JPH04214600A - Sound synthesizing method - Google Patents

Abstract

PURPOSE:To enhance the naturality and crispness of synthesized sounds by previously forming the pitch fluctuation data of original sounds in a synthesizing section and superposing this data on the target pitch value at the time of synthesis. CONSTITUTION:A fluctuation data section 21 is previously formed in the voice synthesizing section 9. The average pitch values of the intervals between the n points of the pitch patterns of the original sounds of the synthesized sounds and the differences between the average pitch values and the respective pitches are stored as the fluctuation data of respective pieces of the data into the fluctuation data section 21. The data of the fluctuation data section 21 obtd. in such a manner are superposed with the pitch patterns formed in a pitch pattern forming section 5 in a superposing section 22 at the time of sound synthesis and the monotonousness of the pitches is decreased.

Description

[Detailed description of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of the Invention This invention relates to a speech synthesis method for synthesizing speech from text containing kanji and kana.

0002

2. Description of the Related Art Ruled speech synthesis is a means of synthesizing arbitrary words, sentences, etc. into speech from text mixed with kanji and kana. FIG. 3 is an explanatory diagram showing an outline of a general speech synthesis device. First, the text input to the text input section 1 is converted into a phoneme symbol string by the Japanese language processing section 2. Next, a prosodic pattern (duration pattern, pitch pattern, energy pattern) is generated from this phoneme symbol string. That is, the time length pattern generation unit 3 refers to the time length pattern database 4 and generates a time length pattern indicating the duration of the audio.

Similarly, the pitch pattern generator 5 generates
Referring to the pitch pattern database 6, a pitch pattern indicating the pitch of the voice is generated. Similarly, the energy pattern generation unit 7 refers to the energy pattern database 8 and generates an energy pattern indicating the strength of the voice. Based on each phoneme pattern obtained in this way, the speech synthesis section 9 refers to the speech database 10 and synthesizes a speech waveform. Note that 11 is an audio output unit that outputs synthesized audio.

[0004] The speech synthesizer configured as described above outputs synthesized speech through processing as shown in FIG. At this time, the intonation (pitch) pattern shown in FIG. 4 was used. That is, in Figure 4, 1
Four pitch target values P10 to P34 are given for each mora, interpolation is performed between the target values in a straight line, and the pitch of each frame is finally given and supplied to the speech synthesis section 9.

[0005]

[Problem to be solved by the invention] In the intonation control method as described above, if the time length of each mora is short, it will not be affected much, but if it becomes longer, the time taken to synthesize pitches of similar values will increase. I end up. In other words, there is a problem that the pitch changes become monotonous, the synthesized sound tends to become mechanical, and the clarity also deteriorates. Furthermore, even when the mora time length is relatively short, there is a problem in that the pitch pattern is monotonous with little variation when compared to the pitch pattern of an actual human voice.

The present invention has been made in view of the above circumstances, and provides a speech synthesis method that improves the naturalness and clarity of synthesized speech by imparting natural fluctuations to the pitch of synthesized speech. The purpose is to

[0007]

[Means for Solving the Problems] In order to achieve the above object, the present invention analyzes a text input of a sentence containing kanji and kana in a Japanese processing unit, converts it into a phoneme string, and uses the phoneme string to calculate the time. In a method in which long patterns, pitch patterns, and energy patterns are generated by referring to each database, and synthesized speech is generated in a speech synthesis section based on these generated patterns, pitch fluctuation data of the original sound is stored in the speech synthesis section in advance. This data is created in advance, and this data is superimposed on the pitch target value of the pitch pattern at the time of synthesis.

[0008]

[Operation] When fluctuation data is superimposed on the pitch pattern, the monotony of the pitch is reduced and the naturalness of the synthesized sound is improved.

[0009]

DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. The same parts as those in FIG. In FIG. 1, 21 is a fluctuation data section created in advance in the speech synthesis section 9;
1 is created as follows. For example, when the pitch pattern of the original sound of the synthesized sound ``a'' is as shown in FIG.
Store in. Thereafter, fluctuation data is obtained for each sound in the same manner and stored in the fluctuation data section 21.

Fluctuation data section 21 obtained as described above
The data is superimposed on the pitch pattern generated by the pitch pattern generation section 5 at the time of speech synthesis in the superimposition section 22. This makes it possible to reduce pitch monotony, which is a drawback of the conventional method, and improve the naturalness of synthesized sounds.

[0011]

[Effects of the Invention] As described above, according to the present invention,
By superimposing the fluctuation data on the pitch pattern, there is an advantage that the naturalness of the synthesized sound can be improved and the clarity can be improved.

[Brief explanation of the drawing]

FIG. 1 is a schematic configuration diagram showing an embodiment of the present invention.

[Fig. 2] Pitch pattern diagram of the original sound of the synthesized sound.

FIG. 3 is a schematic explanatory diagram of a general speech synthesis device.

[Fig. 4] Pitch pattern diagram of synthesized sound.

[Explanation of symbols]

5...Pitch pattern generation section 9...Speech synthesis section 11...Audio output section 21... Fluctuation data section 22...Superimposed part

Claims (1)

[Claims] [Claim 1] A text input containing kanji and kana is analyzed by a Japanese processing unit and converted into a phoneme string, and based on this phoneme string, a time length pattern, a pitch pattern, and an energy pattern are generated by referring to each database. However, in the method of generating synthesized speech in the speech synthesis section based on these generated patterns, pitch fluctuation data of the original sound is created in advance in the synthesis section, and this data is superimposed on the pitch target value during synthesis. A speech synthesis method characterized by the following.