JP6633466B2 - Pause length control device, pause length control method, and program - Google Patents

Description

  The present invention relates to a speech synthesis technique, and more particularly to a technique for controlling the length of a pause inserted between sentences to be synthesized.

  In recent years, the naturalness of synthesized speech corresponding to each sentence has been improved by introducing a corpus-based method and a statistical method. However, it is not clear how to control the length of the pause between sentences. According to the speech analysis, the length of the pause between sentences is not constant. To make the synthesized speech content, which is composed of multiple synthesized speeches with poses of appropriate and natural length interposed, closer to the speech content created by human narration, the length of the pause between sentences is appropriate. Control technology is required.

  Non-Patent Document 1 discloses knowledge for controlling the length of a pause between sentences. According to Non-Patent Document 1, as a result of analysis based on the layout structure of a document, 1) When the title is a sentence, the pause length between the title and the sentence is long. However, this is different when the title is a phrase, and 2) that the pause length between the sentences of the bulleted item takes a value almost constant. Based on these, it is possible to control the length of the pause between those sentences.

Hideharu Nakajima, Noboru Miyazaki, Sumyu Sakauchi, "Analysis of Pause Length in Speech Speaking to Elderly People," Proc. Of the Autumn Meeting of the Acoustical Society of Japan, 3-Q-49, September 2015

  However, the knowledge of Non-Patent Document 1 alone cannot control the length of the pause between sentences in a paragraph, which cannot be differentiated by the document layout structure.

  In view of the foregoing, it is an object of the present invention to provide a pause length control technique that can appropriately control the length of a pause between sentences based on the structure of each sentence to be combined.

  In order to solve the above-mentioned problem, a pause length control device according to the present invention is a model that stores an inter-sentence pause length prediction model that outputs a pause length between sentences according to a feature amount extracted from a dependency analysis result of a sentence. A storage unit, a feature amount extraction unit that extracts a feature amount of each sentence included in the synthesis target document from a dependency analysis result of each sentence included in the synthesis target document, and a synthesis target document using an inter-sentence pause length prediction model And a pose length predicting unit for predicting a pause length corresponding to each sentence included in the synthesis target document from the feature amount of each sentence included in the document.

  According to the present invention, the length of a pause between sentences can be predicted from only the structure information of the sentences before and after each pause position. As a result, the structure of the document can be conveyed to the listener based on the existence and length of the pose, and a higher understanding can be given.

FIG. 1 is a diagram illustrating a functional configuration of the pause length control device. FIG. 2 is a diagram illustrating a processing flow of a pause length control method. FIG. 3 is a diagram illustrating an example of an inter-sentence pause length prediction model.

  Hereinafter, embodiments of the present invention will be described in detail. In the drawings, components having the same function are denoted by the same reference numerals, and redundant description is omitted.

  The pause length control device according to the embodiment receives a dependency analysis result of each sentence of a document to be subjected to speech synthesis including a plurality of sentences (hereinafter, referred to as a synthesis target document) and generates a synthesis generated from the synthesis target document. This is a device that outputs the pause length between sentences in voice. As shown in FIG. 1, the pose length control device 1 includes a model storage unit 10, a feature amount extraction unit 11, and a pose length prediction unit 12. The pause length control method of the embodiment is realized by the pause length control device 1 performing the processing of each step described later.

  The pause length control device 1 is configured by reading a special program into a known or dedicated computer having, for example, a central processing unit (CPU), a main storage device (RAM: Random Access Memory), and the like. It is a special device. The pause length control device 1 executes each process under the control of a central processing unit, for example. The data input to the pause length control device 1 and the data obtained in each process are stored in, for example, a main storage device, and the data stored in the main storage device is read out as necessary and used for other processes. Used. Further, at least a part of each processing unit of the pause length control device 1 may be configured by hardware such as an integrated circuit. Each storage unit of the pause length control device 1 is, for example, a main storage device such as a RAM (Random Access Memory), an auxiliary storage device including a semiconductor memory device such as a hard disk, an optical disk, or a flash memory, or It can be configured by middleware such as a relational database or key-value store.

