JPS63131195A - Voice synthesizer - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION [Object of the Invention] (Industrial Field of Application) The present invention relates to speech synthesis in which an arbitrary input character string is systematically synthesized and converted into speech.

(Prior Art) A speech synthesis device that synthesizes an arbitrary input character string in a regular manner and converts it into speech analyzes the input character string to obtain a phonetic symbol string and prosodic information, and based on the phonetic symbol string, C( consonant)V(
The system selects and connects syllable parameters (vowels) and generates and outputs synthesized speech that takes into account the prosodic information.

By the way, when looking at the spectral envelope characteristics of natural speech, even for the same C syllable, the changes in the spectral envelope characteristics at the junction of two syllables can vary considerably depending on the type of vowel added immediately before. ing. However, since conventional speech synthesis devices combine syllables by simple interpolation, they have the disadvantage of lacking smoothness at syllable connections.

(Problem to be Solved by the Invention) In this way, in the conventional speech synthesis device, simple interpolation and combination are performed at the connection part of the C syllable, so the change in the spectral envelope characteristic at that part is similar to that of the spectral envelope characteristic of natural speech. The sound was different from the previous one, and the sound was dull and unnatural.

SUMMARY OF THE INVENTION An object of the present invention is to provide a speech synthesizer that can eliminate the above-mentioned drawbacks and obtain high-quality synthesized speech with natural continuity.

[Structure of the Invention] (Means for Solving the Problems) The present invention includes a character string analysis device that analyzes an input character string to generate a phonetic symbol string and prosody information, and a writing W1 device that stores syllable parameters. a speech parameter string generation device that generates a speech parameter string from the generated phonetic symbol string with reference to the syllable parameters; and a prosodic parameter string generation device that generates a prosodic parameter string based on the generated prosodic information. and a speech synthesizer that generates and outputs synthesized speech according to the speech parameter string and the prosodic parameter string, wherein the syllable parameter represents a syllable without a vowel immediately before the syllable parameter; The system is characterized by the provision of a syllable parameter representing a syllable to which the transition part of the preceding vowel is added.

(Function) In this invention, when generating a speech parameter string, the syllable before the input character string is considered, and if the input character string is the beginning of a word, a syllable parameter without a vowel is written immediately before it.
If the syllable is not at the beginning of a word, it looks at the syllable immediately before the syllable to be combined, and does not retrieve the syllable parameter to which the transition part with the vowel of the syllable immediately before is added from the storage device. Then, these extracted syllable parameters are interpolated and combined to generate a speech parameter sequence.

According to this invention, when the input character string does not have the beginning of a word, the syllable parameter to be retrieved is a syllable parameter that includes a preceding syllable. This is an audio parameter string containing information on the spectral inclusive characteristics of a portion. Therefore, the synthesized speech that is output is of high quality and has natural continuity.

(Example) An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing a speech synthesis device according to this embodiment.

The character string analysis device 1 analyzes an input character string and generates a phonetic symbol string and a prosodic symbol string. Among these, the phonetic symbol string is sent to the speech parameter string generation device 2, and the prosodic symbol string is sent to the prosodic parameter string generation device @3.

The speech parameter string generating device 2 generates a speech parameter string expressing the vocal tract characteristics of speech according to the input phonetic symbol string. The method of its generation is as follows. In addition,
Here, consonants (C) and vowels (V) are used as the middle level of speech synthesis.
Use Cv syllables consisting of combinations of. For example, if the input character string is "sunflower", the phoneme sequence at this time is represented by [hi rma wa ri]. Here, /h/, /l/.

/v/, /r/ are the phonetic symbols of consonants, /i/, /a
/ is the phonological symbol for a vowel. The speech parameter string generation device 2 converts the input character string into [hi-n+a-wa-r
i] (* indicates a syllable break).

Next, according to the divided single syllable units, syllable parameters are sequentially retrieved from the file in which the syllables are parameterized and combined, but in this Hlit, there are not one type of syllable parameter storage device used here, but six types, and the syllable parameters are combined. One storage device can be selected from six types of syllable parameter storage devices by checking whether the syllable to be used is at the beginning of a word or not, and if it is not at the beginning of a word, by referring to the type of vowel of the syllable immediately before the syllable.

Next, the six types of syllable parameter storage devices 4 to 9 will be explained.

C■Syllable parameter record/inhibition device 4 is a syllable parameter file created by analyzing a single syllable of natural speech uttered alone by an announcer, for example, and is a syllable parameter file created by analyzing a single syllable of natural speech uttered alone by an announcer. It stores the syllable parameters for the beginning of a word, which are parameterized.

