JPH04190398A - Sound synthesizing method - Google Patents

Abstract

PURPOSE:To obtain a sound quality resembling a recorded composite tone by determining a phoneme boundary in the environment of a vocalized word in accordance with phoneme time length data obtained by analyzing an actual voice and a pitch pattern obtained by analyzing an actual voice pitch pattern. CONSTITUTION:A vocalizer vocalizes a word or a sentence stored in a storage 21, and the pitches of the word or the sentence are analyzed by a pitch analyzing part 22. Further, phoneme time length data stored in storage part 23 for each phoneme environment can be obtained by previously analyzing huge actual voices in a time length pattern data base 4, and are dependent upon the data in the storage part 23 and the result of analysis by the pitch analyzing part 22. The pitch pattern is inputted to a time length normalizing part 24 so as to a phoneme boundary in the phoneme environment of the vocalized word. Further, the pitches at the time when each phoneme is quarterly divided are stored as a composite tone pitch pattern in a pitch pattern data base 6, and accordingly, a composite tone is obtained therefrom. Thereby it is possible to obtain a sound quality resembling a recorded composite sound.

Description

DETAILED DESCRIPTION OF THE INVENTION A. Field of Industrial Application This invention relates to a method of synthesizing speech from text containing kanji and kana, and more particularly to a method of generating a database of pitch patterns of speech.

B. Summary of the Invention This invention provides a method for generating a database of speech pitch patterns, which determines phonological boundaries from pitch patterns obtained by analyzing the pitch of uttered words and phonological duration data for each phonological environment, and generates a pitch pattern database. By generating , it is possible to obtain a rule synthesis method with sound quality close to that of recording synthesis.

C. Conventional technical rule speech synthesis is a method of synthesizing a single sentence of arbitrary words into speech from text containing kanji and kana.

FIG. 3 is an explanatory diagram showing an outline of a simple voice synthesis device.

First, a text input into the text input section 1 containing kanji and kana is converted into a phoneme symbol string by the Japanese 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 data hese 4 and generates a time length pattern indicating the duration of the audio.

Similarly, the pitch pattern generation unit 5 refers to the pitch pattern database 6 and generates a pitch pattern indicating the pitch of the voice.

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 manner, the speech synthesis section 9 refers to the speech database IO and synthesizes a speech waveform. Note that 11 is an audio output unit that outputs synthesized audio.

In the speech synthesis device configured as described above, the pitch pattern for the sentence "Aoi", for example, generated by the pitch pattern generation unit 5 with reference to the pitch pattern database 6, is based on the I mora as shown in FIG. 4
The pitch target value of the points was given and synthesized. For this reason, conventional speech synthesizers suffer from monotonous and unnatural intonation.

D. Problems to be Solved by the Invention Although the conventional speech synthesis device described above has the above-mentioned problems, the sound quality of the device has improved in recent years and is reaching a level where there is no problem in terms of clarity.

However, when compared to so-called recording synthesizers that compress the original sound to some extent, the sound quality is slightly lower.

However, since recording synthesis cannot make any sentence uttered, it requires a huge amount of memory and is expensive in order to utter a large number of sentences. In addition to this drawback, there are also the following problems.

(b) When changing one word in a certain sentence to another, it must be recorded each time, which is very complicated and expensive.

(b) It is not always possible to record from the same speaker who has already recorded.

(c) Even if a sound can be recorded from the same speaker, the intonation may differ from the already recorded word or sentence.

(d) As a result of (b) and (c) above, the intonation of the entire sentence becomes unnatural, reducing the naturalness and clarity of the synthesized speech.

The present invention has been made in view of the above circumstances, and an object of the present invention is to provide a speech synthesis method having a sound quality close to that of recording synthesis.

