JPH03225400A - Pause length determining system - Google Patents

Abstract

PURPOSE:To finely control pause length corresponding to the structure of a sentence to be converted into a voice by using pause length right after a word corresponding to the part of speech of a word right before a boundary where a pause is inserted. CONSTITUTION:A character string including plural words is inputted from a character string input terminal 1. The character string which is inputted is sent to a pause insertion position determination part 12 and a morpheme analyzing part 13. The determination part 21 determines the boundary where a pause is to be inserted. Information on the determined boundary is sent to a pause length determination part 15. The analyzing part 13 divides the sentence into word strings to determine the parts of speech of the respective words. The word strings accompanied by the part-of-speech information are sent to a determination part 15. The determination part 15 reads pause length out of a pause length storage part 14 inserted behind the word of the part of speech corresponding to the part of speech of the word right before the pause insertion boundary determined by the determination part 12 to determine the pause length and outputs the result from a pause length output terminal 16. Consequently, the proper pause length can be determined.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention is applicable to word boundaries in speech generated based on an input string containing multiple words in regular speech synthesis, etc. that converts a string of characters into speech. The present invention relates to a pause length determination method for determining the length of a pause to be inserted.

(Prior art) In regular speech synthesis that converts arbitrary sentences into speech,
It is necessary to insert a pause of an appropriate length at an appropriate position in a sentence, and this is important in improving the naturalness of synthesized speech. In order to convey the break in meaning to the listener, the speaker uses appropriate boundaries between clauses or words that have weak connections in terms of meaning (hereinafter, the term "boundary" simply refers to the boundaries between clauses or words). put a pose. Additionally, there is a limit to how long a human can vocalize in one breath, so breath breaks are necessary. Pause is also a phenomenon caused by such physiological demands. Therefore, when generating synthesized speech, if a pause of an appropriate length is not inserted into the synthesized speech, the synthesized speech will lack human-likeness, and the listener will feel that the synthesized speech is unnatural.

Conventionally, the strength of the semantic connection between adjacent clauses was used to determine the pause length.

The strength of the bond at the boundary is expressed by the number of clauses from the preceding clause to the receiving clause, and this measure is called the degree of separation. A large value of the degree of separation indicates that a certain phrase is connected to a phrase that is further away, and that the connection with adjacent phrases is weak. Therefore, it can be said that a pose is likely to be inserted at a boundary with a high degree of separation, and the length of the inserted pose tends to be long. Furthermore, there is a limit to the length of speech that humans can make in one breath. Therefore, the longer the length of the exhalation paragraph (a group of breaths uttered between pauses) before and after the boundary, the longer the inserted pause tends to be. From the above, the value of the product of the degree of separation and the total number of moras of exhalation paragraphs before and after the boundary is used as an evaluation measure for determining the pause length to be inserted. Several thresholds are set for this product value, a pause length corresponding to each threshold is determined in advance, and the pause length is determined based on which threshold range the calculated product value falls within. .

A method for determining such a pause length is described in detail in the Acoustical Society of Japan's Speech Study Group Material 378-07 (1978-4) "Study of intonation rules for r-sentence speech".

(Problems to be Solved by the Invention) In the conventional method, in order to calculate the degree of separation, it was necessary to accurately analyze the dependencies between clauses that make up a sentence. However, since it is difficult to always accurately analyze the relationship of this modification, it is difficult to determine an appropriate pause length, which poses a problem in that the synthesized speech may become unnatural. Furthermore, since the length of the pause to be inserted at the boundary is selected from a small number of predetermined pause lengths, it is not possible to perform detailed control of the pause length according to the structure of the sentence.

In contrast, an object of the present invention is to provide a pause length determination method that does not require analysis of text dependencies and can perform detailed pause length control according to the structure of the text.

