JPH0227396A - Accent type designating system - Google Patents

Abstract

PURPOSE:To easily and exactly designate an accent type by designating the accent type by a movement of a cursor, an input of a numeral or a selection of a candidate, etc., bringing its word to speech synthesis immediately by the designated accent type and calling a user's confirmation. CONSTITUTION:When a user adds newly a word to a dictionary, or in the case of changing an accent type of a word in the dictionary, the accent type in designated by a movement of a cursor, an input of a numeral, or a selection of a candidate, etc., and by the designated accent type, its word is brought to speech synthesis immediately, and a user's confirmation is called. In this case, by adding a case particle 'ga' being a subordinate attached word to the word to be registered or changed, on O type and an M type can be recognized. By hearing such a composite word, a user can confirm whether the designated accent type is as he has intended or not. Its result is informed to a system, and in the case of a synthetic voice being different from an accent which has been intended, a desired synthetic voice is obtained by designating the accent once again.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to an accent type specification method for a word dictionary for text-to-speech synthesis.

The word dictionary used in the text-to-speech synthesis system usually includes items such as nine parts of speech and accent type. The number of words that can be recorded in a word dictionary is limited, and it is not possible to record all the words used by users. In particular, with respect to proper nouns, new words, coined words, abbreviations, etc. that are used differently by each user, the user is forced to newly register the words in the dictionary. Furthermore, even in a common language, there may be multiple accent types, not just one for each word, and even more so, different accent types may be used daily depending on the region or individual. many. Therefore, it is natural that there are people who feel uncomfortable with the uniform accent type registered in the dictionary and want to change it to their own natural accent type.

In such cases, the user must specify the accent type of the word to be added or changed.Accent types are usually expressed as numbers from O to the number of moras of the word, so it is not possible to directly multiply this number. The kind of system you request (PC talk operation manual 198
7.11.20. Hudson Co., Ltd. +pplOO,
101) is also available. However, the concept of accent type is unfamiliar to general users, and it is difficult to specify it accurately. Another proposed method is to have the user say the word to change uR5, input it into the system through a microphone, and identify the user's intended accent type, but this method requires techniques such as pitch extraction and phonological segmentation. This is necessary and difficult to implement.

1-1 The present invention has been made in view of the above-mentioned circumstances,
In particular, in text-to-speech synthesis systems, the purpose is to allow users to easily and accurately specify accent types when adding new words to the dictionary or changing the accent type of words in the dictionary. It is something that

In order to achieve the above object, the present invention converts an input text into a morpheme string by morpheme analysis processing using a word dictionary, and divides the obtained morpheme string into multiple accent phrases by accent phrase division processing. an accent case position for each accent phrase is determined by accent combination processing, and a speech text synthesis device that synthesizes speech from a phonological symbol string obtained from the morpheme string and a prosodic symbol string obtained from the accent case position. When a user adds a new word to the dictionary or changes the accent type of a word in the dictionary, the user specifies the accent type by moving the cursor, entering a number, or selecting a candidate. The feature is that the word is immediately synthesized into speech based on the accent type, and the user is prompted to confirm the word.

Hereinafter, the present invention will be explained based on examples.

FIG. 1 is a diagram for explaining the flow of processing in an embodiment of the present invention. In the following explanation, the number of moras of words to be registered or changed is assumed to be M.

Therefore, 1. In order to specify the word #writing accent type in a way that is easy for the user to understand, the present invention feeds back synthesized speech based on the specified accent type to the user, prompts the user to confirm, and determines the optimal accent type. This relates to an accent type specification method that allows the user to repeat the specification until the accent type is specified.First, the system indicates to the user the accent type candidates for the word to be registered or changed. Considering that there are (M+1) types of word accent types and that most Japanese words are 107-ra or less, it is sufficiently realistic to present all candidates to the user. At this time, the accent type can be displayed by using an accent symbol to indicate the accent case position, or by using a number to indicate the accent type. In addition to selecting candidates like this, you can also specify the accent type by directly entering a number or by moving the cursor to specify the accent case position. is possible.

Next, the words are synthesized into speech according to the specified accent type, and when O-type and M-type words are uttered individually, they have exactly the same accent.

These two exhibit different accents when a subordinate adjunct is added later; the former becomes a flat accent, and the latter produces an accent case in the M-mora. In addition,
A dependent adjunct is an adjunct that does not move or eliminate the accent case of the preceding word, nor does it generate a new accent case; in other words, an adjunct word that retains the accent case of the preceding word as it is. Therefore, in this embodiment, the case particle r, which is a subordinate adjunct to the word to be registered or changed, is
By adding "ga" and synthesizing speech, it is possible to distinguish between type 0 and type M. Of course, it is also possible to replace the case particle ``ga'' with another subordinate adjunct number.

By listening to this synthesized speech, users can confirm whether the specified accent type is what they intended. The system is notified of the confirmation result, and if the synthesized speech differs from the intended accent, the accent type is specified again.

In this way, the designation is repeated until the synthesized voice based on accent 1- that the user intended is obtained.

As is clear from the above description, according to the present invention, by feeding back to the user synthesized speech based on the accent type specified by the user, it is possible to easily and accurately specify the accent type.

[Brief explanation of the drawing]

FIG. 1 is a diagram for explaining the flow of processing in an embodiment of the present invention. Patent applicant Ricoh Co., Ltd.

Claims (1)

[Claims] 1. Convert the input text into a morpheme sequence using a morphological analysis process using a word dictionary, divide the obtained morpheme sequence into multiple accent phrases using an accent phrase division process, and calculate the accent case position of each accent phrase using an accent combination process. In a speech text synthesis device that determines the morpheme string and synthesizes speech from the phonetic symbol string obtained from the morpheme string and the prosodic symbol string obtained from the accent case position, when the user adds a new word to the dictionary, or When changing the accent type of a word in the dictionary, specify the accent type by moving the cursor, entering a number, or selecting a candidate, and the word is immediately synthesized into speech using the specified accent type, and the user Accent type designation method that is characterized by prompting confirmation.