JPH0833745B2 - Speech synthesizer - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION Object of the Invention (Industrial field of use) The present invention relates to a speech synthesizer capable of effectively producing smooth synthetic speech.

(Conventional Technology) There is a voice synthesis output as a technology that plays an important role of a man-machine interface.

Conventionally, this voice synthesis has been created by editing the voice that has been recorded in advance, but there is a problem that the types and number of words and phrases that can be synthesized and output are limited.

Therefore, a method has been developed in which an arbitrary input character string is analyzed to obtain phonological information and prosody information thereof, and synthetic speech is generated from these information based on a predetermined rule. This method is called a rule synthesizing method, and has an advantage that a synthesized voice of an arbitrary word or phrase can be generated relatively easily.

However, there is a drawback in that the quality of the synthesized speech generated by this is lower than that of the speech generated by the recording and editing method described above. For example, in terms of intelligibility, it is possible to generate a synthesized voice having a considerably high quality, but there is a problem in smoothness and a problem that it is difficult to hear.

(Problems to be Solved by the Invention) The present invention has been made in consideration of such circumstances.
It is an object of the present invention to provide a voice synthesizing device capable of improving smoothness of regularly synthesized voices and improving audibility thereof.

[Configuration of the Invention] The present invention, when generating a voice parameter from a phoneme symbol string obtained by analyzing an input character string, considers an environment in which a voice unit, which is a unit of voice synthesis, is placed, and independently. A speech unit file for the beginning of a word that stores existing speech units, a speech unit file that stores the speech units determined according to the immediately preceding vowels, and a speech unit determined according to the unvoiced immediately preceding vowels. By using a parameter file consisting of the stored speech segment files, the speech parameters are obtained by selectively using the speech segment files according to the type of the vowel immediately preceding the phoneme symbol string.

Embodiment An embodiment of the present invention will be described below with reference to the drawings.

 FIG. 1 is a schematic configuration diagram of the apparatus of the embodiment.

A word or phrase provided for speech synthesis is input as a character string expressing it. The character string analysis unit 1 analyzes the input character string and generates a phoneme symbol string and a prosodic symbol string corresponding to the input character string, respectively.

The speech parameter sequence generation device 2 inputs the above phoneme symbol sequence, refers to the parameter file 3 (speech segment files 3a, 3b, to 3h), which is a unit of speech synthesis, to obtain the speech segment parameters, A speech parameter string that approximates the vocal tract characteristics of the speech is generated by combining the speech unit parameters. The combination of the above speech unit parameters is
For example, a linear interpolation method or the like is used.

Specifically, for example, when a speech unit is a syllable, for each syllable word detected from the phonological symbol string, the vowel of the syllable immediately before that is determined, and the speech unit file 3a, 3b, according to the determination result. The syllable parameters are obtained by selectively referring to ~ 3h, and these syllable parameters are combined to generate a speech parameter string.

On the other hand, the prosody parameter string generation device 4 generates the prosody parameter string according to the prosody symbol string obtained by the character string analysis unit 1.

The voice synthesizer 5 applies a predetermined synthetic voice rule in accordance with the voice parameter sequence and the prosody parameter sequence thus generated to generate a synthetic voice corresponding to the input character string.

Here, the eight speech unit files 3a, 3b, to 3h forming the parameter file 3 will be described.

The speech unit file 3a stores a speech unit for the beginning of a word, and stores, for example, a syllable unit obtained by cutting out the first syllable data of a single uttered word and analyzing it.

In addition, the speech unit files 3b to 3f obtain syllable data when the immediately preceding vowel is voiced, and the speech units obtained by analyzing this speech data are classified according to the type of the vowel. Is stored. For example, the last vowel is / i /
For the syllable speech unit, the natural speech such as "ike;ike" is analyzed, and the analysis result of the second syllable / ke / portion of the phonological symbol string "ike" is extracted and parameterized. The speech unit is required.

By such a voice segment parameter extraction process,
The voice unit file 3b is the voice unit when the last vowel is / a /, the voice unit file 3c is the voice unit when the last vowel is / i /, and the voice unit file 3d is the last vowel. Is / u /, the last vowel is / e in the speech unit file 3e.
The voice unit when / is stored, and the voice unit file 3f stores the voice unit when the immediately preceding vowel is / o /.