E. Means for Solving the Problems This invention analyzes text input of sentences containing kanji and kana in a Japanese language processing unit, converts it into a phoneme string, and then converts each time length pattern, pitch pattern, and energy pattern based on this phoneme string. In the method of generating synthesized speech based on the generated patterns by referring to the data storage section, the pitch of the uttered word is analyzed, and this pitch analysis pattern and the time length pattern stored in the data base are The present invention is characterized in that a pitch pattern database is generated which is referred to when generating a pitch pattern by determining phoneme boundaries from phoneme duration data for each phoneme environment.

F. Have the speaker say the words and sentences to be uttered and analyze their pitch. A phonological boundary is determined from the phonological environment of the uttered word using the pitch pattern obtained as a result of the analysis and the phonological time length data for each phonological environment obtained in advance by analyzing real speech. Then, the pitch at the time when each phoneme is divided is stored as an intonation pattern at the time of synthesis, and this pattern is used to obtain a synthesized sound.

G Example Embodiments of the present invention will be described below based on the drawings.

In FIG. 1, 21 is a storage unit for words or sentences to be uttered. A speaker is asked to utter the words or sentences in this storage unit 21, and the pitch analysis unit 22 analyzes the pitch. Reference numeral 23 denotes a phoneme time length data storage unit for each phoneme environment, and the phoneme time length data in this storage unit 23 is obtained by previously analyzing a huge amount of real speech into the time length pattern database 4 shown in FIG. It is. The data in the storage unit 23 and the pitch pattern shown in FIG. 2 based on the results analyzed by the pitch analysis unit 22 are input to the time length normalization unit 24,
Here, the phonological environment of the uttered word and the phonological boundaries are determined.

Then, the pitch at the time when each phoneme is divided into four is stored in the pitch pattern database 6 as a pitch pattern at the time of synthesis, and this is used to obtain a synthesized sound.

If the pitch pattern database obtained as described above is used, "a" and "o≦" as shown in FIG.

Phonological boundaries such as :i=・ can be recognized correctly, and the pitch target value can also be positioned accurately. Therefore, the synthesized sound has a natural intonation, and the sound quality is close to that of recorded synthesis.

Effects of the Invention As described above, according to the present invention, phonetic duration data by phonetic environment obtained by analyzing real speech, and pinch patterns obtained by analyzing pitch patterns of real speech. By determining phonological boundaries from the phonological environment of spoken ejectives and synthesizing them with pitch patterns, a regular speech synthesis method can be obtained that can obtain sound quality close to that of recorded synthesized speech.

[Brief explanation of the drawing]

FIG. 1 is a schematic configuration diagram showing an embodiment of the present invention, FIG. 2 is an explanatory diagram of pitch patterns obtained by pitch analysis according to the embodiment of this invention, and FIG. 3 is a schematic configuration diagram of a conventional regular speech synthesis device. FIG. 4 is an explanatory diagram of a pitch pattern used in a conventional regular speech synthesizer. 21. Storage unit for words or sentences to be uttered, 22.
・Pitch analysis section, 23 - Phonological duration data storage section, 24
...Time length normalization section. Fig. 1 6 Pitch pattern data/\-space 21 Storage section 22 for words or sentences that occur without occurrence Pitch analysis section 23 Phonological time length nata storage section 24 Time length normalization section Fig. 3 4 Time length pattern/database 7 , 6 pitch butter 7 database 8 energy pattern database 10 audio database 11 audio output section

Claims (1)

[Claims] (1) The Japanese processing unit analyzes the text input of a sentence containing kanji and kana, converts it into a phoneme string, and generates a time length pattern, pitch pattern, and energy pattern based on this phoneme string by referring to each database, In the method of generating synthetic speech based on these generated patterns, the pitch of the uttered word is analyzed, and the pitch of the uttered word is analyzed from the pitch analysis pattern and the phonetic duration data for each phonetic environment stored in the duration pattern database. A speech synthesis method characterized in that a pitch pattern database is generated which is referred to when generating pitch patterns by determining phoneme boundaries.