JPH0337700A - Pause insertion position determining system - Google Patents

Abstract

PURPOSE:To determine a natural pause insertion position by discriminating whether a pause should be inserted to a word boundary or not based on the numerical value which corresponds to the part of speech of the word of the boundary of each word constituting a sentence expressed with an inputted character string and indicates the preliminarily stored degree of easiness of pause insertion just after each word. CONSTITUTION:The character string to be converted to a voice is inputted from a character string input terminal 11. The inputted character string is sent to a morpheme analyzing part 12, and the sentence expressed with the input character string is divided to word strings, and parts of speech and readings of respective words are determined. A pause insertion position determining part 15 reads out the pause insertion probability which corresponds to the part of speech of the word just before the boundary of each word string having a part of speech and is stored in a pause insertion settlement storage part 14. The read-out pause insertion probability and the length of each word calculated by a word length calculating part 13 are used to determine the pause insertion position, and the result is outputted from a pause insertion position output terminal 16. Thus, the natural pause insertion position is determined.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to a pause insertion position determining method for determining a position to insert a pause in regular speech synthesis for converting character strings into speech.

(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 other words, there is a limit to how long a human voice can be uttered in one breath, and the speaker pauses at the boundaries of phrases or words that have weak connections in terms of meaning, either to take a breather or to convey a break in meaning to the listener. put

Therefore, when generating synthesized speech, the listener will feel unnatural if appropriate pauses are not inserted.

Conventionally, the strength of the connection between adjacent bunsetsu has been used to determine the pose insertion position.

In other words, it utilizes the property that the weaker the semantic connection between the preceding and following phrases, the easier it is for a pause to be inserted.

First, the strength of the bond at the boundary between bunsetsu and bunsetsu is expressed by the number of clauses from the preceding clause to the receiving clause,
This measure is called 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, we believe that there is a high possibility that a pause will be inserted at a bunsetsu boundary with a high degree of separation. Furthermore, humans have a limit to how long they can vocalize in one breath, so the speaker pauses at an appropriate position to catch their breath. Therefore, pauses are likely to be inserted when the total number of moras in the clauses before and after the clause boundary is large. Based on the above, the product of the degree of separation and the total number of moras is used as an evaluation measure to determine whether or not to insert a pause, and if this exceeds a certain predetermined threshold, a pause is inserted. shall be taken as a thing.

Regarding the method of determining such a pause insertion position, please refer to the Acoustical Society of Japan Speech Research Group Sample 878-07 (1978-4).
Examination of intonation rules for one-sentence speech] (Reference 1).

(Problems to be Solved by the Invention) In the conventional method, it was necessary to accurately analyze the dependencies between clauses that make up a sentence. However, it is difficult to always accurately analyze the relationship of this modification, and as a result, there is a problem that the selected pose insertion position may become unnatural.

In contrast, an object of the present invention is to provide a pose insertion position determination method that can determine a natural pose insertion position without analyzing such dependency relationships.

(Means for Solving the Problem) The first aspect of the present invention provides, when converting an input character string into speech, a sentence represented by the input character string is decomposed into its constituent words, In the pause insertion position determination method that determines whether or not to insert a pause at the boundary of each decomposed word, a numerical value indicating how likely a pause is to be inserted immediately after a word is stored in advance for each part of speech of the word. , the word boundary is determined based on a numerical value representing the ease of inserting a pause immediately after the memorized word according to the part of speech of the word immediately before the boundary of each word constituting the sentence represented by the input character string. It is characterized by determining whether or not to insert a pose into the image.

A second aspect of the present invention, when converting an input character string into speech, decomposes a sentence represented by the input character string into its constituent words, and pauses at the boundary of each decomposed word. In the pause insertion position determination method that determines whether or not to insert a pause, a numerical value indicating how likely a pause is to be inserted immediately before a word is memorized for each part of speech of the word, and the number expressed by the input character string is Determine whether or not to insert a pause at the word boundary based on a numerical value representing the ease with which a pause immediately before the memorized word is inserted according to the part of speech of the word immediately after the boundary of each word constituting the sentence. It is characterized by

A third aspect of the present invention is that when converting an input character string into speech, the sentence represented by the input character string is decomposed into its constituent words, and pauses are made at the boundaries of each decomposed word. In the pause insertion position determination method that determines whether to insert a pause, a numerical value indicating how likely a pause is to be inserted immediately after a word and a numerical value indicating how likely a pause is to be inserted immediately before a word are determined in advance. , each part of speech of each word is memorized, and a pause immediately after the memorized word is inserted corresponding to the part of speech of the word immediately before the boundary of each word constituting the sentence represented by the input character string. Whether to insert a pause at the word boundary based on a numerical value representing the ease of insertion and a numerical value representing the ease with which a pause immediately before the memorized word can be inserted according to the part of speech of the word immediately after the word boundary. It is characterized by determining.