  The processing procedure of the pause length control method according to the embodiment will be described with reference to FIG.

  The result of the dependency analysis of each sentence of the document to be synthesized is input to the pause length control device. The dependency analysis result may be a dependency analysis result between clauses constituting each sentence, or a dependency analysis result between accent phrases forming each sentence. The dependency analysis method may use any existing dependency analysis technology. The input dependency analysis result is input to the feature amount extraction unit 11.

  The model storage unit 10 stores an inter-sentence pose length prediction model that outputs the inter-sentence pause length according to the feature amount extracted from the sentence dependency analysis result. Hereinafter, the feature amount extracted from the dependency analysis result and the sentence pause length prediction model will be described in detail.

  The feature amount extracted from the dependency analysis result will be described. From the dependency relationship between sentences, that is, the dependency relationship between the phrases that constitute the sentences on the front and rear sides of the pause position, the number of clause boundaries whose clause is not the immediately following phrase is counted as the number of indirect dependencies. If the input dependency analysis result is a dependency analysis result between accent phrases, the number of indirect dependencies is counted from the dependency relationship between accent phrases. The number of indirect dependencies is not only the sentence immediately before and after the sentence, but also the sentence before and after the sentence, the sentence before and after the sentence, and so on within a certain width. May be extracted from the sentence. That is, the length from the sentence N to the sentence N between the sentences whose pause length is to be predicted is set as a predetermined width, and the number of indirect dependencies extracted from each sentence included in the predetermined width is set. (N + M) sets are extracted as feature amounts. This feature can be represented by the following equation.

Here, x indicates the number of indirect dependencies of each sentence, and the subscript indicates the relative position with respect to the inter-sentence assuming that the interval between the sentences to be predicted is 0. That, x _-1 is the number of receive indirect dependency sentence of the previous sentences to be predicted, x ₁ is an indirect dependency number of statements immediately following, x _-N undergoes indirect dependency of N th previous sentence X _M is the number of indirect dependencies of the sentence after M. Note that the values of N and M may be different or the same.

The inter-sentence pause length prediction model is, for example, a regression equation in which the feature amount is an explanatory variable (input variable), and the pause length to be predicted is a dependent variable (output variable). Sentences to be predicted, i.e., when the pause length position between the x _-1 and x ₁ and Y, sentences pause length prediction model for predicting the Y can be represented by the linear equation below.

  Here, a is a coefficient of each explanatory variable x, and b is a constant term. These linear coefficients and constant terms can be calculated by a least squares method from a large amount of learning data in which the number of indirect dependencies and the pause length are paired. In addition to the linear regression equation, a non-linear regression equation having the same input variables and output variables, or a regression model based on machine learning such as a regression tree or a neural network may be used.

  The inter-sentence pause length prediction model may be, for example, a step-like function having only the feature amount obtained from the sentence immediately before the sentence to be predicted as an argument. FIG. 3 shows an example of the step function. In FIG. 3, the horizontal axis represents the feature quantity, that is, the number of indirect dependencies obtained from the sentence immediately before the sentence, and the vertical axis represents the pause length corresponding to the number of indirect dependencies. In the step-like function, for example, as shown in FIG. 3, when the number of indirect dependencies is 1 or 2, the same pause length (1.25 seconds) is used, and when the number of indirect dependencies is 3 to 5, another pause value is used. 1.45 seconds).

  The inter-sentence pause length prediction model has, for example, the height of the vertical axis in FIG. 3 as the average value of the pause lengths in which the number of indirect dependencies is 1 and 2, and has a variance thereof. It is also possible to model as a probability distribution such as a normal distribution according to the variance, and set the pose length stochastically.

  In step S1, the feature value extraction unit 11 extracts the feature value of each sentence included in the synthesis target document from the dependency analysis result of each sentence included in the input synthesis target document. Table 1 shows an example of the feature amount extracted from each sentence.