In addition, other five vCV syllable parameter storage devices 4 to 9
is, for example, the necessary back syllable (CV) from a two-syllable natural voice uttered by an announcer with the vowel /a/, /i/, /u/, /e/, 10/ immediately before it.
This is a vCV syllable parameter file created by cutting out and analyzing the VCV including the transition part (V) with the preceding vowel. For example, the vowel /a / the syllable with the transition part added /
To explain the case of creating vCV sound D parameters for Wa/ using Figure 2, first, tell the announcer /a
Wa/ is uttered to obtain a natural speech spectrum as shown in FIG. Here, ra J "waJ" is the syllable part, and [a sword is the transition part of the voice from ra J to "14".

The conventional CV parameter storage device stores only the analysis parameters of the "a" and [waJ parts. However, here, the voice transition part “a” from raJ to r14J is further explained.
The analysis result of "o" is also stored as a vCV syllable parameter file in the form of [a'14aJ, which includes several frames.

Next, we will explain how to generate speech parameters from a phoneme sequence divided into syllables.

When the syllable to be combined is at the beginning of a word, the syllable parameter string is retrieved from the CV syllable parameter storage device 4, and when it is not at the beginning of the word, the syllable immediately before the syllable to be combined is focused on, and the transition part for the same vowel as that vowel is included. A syllable parameter string is retrieved from the vCV syllable parameter storage devices 5-9.

For example, the single sound ``Himawari'' [hi・rm
a -wa -ri], for example, /
hi/ is a CV sound immediate parameter storage device 4 for the beginning of a word fi
t and other syllable parameter strings are retrieved, /1a/ is retrieved from the VCV syllable parameter storage device 6 whose immediately preceding vowel is /1/, and /V4a/ and /rr/ are retrieved from the vCV syllable parameter storage device 6 whose immediately preceding vowel is /a/. A syllable parameter string is extracted from each. As a result, /ia/, /wa/, /r
For i/, a syllable parameter sequence is obtained in which there is a transition part such as [do J, ral] at the beginning. Then, these syllable parameter sequences are interpolated and continued in the interpolation interval indicated by diagonal lines to generate a speech parameter sequence. Incidentally, in the conventional syllable parameter recording device, only parameters without a spanning part are used as shown in FIG. 4, and crossing parts are also generated only by performing interpolation in the interpolation interval.

The speech parameter sequence generated by the method described above is input to the speech synthesizer 10. On the other hand, the I11 Ritsuki Monthly is converted into a prosodic parameter string by the prosodic parameter generation device 3, but this prosodic parameter string is also converted to the speech synthesizer 10.
input. Then, the speech synthesizer 10 operates based on the input of the above two types of parameter sequences, and outputs smooth and highly natural synthesized speech.

As described above, according to the speech synthesis device according to the present embodiment, the CV syllable parameter file created from the analysis of a single syllable uttered in isolation as a word-initial character is used, with an eye to the articulatory combination phenomenon of speech. By using five types of vCV syllable parameter files including front and back parts, a speech parameter string is generated by performing interpolation and combination taking into account the surrounding environment of the syllable. For this reason, CV
The change in the spectral envelope characteristics of the connecting portion of syllables becomes close to that of natural speech, resulting in the excellent effect of obtaining high-quality synthesized speech with natural continuity.

[Effects of the Invention] As described above, according to the present invention, since the syllables are continuous by interpolation processing including the parts where they cross each other, it is possible to produce a more natural sound.

[Brief explanation of the drawing]

Fig. 1 is a block diagram of a speech synthesis device according to an embodiment of the present invention, Fig. 2 is a speech analysis diagram for explaining the procedure for creating a syllable parameter file in the same device, and Fig. 3 is a speech parameter diagram in the same device. FIG. 4 is a diagram for explaining the procedure for creating a sequence of speech parameters in a conventional speech synthesis device. DESCRIPTION OF SYMBOLS 1... Character string analysis device, 2... Audio parameter string generation device, 3... Prosody parameter string generation device, 4...
Cv syllable parameter record device, 5 to 9 vCV syllable parameter storage device, 10 speech synthesizer. Applicant's Representative Patent Attorney Takehiko Suzue Power □ Time Figure 2 Figure 3 Figure 4

Claims (2)

[Claims] (1) A character string analysis device that analyzes an input character string to generate a phonetic symbol string and prosodic information, a syllable parameter representing a syllable without a vowel immediately before it, and a syllable with a transition part of the previous vowel added. a storage device that stores a syllable parameter representing a syllable; a phonological parameter string generation device that generates a phonological parameter string from the generated phonetic symbol string by referring to the syllable parameter; a prosodic parameter sequence generation device that generates a sequence;
A speech synthesis device comprising a speech synthesizer that generates and outputs synthesized speech according to the speech parameter string and the prosodic parameter string. (2) The syllable parameter representing the syllable without a vowel immediately preceding it is created from monosyllabic analysis of naturally occurring natural speech, and is a syllable representing the syllable to which the transition part of the immediately preceding vowel is added. The parameters are generated by analyzing natural speech of two syllables with a vowel immediately before it, including the transition part between the vowel and the previous vowel. Speech synthesizer.