(Action) A pause is a boundary between a phrase or a word (hereinafter, the term "boundary" simply refers to a boundary between a phrase or a word)
, words are likely to be inserted into a boundary where the words before and after the boundary have a weak connection in terms of meaning. In the conventional method, the strength of this connection was calculated from the structure of the text's dependencies. On the other hand, when examining the parts of speech of words before and after the boundary where a pause is inserted, it is found that there are parts of speech in which pauses are easily inserted and parts of speech in which it is difficult to insert. Therefore, it is thought that by utilizing such properties of parts of speech, it is possible to determine a natural pause insertion position without using the structure of sentence dependencies.

Therefore, in advance, we check the parts of speech of the words immediately after or immediately before the boundary where the pause is inserted from the audio of several sentences read aloud by humans (hereinafter referred to as the sentence audio database), and then analyze the parts of speech for each part of speech. Find out how likely a pause is to be inserted immediately before or after a word. As the ease with which a pause is inserted, for example, the ratio of occurrences of a certain part of speech accompanied by a pause to all occurrences of that part of speech in the sentence speech database may be used. That is, the probability P,(PS) that a pause will be inserted immediately before a word whose part of speech is PS, and the probability P<PS) that a pause will be inserted immediately after, are defined as follows.

P, (PS) = (Part of speech PS accompanied by a pause immediately before
(number of occurrences of part of speech ps) / (total number of occurrences of part of speech ps) ,,,, (1
)P〆PS)=(Number of occurrences of part-of-speech PS that immediately follows a pause)/(Total number of occurrences of part-of-speech ps) ,,,, (2)
In the following, an example will be explained in which the pause insertion probability Pp (PS), ppps) is used as a numerical value representing how likely a pause is to be inserted immediately before or after a word.

These pause insertion probabilities Pp(PS), P〆PS), that is, the ease with which a pause is inserted, are determined for each part of speech.

There are different classification methods for parts of speech, such as nouns, verbs, and adverbs. Furthermore, for parts of speech that are inflected such as verbs, adjectives, and auxiliary verbs, it is also possible to classify them into conjugated forms such as unnatural form, conjunctive form, and adjunctive form.

In the first aspect of the present invention, as a criterion for determining the boundary where a pause should be inserted in a sentence to be converted into speech,
It is shown by equation (2). Pause insertion probability P〆PS according to the part of speech PS of the word immediately before the boundary at each boundary
, ) is used. As a method of actually determining the pose insertion position using such a determination criterion, for example, a method as shown in (Reference 1) can be considered. That is, the evaluation scale is a value calculated from the judgment scale 1 degree and the lengths of the clauses before and after the boundary (for example, the number of moras). A threshold is set for this evaluation scale, and depending on whether the evaluation scale at each boundary exceeds the threshold, it is determined whether or not to insert a pause at the boundary. Alternatively, the number of pauses to be inserted may be determined first based on the length of the sentence to be converted into speech, and the pauses may be inserted in order from the boundary with the largest criterion.

In the second aspect of the present invention, the pause insertion probability P, (PSr) corresponding to the part of speech PSf of the word immediately following the boundary in each boundary is used as the criterion for the determination.

Further, in the third aspect of the present invention, (2
), the pause insertion probability P1(PS,
) and the pause insertion probability Pp (PSf) according to the part of speech PSr of the word immediately after the boundary, which is shown by equation (1). As a judgment scale using both, for example,
A possible example is the sum of PdPS, ) and P, (PSF).

By using the above method, it is possible to determine a natural pause insertion position by simply using the part of speech of the word obtained through morphological analysis, without using the structure of the dependency of the sentence to be converted into speech. Become.

(Embodiment) FIG. 1 is a block diagram showing an embodiment for realizing the pose insertion position determination method according to the first invention.