(Means for Solving the Problems) A first aspect of the present invention provides a pause length determination method for determining the length of a pause to be inserted at a word boundary in speech generated based on an input character string including a plurality of words. , for a plurality of words, store in advance the length of the pause to be inserted immediately after the word for each part of speech, and select the length of the pause to be inserted immediately after the word that is located immediately before the word boundary. The present invention is characterized in that the length of the pause inserted at the word boundary is determined based on the length of the pause inserted immediately after.

The second aspect of the present invention provides a pause length determination method that determines the length of a pause to be inserted at a word boundary in speech generated based on an input character string including a plurality of words. The length of the pause inserted immediately before each word is stored in advance, and the length of the pause inserted immediately before the stored word corresponding to the part of speech of the word located immediately after the word boundary is stored in advance. The length of the pause to be inserted at the word boundary is determined based on the word boundary.

The third aspect of the present invention provides a pause length determination method that determines the length of a pause to be inserted at a word boundary in speech generated based on an input character string including a plurality of words. The length of the pause inserted immediately after the word and the length of the pause inserted immediately before the word are respectively stored in advance for each word, and the length of the pause inserted immediately before the word is stored in advance. at the word boundary based on the length of the pause inserted immediately after the word that is located, and the length of the pause inserted immediately before the stored word that corresponds to the part of speech of the word located immediately after the word boundary. It is characterized by determining the length of the inserted pose.

The fourth aspect of the present invention provides a pause length determination method for determining the length of a pause to be inserted at a word boundary in speech generated based on an input character string including a plurality of words.
For a word string consisting of two words, the length of a pause to be inserted between one and the other word of the word string is stored in advance for each combination of one part of speech and the other part of speech of the word, The length of the pause to be inserted at the word boundary is determined based on the length of the pause of the stored word string corresponding to a combination of parts of speech of words located before and after the word boundary.

(Effect) Pauses tend to be inserted at boundaries between clauses or words, where the words before and after the boundary have a weak connection in terms of meaning, and the length of the inserted pauses tends to be long. . In the conventional method, the strength of this connection was calculated from the structure of the text's dependencies. On the other hand, if you examine the parts of speech of the words before and after the boundary where the pause is inserted,
It can be seen that there are parts of speech in which pauses are easily inserted, that is, the connections are weak, and parts of speech that are difficult to insert, that is, the connections are strong. Therefore, it is thought that by utilizing these properties of parts of speech, it is possible to determine natural pause lengths for sentence dependencies without using structure.

Therefore, in advance, we have prepared the sounds of some sentences read out loud by humans (hereinafter referred to as the "sentence speech database").
Then, the part of speech of the word immediately before or after the boundary where the pause is inserted is checked, and for each part of speech, the length of the pause inserted immediately before or after that word is checked. As the pause length, it is possible to use, for example, the average value of the pause lengths of all occurrences of a certain part of speech accompanied by a pause in a text speech database.

For the sake of explanation, let Lf (PS) be the average value of the pause lengths inserted immediately after a word of a certain part of speech PS, and Lp (PS) be the average value of the pause lengths inserted immediately before. Also,
Let L (PSp, PSf
). These pose length average values Lf (PS), L
p (PS) for each part of speech ps, and L (PSp, P
Sf) is obtained for each combination of parts of speech PSp and PSf. At this time, the types of parts of speech can be classified into nouns, verbs, and adverbs. In addition, verbs and adjectives
For parts of speech that are conjugated, such as auxiliary verbs, the unnatural form,
It is also possible to classify them by conjugated form, such as continuous form and adnominal form.

In the first aspect of the present invention, in a sentence to be converted into speech, an average pause length Lf (PSp) immediately after a word is used, which corresponds to the part of speech PSp of the word immediately before the boundary where a pause is inserted. Furthermore, as described in the (Prior Art) section, the pause length is also affected by the length of the exhalation paragraphs before and after the pause. Therefore, it is possible to deal with the influence of the length of the exhalation paragraphs before and after the pause by calculating Lf(PSp) for each length of the expiration paragraphs before and after the pause (for example, the number of moras).