The sentence number is an index of each sentence included in the synthesis target document, and the feature amount is a feature amount corresponding to each sentence. In the example of Table 1, the number of indirect dependencies of each sentence is set as a feature amount. In the case where the number of indirect dependencies of a sentence included in a predetermined width is included in the feature amount, for a sentence having no preceding sentence, such as sentence 1 located at the beginning of the document, the number of indirect dependencies of the front N is determined by For example, 0 is set. The extracted feature amount is sent to the pose length prediction unit 12.

  In step S2, the pause length prediction unit 12 reads the inter-sentence pause length prediction model stored in the model storage unit 10, and uses the inter-sentence pause length prediction model to calculate the feature amount of each sentence included in the synthesis target document. Then, the pause length corresponding to each sentence included in the document to be synthesized is predicted. Table 2 shows an example of the pose length predicted from the example of the feature amount in Table 1.

  In the example of Table 2, the pause length prediction value represents the pause length between sentences immediately after the sentence on the same line in seconds. That is, the pause length between sentences 1 and 2 is 1.2 seconds, and the pause length between sentences 2 and 3 is 1.3 seconds.

  The pause length control device outputs a pause length between the sentences of the document to be synthesized, which is predicted by the pause length prediction unit 12. By synthesizing the document to be synthesized using the pause length, it is possible to generate a synthesized speech in which a pause having an appropriate length is inserted.

  With this configuration, the pause length control device of the present invention can predict the length of a pause between sentences only from the structure information of the sentences before and after each pause position. As a result, the structure of the document can be conveyed to the listener based on the existence and length of the pose, and a higher understanding can be given.

  As described above, the embodiments of the present invention have been described. However, the specific configuration is not limited to these embodiments, and the design may be appropriately changed without departing from the spirit of the present invention. It goes without saying that this invention is included. The various processes described in the embodiments may be executed not only in chronological order according to the order described, but also in parallel or individually according to the processing capability of the device that executes the processes or as necessary.

[Program, recording medium]
When the various processing functions of each device described in the above embodiment are implemented by a computer, the processing content of the function that each device should have is described by a program. Then, by executing this program on a computer, various processing functions in each of the above devices are realized on the computer.

  A program describing this processing content can be recorded on a computer-readable recording medium. As a computer-readable recording medium, for example, any recording medium such as a magnetic recording device, an optical disk, a magneto-optical recording medium, and a semiconductor memory may be used.

  The distribution of the program is performed by, for example, selling, transferring, lending, or the like, a portable recording medium such as a DVD or a CD-ROM on which the program is recorded. Further, the program may be stored in a storage device of a server computer, and the program may be distributed by transferring the program from the server computer to another computer via a network.

  A computer that executes such a program first stores, for example, a program recorded on a portable recording medium or a program transferred from a server computer in its own storage device. Then, when executing the processing, the computer reads the program stored in its own recording medium and executes the processing according to the read program. As another execution form of the program, the computer may directly read the program from the portable recording medium and execute processing according to the program, and further, the program may be transferred from the server computer to the computer. Each time, the processing according to the received program may be sequentially executed. A configuration in which the above-described processing is executed by a so-called ASP (Application Service Provider) type service that realizes a processing function only by executing the program and acquiring the result without transferring the program from the server computer to the computer. It may be. It should be noted that the program in the present embodiment includes information used for processing by the computer and which is similar to the program (data that is not a direct command to the computer but has characteristics that define the processing of the computer).

  In this embodiment, the present apparatus is configured by executing a predetermined program on a computer. However, at least a part of the processing may be realized by hardware.