First, the character string to be converted into speech is input to the character string input terminal 11.
Enter from. The input character string is sent to the morphological analysis section 2, which divides the sentence represented by the input character string into word strings and determines the part of speech and pronunciation of each word. This word string with part of speech and pronunciation is sent to the word length calculation section 13 and the pause insertion position determination section 15. When the word string is given, the word length calculation section 13 calculates the length of each word (for example, the number of moras) constituting the sentence using the pronunciation of the word, and sends it to the pause insertion position determination section 15. .

The pause insertion probability storage unit 14 stores, as expressed by equation (2),
The probability that a pause will be inserted after a part of speech (pKps) is stored in advance for each part of speech.

The pose matchmaker position determination unit 15 first determines the pose suppression probability PrCPS stored in the pause insertion probability storage unit 14, which corresponds to the part of speech PS of the word immediately preceding each boundary using the word string with the part of speech. , ). Using the read pause insertion probability P [PSp) and the length of each word calculated by the word length calculation unit 13, the position at which a pause is inserted is determined by the method described in the (effect) section. is determined, and the result is output from the pause insertion position output terminal 16.

In order to realize the second aspect of the present invention, the pause insertion probability stored in the pause insertion probability storage section 14 in FIG. 1 is set to Pp (PS) expressed by equation (1). and,
The pause insertion position determination unit 15 reads out the pause insertion probability Pp (PSr) stored in the pause insertion probability storage unit 14 according to the part of speech PSf of the word immediately after each boundary, and reads out the pause insertion probability Pp (PSr) stored in the pause insertion probability storage unit 14. The position to insert the pose can be determined in the same way.

In addition, in order to realize the third aspect of the present invention, it is necessary to store P expressed by equation (1) in the pause insertion probability storage unit 14 in FIG.
, (PS) and PdPS> expressed by equation (2) are stored. Then, the pause insertion position determining unit 15 calculates the pause insertion probability PdPS, ) corresponding to the part of speech PSp of the word immediately before each boundary, and the part of speech P of the word immediately after each boundary.
The pose insertion probabilities P, .

(Effects of the Invention) As described above, the present invention determines the insertion position of a pose without using the results of analysis of dependencies that are difficult to analyze correctly. Therefore, it is possible to determine a more accurate and natural pose insertion position than the conventional method. Therefore, the present invention is effective as a pause insertion position 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 for realizing the pose insertion position determination method according to the first invention. In the figure, 11 is a character string input terminal, 12 is a morphological analysis section, 13 is a word length calculation section, 14 is a pause insertion probability storage section, 15 is a pause insertion position determination section, and 16 is a pause insertion position output terminal.

Claims (3)

[Claims] (1) When converting an input character string into speech, do you decompose the sentence represented by the input character string into its constituent words and insert a pause at the boundary of each decomposed word? In the pause insertion position determination method, a numerical value indicating how likely a pause is to be inserted immediately after a word is stored in advance for each part of speech of the word, and a sentence represented by the input character string is constructed. It is characterized in that it is determined whether or not to insert a pause at the word boundary based on a numerical value representing the ease with which a pause is inserted immediately after the memorized word according to the part of speech of the word immediately before the word boundary. A pose insertion position determination method. (2) When converting an input character string into speech, do you decompose the sentence represented by the input character string into its constituent words and insert a pause at the boundary of each decomposed word? In the pause insertion position determination method, a numerical value indicating how likely a pause is to be inserted immediately before a word is stored in advance for each part of speech of the word, and a sentence represented by the input character string is constructed. It is characterized in that it is determined whether or not to insert a pause at the word boundary based on a numerical value representing the ease with which a pause immediately before the memorized word is inserted according to the part of speech of the word immediately after the word boundary. A pose insertion position determination method. (3) When converting an input character string into speech, do you decompose the sentence represented by the input character string into its constituent words and insert a pause at the boundary of each decomposed word? In the pause insertion position determination method that determines whether or not the word It is memorized for each part of speech and represents the ease with which a pause can be inserted immediately after the memorized word according to the part of speech of the word immediately before the boundary of each word constituting the sentence represented by the input character string. Determining whether or not to insert a pause at the word boundary based on a numerical value and a numerical value representing ease of insertion of a pause immediately before the memorized word according to the part of speech of the word immediately after the word boundary. A pose insertion position determination method featuring: