JP2622834B2 - Text-to-speech converter - Google Patents

Description

Description: FIELD OF THE INVENTION The present invention relates to a sentence-to-speech conversion apparatus for automatically converting an input sentence of a word processor into a sound and comparing the sentence with a manuscript by sound or outputting a message communication sound. It is.

(Prior Art) FIG. 2 shows a configuration of a conventional sentence-to-speech conversion device.

1 is an interface for inputting kanji-kana mixed sentence codes to the apparatus, 2 is a kanji-kanji conversion dictionary, 3 is a kanji-kana conversion section, 4 is a monosyllabic file, 5 is a speech synthesis section, 6
Is an audio output speaker.

The kanji / kana mixed code input from the interface 1 is searched by the kanji / kana conversion unit 2 and converted to a kana code string by reading and accenting based on the information.

The speech synthesizing unit 5 searches the single syllable file 4 for a syllable synthesis parameter corresponding to the kana code string, and synthesizes a speech based on this.

The kanji-kana conversion dictionary 2 stores information such as kanji codes, readings, accent types, parts of speech, etc., constituting words.

This dictionary contains words that do not have hiragana between kanji characters (eg, Karaeoka), but do not include words that have hiragana between kanji characters (eg, withered flower, number of objects). For this reason, the reading and accent type given to the words sandwiching the hiragana became unnatural.

For example, FIG. 3 shows a flow of conventional sentence-to-speech conversion for "whether there is an advantage" when a sentence such as (first sentence example) is input. In the figure, the kanji and the hiragana following it are grouped as a unit (// of the ground / whether there is an advantage /). Next, the Kanji-Kana conversion dictionary 2 is searched for each delimiter, grammatical checks are added, and reading and accent are given. Specifically, "no" is determined as a particle whose "no" follows a noun, Readings and accents. In addition, for "whether there is profit," the hiragana part is checked with which particle to join, and "ga" is determined to be a particle following the noun, Readings and accents.

The other sentences (first sentence example) are similarly converted as (second sentence example). (First sentence example) Two opinions were divided on which of the land is more advantageous.

(Second sentence example) However, when humans utter, Readings and accent types should be given.

Here, the most suitable kanji conversion is given based on the word information (part of speech, utilization, etc.) in the kanji / kana conversion dictionary. If there is no kanji conversion, the on-speech reading and the O-type accent are given. Finally, the word and the following hiragana are combined to have one accent type.

As in the conventional example, in a sentence-to-speech conversion method that does not have information on words with hiragana between kanji in the dictionary, reading and accent different from those pronounced by humans are given, and the naturalness and intelligibility of sentences are Had the problem of getting worse.

(Problems to be solved by the invention) The present invention is to solve the above-mentioned problems of the conventional example,
An object of the present invention is to eliminate the unnaturalness of the converted speech and improve the intelligibility of sentences.

(Means for Solving the Problems) According to the present invention, in order to achieve the above-mentioned object, information of a word that includes a hiragana between kanji characters is included in information such as a kanji code, a reading, an accent type, and a part of speech that constitute the word. An additional Kanji-Kana conversion dictionary, an interface for outputting Kanji-Kana mixed codes to this Kanji-Kana conversion dictionary, and searching for the Kanji-Kana conversion dictionary, giving a reading and accent type based on the information Converted to a kana code string, further, when both the kanji before and after the hiragana were not identified as two or more kanji words in the kanji kana conversion dictionary,
A kanji-kana conversion unit that identifies hiragana words between kanji and a syllable synthesis parameter corresponding to a kana code string from this kanji-kana conversion dictionary is searched from a single syllable file and based on the synthesis parameters. A speech synthesizing unit for synthesizing speech is provided to convert sentences into natural and highly intelligible speech.

Example An example of the present invention will be described below.

When a character string in the input sentence is identified by a word in the dictionary, the case where the character string is identified as a word having more characters is more likely to be correctly read and given an accent type.

In this embodiment, in consideration of this point, the dictionary is searched as follows.

In the conventional dictionary, that is, the Kanji-Kana conversion dictionary 2 in the sentence-to-speech conversion device shown in FIG.
Information such as the reading of words and accents sandwiched on one side by one kanji character, such as
If the kanji and the succeeding kanji are not identified as two or more kanji words, the kanji / kana conversion unit 3 searches the kanji / kana conversion dictionary 2 to which the information is added.

For example, FIG. 1 shows the flow of sentence-to-speech conversion in the embodiment for a sentence as described above (first sentence example). 3 differs from the conventional sentence-to-speech conversion flow shown in FIG. 3 in the following points. Since both “chi no no” and “do you have interest” contain only one kanji, And then compare it with a word in the Kanji-Kana conversion dictionary 2 to which the information has been added. Then, "ga" is determined to be a particle following the noun, and since the noun "geichi no ri" can be searched in the dictionary, Is identified.

The same is true for "the two". As a result, (first sentence example) is read and accented as (third sentence example).

(3rd sentence example) Similarly, (fourth sentence example) becomes like (fifth sentence example).

(4th sentence) The opposite of using vacant land is not the number of objects.

(4th sentence example) In the fourth sentence example, since “vacant land” and “use” are identified as two-letter words, the search for “ground advantage” is not performed.

In the present embodiment, since the addition of the dictionary and the addition of the dictionary search are performed as described above, there is an advantage that a natural voice close to human utterance is output and the intelligibility of the sentence is improved. Further, the present embodiment adopts a new dictionary addition and dictionary search method as described above, and adds dictionary information of a word sandwiching hiragana between kanji and searches for this dictionary information. So you can get more natural readings and accents.

(Effects of the Invention) The present invention adds a dictionary and improves a search method as described above so that correct readings and accents are given to words sandwiching hiragana between kanji characters. This has the advantage that the accent is given and the intelligibility of the synthesized speech can be improved.

[Brief description of the drawings]

1 is a diagram showing a flow of a sentence-to-speech conversion in one embodiment of the present invention, FIG. 2 is a diagram showing a configuration of a conventional sentence-to-speech conversion device, and FIG. 3 is a diagram showing a flow of a conventional sentence-to-speech conversion. It is. 1 ... Interface 2 ... Kanji Kana conversion dictionary 3 ...
... Kanji-Kana conversion unit, 4 ... single syllable file, 5 ... Speech synthesis unit, 6 ... Speaker.

Claims (1)

(57) [Claims] 1. A Kanji-Kana conversion dictionary in which information such as Kanji codes, readings, accent types, parts of speech, etc. constituting words are added to information on words sandwiching Hiragana between Kanji characters. An interface for outputting kanji-kana mixed codes and searching the kanji-kana conversion dictionary, converting the kana code string by adding reading and accent type based on the information, and furthermore, the kanji before and after the hiragana A kanji-kana conversion unit for identifying a word sandwiching hiragana between kanji when not identified as a word of two or more kanji in the kanji-kana conversion dictionary; Sentence-to-speech conversion characterized by comprising a speech synthesizing unit for retrieving a syllable synthesis parameter corresponding to a kana code sequence from a monosyllable file and synthesizing speech based on the synthesis parameter. Location.