DESCRIPTION OF SYMBOLS 1 Pose length control device 10 Model storage unit 11 Feature amount extraction unit 12 Pose length prediction unit

Claims (7)

A model storage unit that stores an inter-sentence pause length prediction model that outputs a pause length between sentences according to a feature amount extracted from a sentence dependency analysis result;
A feature amount extraction unit configured to extract a feature amount of each sentence included in the synthesis target document from a dependency analysis result of each sentence included in the synthesis target document;
A pose length prediction unit that predicts a pose length corresponding to between sentences included in the synthesis target document from a feature amount of each sentence included in the synthesis target document using the inter-sentence pause length prediction model;
Including
The inter-sentence pause length prediction model is a step function having only the feature amount extracted from the sentence immediately before the sentence to be predicted as an argument.
Pause length control device. A model storage unit that stores an inter-sentence pause length prediction model that outputs a pause length between sentences according to a feature amount extracted from a sentence dependency analysis result;
A feature amount extraction unit configured to extract a feature amount of each sentence included in the synthesis target document from a dependency analysis result of each sentence included in the synthesis target document;
A pose length prediction unit that predicts a pose length corresponding to between sentences included in the synthesis target document from a feature amount of each sentence included in the synthesis target document using the inter-sentence pause length prediction model;
Including
The feature amount is obtained by combining the indirect dependency number, which is the number of clause boundaries in which the preceding clause is not the immediately following clause between the clauses constituting each sentence, for a predetermined number of sentences before and after each sentence Is a thing,
Pause length control device. A model storage unit that stores an inter-sentence pause length prediction model that outputs a pause length between sentences according to a feature amount extracted from a sentence dependency analysis result;
A feature amount extraction unit configured to extract a feature amount of each sentence included in the synthesis target document from a dependency analysis result of each sentence included in the synthesis target document;
A pose length prediction unit that predicts a pose length corresponding to between sentences included in the synthesis target document from a feature amount of each sentence included in the synthesis target document using the inter-sentence pause length prediction model;
Including
The feature quantity is the number of indirect dependencies between the accent phrases that constitute each sentence, the number of accent phrase boundaries where the preceding accent phrase is not the immediately following accent phrase, and the number of indirect dependencies between each sentence. A combination of sentences,
Pause length control device. The model storage unit stores an inter-sentence pause length prediction model that outputs a inter-sentence pause length according to a feature amount extracted from a sentence dependency analysis result,
A feature amount extraction unit that extracts a feature amount of each sentence included in the synthesis target document from a dependency analysis result of each sentence included in the synthesis target document,
A pose length prediction unit predicts a pose length corresponding to each sentence included in the synthesis target document from the feature amount of each sentence included in the synthesis target document using the inter-sentence pause length prediction model ,
The inter-sentence pause length prediction model is a step function having only the feature amount extracted from the sentence immediately before the sentence to be predicted as an argument.
Pause length control method. The model storage unit stores an inter-sentence pause length prediction model that outputs a inter-sentence pause length according to a feature amount extracted from a sentence dependency analysis result,
A feature amount extraction unit that extracts a feature amount of each sentence included in the synthesis target document from a dependency analysis result of each sentence included in the synthesis target document,
A pose length prediction unit predicts a pose length corresponding to each sentence included in the synthesis target document from the feature amount of each sentence included in the synthesis target document using the inter-sentence pause length prediction model ,
The feature amount is obtained by combining the indirect dependency number, which is the number of clause boundaries in which the preceding clause is not the immediately following clause between the clauses constituting each sentence, for a predetermined number of sentences before and after each sentence Is a thing,
Pause length control method. The model storage unit stores an inter-sentence pause length prediction model that outputs a inter-sentence pause length according to a feature amount extracted from a sentence dependency analysis result,
A feature amount extraction unit that extracts a feature amount of each sentence included in the synthesis target document from a dependency analysis result of each sentence included in the synthesis target document,
A pose length prediction unit predicts a pose length corresponding to each sentence included in the synthesis target document from the feature amount of each sentence included in the synthesis target document using the inter-sentence pause length prediction model ,
The feature quantity is the number of indirect dependencies between the accent phrases that constitute each sentence, the number of accent phrase boundaries where the preceding accent phrase is not the immediately following accent phrase, and the number of indirect dependencies between each sentence. A combination of sentences,
Pause length control method. A program for causing a computer to function as the pause length control device according to any one of claims 1 to 3 .

Images