JP2740510B2 - Text-to-speech synthesis method - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a text-to-speech synthesis system.

Prior Art When synthesizing long sentences by text-to-speech synthesis,
The synthesized speech was very monotonous compared to human speech, and prolonged listening was painful. In addition, in sentences written in characters, in addition to character information, horizontal lines, cross points, angle brackets,
Various emphasis information such as bold, enlarged characters, shading, and modified fonts are usually included, thereby providing important information in a form that is easy for the reader to understand. However, in the conventional text-to-speech synthesis method, such emphasis information is ignored and only character information is used as input information. For this reason, there is no distinction between the emphasized portion and the non-emphasized portion in the output synthesized speech, and the synthesized sound is more monotonous.

Objective The present invention has been made in view of the above-mentioned circumstances, and in particular, in text synthesis of speech, a synthesized speech that tends to be monotonous, by emphasizing important information contained therein and uttering the speech, The purpose is to make it easy to hear.

To achieve the above object, the present invention provides a morphological analysis unit, a phoneme / prosodic symbol generation unit, a prosody table, a prosody generation unit, an emphasis bias table, an emphasis parameter generation unit, an adder, a speech unit file, a phoneme parameter. A text-to-speech synthesis device including a generation unit and a speech synthesizer, wherein the morphological analysis unit outputs morphological information by performing a morphological analysis process on the input text; and the phonological / prosodic symbol generating unit includes a morphological information unit. A prosody symbol string and a phoneme symbol string are generated from the prosody table, the prosody table stores prosody parameters corresponding to the prosody symbol string, and the prosody generation unit reads the prosody parameter from the prosody symbol string with reference to the prosody table,
The emphasis bias table stores emphasis prosody parameters corresponding to emphasis information included in the text, and the emphasis parameter generation unit refers to the emphasis bias table and emphasizes the prosody of a portion where the emphasis information included in the text emphasizes. Generating a parameter, the adder adds a prosodic parameter and an enhanced prosodic parameter, and outputs the sum as a prosodic parameter. The speech unit file stores a phoneme parameter corresponding to a phoneme symbol string. The generation unit refers to the speech unit file, reads a speech unit parameter from the input phoneme symbol string, and the speech synthesizer combines the speech unit parameters according to a combination rule while taking into account the added prosody parameter. It is characterized by generating synthesized speech. Hereinafter, a description will be given based on an example of the present invention.

FIG. 1 is a block diagram for explaining an embodiment of the present invention, in which 1 is a morphological analysis unit, 2 is a phoneme / prosodic symbol generation unit, 3 is a prosody generation unit, 4 is a prosody table, 5 Denotes an emphasis symbol generation unit, and 6 denotes an emphasis prosody generation unit. These emphasis symbol generation units 5 and 7 generate parameters to be emphasized. 7 is an emphasis bias table, 8 is an adder, 9
Is a phoneme parameter generation unit, 10 is a speech unit file, and 11 is a speech synthesizer. According to the present invention, the morphological analysis unit 1 outputs morphological information by performing morphological analysis processing on an input text, The prosody symbol generation unit 2 generates a prosody symbol sequence and a phoneme symbol sequence from the morphological information, the prosody table 4 stores prosody parameters corresponding to the prosody symbol sequence, and the prosody generation unit 3 , Read the prosody parameters from the prosody symbol string, the emphasis bias table 7 stores the emphasis prosody parameters corresponding to the emphasis information included in the text, and the emphasis parameter generation units 5 and 6 execute the emphasis bias table 7, the adder 8 generates a prosody parameter at a position where the emphasis information included in the text emphasizes, and the adder 8 generates the prosody parameter and the prosody parameter. The speech unit file 10 stores the phoneme parameter corresponding to the phoneme symbol string, and the phoneme parameter generation unit 9 refers to the speech unit file 10. ,
The speech synthesizer 11 reads out speech unit parameters from the input phoneme symbol string, and the speech synthesizer 11 combines the speech unit parameters according to a combination rule while taking into account the added prosodic parameters to generate a synthesized speech. By changing the prosody of the emphasis part to utter a sentence containing various emphasis information such as side-points, brackets, bold, enlarged characters, shading, deformed fonts, etc. Information is provided to the listener in an easy-to-understand format.

FIG. 2 to FIG. 4 are diagrams for explaining the embodiment of the present invention, and all show examples in which "this is meat called" marbling "" is uttered, and "marbling""
Here is an example when is highlighted. In the embodiment shown in FIG. 2, the fundamental frequency of the portion (each character of the character string) emphasized by the emphasis information is increased by giving a certain bias to that of the non-emphasized portion. is there. In the embodiment shown in FIG. 3, the power of the character string (marbling) emphasized by the emphasis information (“”) is increased by giving a certain bias to that of the non-emphasized portion, In the embodiment shown in the figure, the utterance speed of the portion (character string) emphasized by the emphasis information is lower than that of the non-emphasized portion, that is, the duration of each phoneme of the emphasized portion is longer than that of the non-emphasized portion. It was done.

Effects As is apparent from the above description, according to the present invention, it is possible to reduce the monotonicity of synthesized speech and emphasize important information to produce utterance. In particular, speech synthesis tends to be monotonous. It is possible to make the synthesized speech easy to hear by voicing with emphasis on important information contained in the synthesized speech. Therefore, by performing the emphasis process in parallel with the prosody generation process before the phoneme generation, Synthetic speech is closer to human utterance to improve naturalness,
The emphasis process is performed by addition, thereby simplifying the emphasis process.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an example of a text-to-speech synthesizing apparatus used for carrying out the present invention, and FIGS. 2 to 4 are views for explaining an embodiment of the present invention. 1 ... morphological analysis unit, 2 ... phonological prosodic symbol generation unit, 3 ...
... Prosody generation section, 4... Prosody table, 5... Emphasis symbol generation section, 6... Emphasis prosody generation section, 7... Emphasis bias table, 8.
…… Speech unit file, 11 …… Speech synthesizer.

Claims (1)

(57) [Claims] 1. A morphological analysis section, a phoneme / prosodic symbol generation section, a prosody table, a prosody generation section, an emphasis bias table, an emphasis parameter generation section, an adder, a speech unit file, a phoneme parameter generation section, and a speech synthesizer. In a text-to-speech synthesis apparatus, a morphological analysis unit outputs morphological information by performing morphological analysis processing on an input text, and a phoneme / prosodic symbol generation unit generates a prosodic symbol sequence from the morphological information,
A prosody table stores a prosody parameter corresponding to the prosody symbol string. The prosody generation unit reads the prosody parameter from the prosody symbol string with reference to the prosody table. The emphasis prosody parameter corresponding to the emphasis information included is stored. The emphasis parameter generation unit refers to the emphasis bias table and generates an emphasis prosody parameter of a portion where the emphasis information included in the text is emphasized. The parameter and the enhanced prosodic parameter are added and output as an added prosodic parameter. The speech unit file stores the phoneme parameter corresponding to the phoneme symbol string. The phoneme parameter generation unit refers to the speech unit file,
A text-to-speech synthesizer that reads out speech unit parameters from an input phoneme symbol string, and a speech synthesizer that combines speech unit parameters according to a combination rule while taking added prosody parameters into account.