JPH03141399A - Voice parameter coupling system - Google Patents

Abstract

PURPOSE:To allow the synthesis of voices to meet the requirements for naturality and distinctness by selectively using plural different tables and cou pling the voice parameters between elementary pieces. CONSTITUTION:The connection point position of the preceding elementary piece A is indicated by the frame number (a) from the final frame and the connection point position of the succeding elementary piece B is indicated by the frame number (b) from the top frame. The method of connection is indicated by an interpolation function (f). The most adequate connection parameters for respective conditions and characteristics are determined at this time and the respective corresponding tables, for example, (a), (b) are prepd. The voice parameter in the i-th frame (0<=i<=N1) of the connecting part is thereafter calculated. The connecting parameter, such as connecting method and connection point position, are selected by the table number (t) stored in a memory device and are used for each of the characteristics, such as use objects of the synthesized voices, required speaking tone and speaker's characteristics. The distinctness and smoothness of the synthesized voices are obtd. with the balance meeting the condition in this way.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a speech parameter combination method for a regular speech synthesizer.

As the use of computers to process Japanese, such as word processors and translation devices, has become widespread, the need for devices that can output arbitrary speech has increased.

Speech output devices are required to have synthesized speech that is clear, fluent and easy to listen to, and has rich sound quality. However, there are many unresolved problems with arbitrary speech synthesis technology, such as the synthesized speech lacking clarity and the sense of rhythm typical of Japanese.

As a speech parameter combination method for regular speech synthesis, for example, ① "Hakode, Ishikawa, LSi) - Study of parameter combination method in CV method.

Acoustical Society of Japan Proceedings l-2-18 (October 1982)
``Moon'', ■``Arai, LSP-C using CV syllable placement rules
V-rule speech synthesis, Journal of the Institute of Electronics, Information and Communication Engineers, V01j7
0-A, N (15, PP836-843”, ■ “Saito,
Oshima, Ruled Speech Synthesis Using Applied Deformation of Elements, Proceedings of the Acoustical Society of Japan 3-4-11 (October 1986) j
etc. are known. In ■, the parameters between the segments are connected with a straight line (however, the position of the connection point is set appropriately depending on the type of phoneme), and in ■, the beginning of the subsequent segment is connected. A bias '7 L/- frame is set nearby, and the parameters of the previous frame are replaced with the parameters of the bias frame and connected. We are trying to realize this.

Therefore, there are two main challenges in connecting speech parameters between segments: one is to be able to clearly express phonology, and the other is to connect smooth segments.

In many cases, these two requirements are contradictory, and it is difficult to completely satisfy both at the same time. In all of the above-mentioned conventional techniques, it is necessary to set connection methods and parameters that moderately satisfy both requirements. However, the balance between these two requirements usually differs depending on the purpose of use of the synthesized voice, the tone required of the synthesized voice, the speaker characteristics required of the synthesized voice, etc. For example, when using it for proofreading a manuscript, it is necessary to pronounce the phonemes clearly, even at the expense of smoothness. Furthermore, if the purpose is to listen to a long sentence and understand its meaning, phonology is not so important, but listening to unsmooth speech for a long time causes listeners to become fatigued.

Aim - The present invention was made in view of the above-mentioned circumstances, and
In particular, this was done with the aim of achieving a balance between clarity and smoothness of synthesized speech that suits the situation.

In order to achieve the above object, the present invention reads out a parameter sequence of a speech segment prepared in advance according to an input symbol string expressing phoneme/prosody, and adjusts the speech segment parameters by a speech parameter combination rule. In a regular speech synthesizer that connects sequences and synthesizes speech by adding prosody according to prosody control rules, connection parameters such as connection method and connection point position when combining speech parameters are prepared in a table and multiple storage methods are used. Speech parameters are combined by selectively using different tables from among the tables created, depending on the purpose of use of the synthesized voice, the tone required for the synthesized voice, the speaker characteristics required for the synthesized voice, etc. It is characterized by

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

FIG. 1 is a flowchart for explaining one embodiment of the present invention, and the specific process flow will be explained below with reference to FIG.

First, the table number t is set based on the purpose of use of the synthesized speech, the tone required for the synthesized speech, the speaker characteristics required for the synthesized speech, etc. Here, the purposes of using synthesized speech include, for example, ``some kind of education or instructions to listeners,'' ``accurate transmission of information such as news and databases,'' ``emergency warnings or guidance,'' and ``dialogue with humans.'' ”, etc. The tone required for synthesized speech is, for example, "command tone,"
These include ``between questions'', ``reading tone'', ``speech that makes one feel like crying'', ``speech that is sweet'', and ``speech that is filled with anger''. The speaker characteristics required for synthesized speech include "gender", "age", and "
Examples include ``place of birth'', ``a way of speaking unique to a specific occupation (bus guide, rakugo storyteller, announcer, etc.)'', ``a specific individual'', etc.

Further, the table stores connection parameters such as the position of connection points and connection method when connecting the pieces.

FIG. 2 shows an example of these connection parameters.

In FIG. 2, A represents a preceding elemental piece, B represents a subsequent elemental piece, and ■ represents a connecting part. Further, N represents the number of frames in the # portion.

Figures 3(a) and 3(b) are diagrams showing examples of multiple tables prepared depending on the situation and characteristics, respectively.Here, to explain the connection parameters stored in the table, a is the connection point position of the preceding segment. is expressed as the number of frames from the last frame, b is the connection point position of the subsequent segment expressed as the number of frames from the first frame, and f
represents the connection method using an interpolation function.

Determine the most appropriate connection parameters for each situation and characteristic as described above, and prepare tables corresponding to each.

After this, the i-th frame of the connection part (0≦l<N I
) is calculated by the following formula.

Note that X[#] means the audio parameter of X in the i-th frame. At this time, connection parameters such as the connection method and connection point position are stored in the storage device as multiple tables for each characteristic such as the purpose of use of the synthesized voice, the tone required for the synthesized voice, and the speaker characteristics required for the synthesized voice. Select and use the stored items based on the table number.

Furthermore, a prosody pattern is set, audio signals are synthesized using a digital filter, etc., and synthesized speech is obtained using a D/A converter, but since these methods are well known, detailed explanation will be omitted.

As is clear from the explanation in Section 2, according to the present invention, a plurality of speech patterns can be used depending on the purpose of use of the synthesized speech, the tone required for the synthesized speech, the speaker characteristics required for the synthesized speech, etc. By selectively using different tables to combine the speech parameters between segments, it is possible to synthesize speech that satisfies different demands for naturalness and clarity depending on the utterance [1-6 B key, speaker characteristics, etc.].

[Brief explanation of the drawing]

FIG. 1 is a flowchart for explaining one embodiment of the present invention, FIG. 2 is a diagram showing an example of connection parameters, and FIG.
Figures (a) and (b) are diagrams showing tables storing connection parameters, respectively.

Claims (1)

[Claims] 1. Read out the parameter series of speech segments prepared in advance according to input symbol strings expressing phoneme and prosody, connect the speech segment parameter series using speech parameter combination rules, and add prosody using prosody control rules to create speech. In a rule speech synthesis device that synthesizes speech parameters, a table is prepared that stores connection parameters such as connection method and connection point position when combining speech parameters, and from among the multiple stored tables,
It is characterized by selectively using different tables to combine voice parameters depending on the speech rate of the synthesized voice, the purpose of use of the synthesized voice, the tone required for the synthesized voice, the speaker characteristics required for the synthesized voice, etc. Audio parameter combination method.