and a classifier, and a phonological change that cannot be handled by the rule. However, since there are only a few combinations of numbers and classifiers that change phonologically that cannot be handled by the rules, all of them are registered in the dictionary in advance as exceptional readings, and are correct when morphological analysis is performed. Yomi is given, standard readings are given to those not registered in the dictionary as exceptional readings, and finally, correspondences with phonological changes are made according to the rules based on the phonological arrangement of the connecting part between the number and the classifier.
Simplify the process of giving a number reading without lowering the accuracy of the addition of the number reading, because it uses the connection check of adjacent words performed during morphological analysis to distinguish between the case of reading a Japanese word and the case of not reading it. And can be
Even if an unknown classifier is included in the input sentence, it is possible to provide a reading that can respond to general phonetic changes as long as the standard reading of the classifier is given.

[Brief description of the drawings]

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

Claims (2)

(57) [Claims] 1. A speech number synthesizing apparatus comprising a word dictionary and a rule table, and performing morphological analysis processing and numeric word reading addition processing. A word dictionary stores a word part of speech and a reading, and includes a numeral and a classifier. Is stored as a one-word Japanese word reading, the rule table stores the phoneme change for the phoneme sequence of the connected part of the number and the classifier, and the morphological analysis process uses the word dictionary to input the input Japanese sentence. In addition to dividing words into words and giving parts of speech and readings, the numerical part reading processing is performed when the parts of speech and readings of words given in the morphological analysis processing match the phonemes of the rule table.
Numeral reading device in a rule-based speech synthesizer that changes readings to corresponding phonological changes. In the morphological analysis processing, if the sequence of the divided words is a numeral word and a Japanese word reading, the Japanese word reading word is a numeral word and
The numerical-numerical-reading apparatus in the rule-based speech synthesizer according to claim 1, wherein the numerical-numerical number is divided and the reading is assigned.