In the second aspect of the present invention, the average pause length Lp (PSf) immediately before a word is used, which corresponds to the part of speech PSf of the word immediately after the boundary where a pause is inserted. The same method as described in the first aspect of the present invention can be considered as a method for realizing the influence of the length of the exhalation paragraphs before and after a pause.

In the third aspect of the present invention, the pause length average value Lf (PSp) immediately after a word is determined according to the part of speech PSp of the word immediately before the boundary where a pause is inserted, and the part of speech PSf of the word immediately after the boundary is determined. In addition, the average pause length immediately before the word Lp(
PSf) is used.

A possible method for determining the pause length using both is, for example, taking the arithmetic average of Lf (PSp) and Lp (PSf). The same method as described in the first aspect of the present invention can be considered as a method for realizing the influence of the length of the exhalation paragraphs before and after a pause.

In the fourth aspect of the present invention, the average value L of pause lengths corresponds to the combination of the part of speech PSp of the word immediately before the boundary where the pause is inserted and the part of speech PSf of the word immediately after.
, f) is used. The same method as described in the first aspect of the present invention can be considered as a method for realizing the influence of the length of the exhalation paragraphs before and after a pause.

In the first and second aspects of the present invention, the pause length is determined by considering the parts of speech of words either before or after the boundary where the pause is inserted.

On the other hand, in the third aspect of the present invention, the influence of the parts of speech of the preceding and following words on the pause length is independently considered, and by taking the arithmetic average of the pause lengths determined from each word, the parts of speech of the preceding and following words are The pause length is determined by considering the influence of As a result, it can be expected that the inserted pause length will be more suitable for the boundary than in the first and second aspects of the present invention.

However, it is conceivable that the pause length obtained by taking the arithmetic average of the pause lengths determined according to the parts of speech of the preceding and following words will not necessarily be an appropriate value.

It is also considered that there may be cases where it is necessary to put more weight on either the front or the rear. Furthermore, the weighting method may also differ depending on the combination of the preceding and following parts of speech. Therefore, in the fourth aspect of the present invention, in order to solve this problem, the length of the pause to be inserted is determined for each combination of the preceding and succeeding parts of speech. Then, a predetermined pause length is used according to the combination of word parts of words before and after the boundary where the pause is inserted. Such a method makes it possible to determine a more suitable pause length than the third invention.

By using the above method, it is possible to determine a natural pause length by simply using the part of speech of the word instead of using the structure of the dependency of the sentence to be converted into speech9.
By controlling each part of speech, it becomes possible to finely control the pause length according to the structure of the sentence.

(Embodiment) FIG. 1 is a block diagram showing an embodiment of the pose length determination method of the first invention.

First, input a character string containing multiple words to the character string input terminal 11.
Enter from. The input character string is sent to the pause insertion position determination section 12 and the morphological analysis section 13. The pose insertion position determining unit 12 determines a boundary for inserting a pose using a well-known method. Information on the boundary determined as the pose insertion position is sent to the pause length determining section 15. The morphological analysis unit 13 divides the sentence represented by the input character string into word strings, and determines the part of speech of each word. The word string with the part of speech information is sent to the pause length determining section 15. The pause length storage unit 14 stores in advance the pause length Lf (PSp) to be inserted after the word of the part of speech PSp, or for each part of speech PSP. The pause length determination unit 15 determines the pause length Lf (PSp) stored in the pause length storage unit 14 according to the part of speech PSp of the word immediately before the pose insertion boundary determined by the pause insertion position determination unit 12. readout, (action)
The pause length to be inserted is determined by the method described in the above section, and the result is output from the pause length output terminal 16.

In order to realize the second aspect of the present invention, the pause length storage unit 14 in FIG. 1 stores a pause length Lp (PSf) to be inserted immediately before the word of the part of speech PSf for each part of speech PSf. The pause length determination unit 15 determines the pose to be inserted immediately before the word stored in the pause length storage unit 14, depending on the part of speech of the word immediately after the pose insertion boundary determined by the pause insertion position determination unit 12. It is sufficient to read the length Lp (PSf) and determine the pause length to be inserted using the method described in the (effect) section.

In order to realize the third aspect of the present invention, the pause length Lf (PSp) inserted immediately after the word of the part of speech PSp and the pause length Lf (PSp) inserted immediately before the word of the part of speech PSf are stored in the pause length storage unit 14 in FIG. Pose length LP (PSf) and part of speech PSp
, PSf respectively. Pose length determination section 1
5, the pause length Lf( PSp).

Also, depending on the part of speech PSf of the word immediately after the insertion boundary, the pause length 1. Read p (PSf).

These two pose lengths Lf (PSp), Lp (PSf
) to determine the length of the pause to be inserted using the method described in the (effect) section.

In addition, in order to realize the fourth aspect of the present invention, it is necessary to store the pose length L inserted at the boundary in the pose length storage unit 14 in FIG.
(PSp, PSf) is the part of speech PS of the words before and after the boundary
It is stored for each combination of p and PSf. The pause length determination unit 15 determines the combination of parts of speech (PS) of words before and after the pause insertion boundary determined by the pause insertion position determination unit 12.
p, PSf), the pause length L (PSp, PSf) stored in the pause length storage section 14 is read out, and the pause length to be inserted is determined by the method described in the (effect) section. good.

(Effects of the Invention) As described above, according to the present invention, the pause length of a dependency that requires accurate analysis is determined without using the analysis result, and the part of speech of the word is used to determine the pause length. Since the pause length is finely controlled according to the structure of the sentence, it is possible to determine a more appropriate pause length than in the conventional method. Therefore, the present invention is effective as a pause length determination method in a speech synthesis device or the like that converts an arbitrary sentence given as a character string into speech.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an embodiment of the first pose length determination method of the present invention. DESCRIPTION OF SYMBOLS 11... Character string input terminal, 12... Pause insertion position determination section, 13... Morphological analysis section, 14... Pose length storage section, 15... Pose length determination section, 16... Pause Long output terminal.

Claims (4)

[Claims] (1) In a pause length determination method that determines the length of a pause to be inserted at a word boundary in speech generated based on an input character string containing multiple words, The length of the pause to be inserted immediately after is stored in advance, and the word is selected based on the length of the pause to be inserted immediately after the stored word that corresponds to the part of speech of the word located immediately before the word boundary. A pose length determination method characterized by determining the length of a pose inserted at a boundary. (2) In a pause length determination method that determines the length of a pause to be inserted at a word boundary in speech generated based on an input string containing multiple words, the pause length determination method determines the length of a pause inserted at a word boundary in speech generated based on an input string containing multiple words. The length of the pause to be inserted immediately before is stored in advance, and the word is selected based on the length of the pause to be inserted immediately before the stored word that corresponds to the part of speech of the word located immediately after the word boundary. A pose length determination method characterized by determining the length of a pose inserted at a boundary. (3) In a pause length determination method that determines the length of a pause to be inserted at a word boundary in speech generated based on an input string containing multiple words, the pause length determination method The length of the pause inserted immediately after the word and the length of the pause inserted immediately before the word are stored in advance, respectively, and the length of the pause inserted immediately before the word is stored in advance, and the length of the pause inserted immediately before the word is stored in advance. a pause inserted at the word boundary based on the length of the pause inserted immediately after and the length of the pause inserted immediately before the stored word corresponding to the part of speech of the word located immediately after the word boundary; A pose length determination method characterized by determining the length of. (4) In a pause length determination method that determines the length of a pause to be inserted at a word boundary in speech generated based on an input character string containing multiple words, for a word string consisting of two consecutive words, storing in advance the length of a pause to be inserted between one and the other of the words in the word string for each combination of one part of speech and the other part of speech of the word;
A pause characterized in that the length of the pause to be inserted at the word boundary is determined based on the length of the pause of the stored word string corresponding to a combination of parts of speech of words located before and after the word boundary. Long decision method.