In the speech unit files 3g and 3h, syllable data when the immediately preceding vowel is unvoiced is obtained, and the speech units obtained by analyzing this speech data are classified according to the type of the vowel, Is stored. Specifically, the speech unit file 3g stores the speech unit when the immediately preceding vowel is unvoiced //, and the speech unit file 3h when the immediately preceding vowel is unvoiced // Is stored.

The speech segment in the case where the immediately preceding vowel is unvoiced is also performed in the same manner as the above analysis processing. For example, a natural voice such as “forehead; taitai” is analyzed, and the second part of the phoneme symbol string “hitai”
The analysis result of the syllable / ta / part is extracted, and the speech unit is obtained by parameterizing this.

Then, the speech parameter sequence generation device 2 determines the type of vowel existing immediately before each syllable of the given phoneme symbol sequence, and according to the determination result, the speech unit files 3a, 3b, to 3h described above are generated. It is selectively used to obtain the voice parameter.

Specifically, when an input character string "appropriate;precise" is given as shown in FIG. 2 (a), the phoneme symbol string [tekikaku] is first obtained. Here, the voice unit, which is a unit of voice synthesis, is, for example, a consonant (C) and a vowel (V).
Shall be defined by the combination with (CV). The voice parameter generation device 2 uses this phoneme symbol string [tekikaku].
The consonant phoneme symbol / t / / k / and the vowel phoneme symbol / e / / i / / a
/ / u / is judged respectively and this is divided into syllable units according to the above definition. If the syllables are delimited by [.], The phonological symbol string [tekikaku] is decomposed into four syllables [te.ki.ka.ku]. For each of these syllables, its syllable parameter is obtained as follows.

That is, since there is no vowel immediately before the first syllable [te], the syllable parameter is obtained using the speech segment file 3a for the beginning of the word. Next, since the vowel immediately before the second syllable [ki] is / e /, its syllable parameter is obtained using the speech unit file 3e. Similarly, for the third syllable [ka], since the vowel immediately before it is / i /, the speech unit file 3c
Is used to determine the syllable parameter. For the fourth syllable [ku], the vowel immediately before that is / a /, and therefore the syllable parameter is determined using the speech unit file 3b.

Then, these syllable parameters are combined to obtain a voice parameter string corresponding to the input character string "appropriate;

When an input character string “Bakufu; Bakufu” is given as shown in FIG. 2 (b), a syllable sequence [ba, ku, fu] is similarly obtained from the phonological symbol string, For the first syllable [ba], since there is no vowel immediately before that, the syllable parameter is obtained using the speech segment file 3a for the beginning of the word, and for the second syllable [ku], the immediately preceding vowel is /. Since it is a /, the syllable parameter is obtained using the speech unit file 3b, and the third syllable [fu]
With respect to, since the vowel immediately before it is unvoiced //, its syllable parameter is obtained using the speech unit file 3h.

Thus, according to the present apparatus configured as described above, the syllable parameter obtained for each syllable that is a speech unit is considered in consideration of the influence of the vowel of the syllable immediately before it and the influence of the environmental change. Fragment file 3
The synthesized syllables that are regularly synthesized based on the speech units obtained from a, 3b, and 3h are very smooth. Especially, not only the type of vowel existing immediately before but also the voiced or unvoiced vowel is used to select the voice unit file to be searched for that syllable. It will be smooth and highly natural.

Therefore, it is possible to easily and effectively obtain a synthetic voice having a very high intelligibility, smoothness and high naturalness.

The present invention is not limited to the above embodiment. Here, the description has been made assuming that the phoneme as a unit of rule synthesis is a syllable, but the phoneme may be a phoneme as well. In this case, it is a phoneme connection (C
V) and (VCV) should be taken into consideration when constructing the speech unit file. Further, in the embodiment, the voice unit files are separately configured according to the type of the voiced immediately preceding vowel,
For example, the vowel / i // e /, which is accompanied by the lateral spread of the mouth, is excluded from the characteristics of the near-term voice, and the / a / / o / / u / files are grouped together to simplify the configuration of the parameter file 3. It may be possible to achieve this. In addition, the present invention can be variously modified and implemented without departing from the gist thereof.

[Effects of the Invention] According to the present invention as described above, a speech unit parameter is obtained according to the type of vowel existing immediately before it, so that a smooth and highly natural synthesized speech according to the environment of the speech. Can be generated. Therefore, it becomes possible to improve the easiness of hearing, and a great effect can be obtained as a man-machine interface.

[Brief description of drawings]

FIG. 1 is a schematic configuration diagram of an embodiment apparatus of the present invention, and FIGS. 2 (a) and 2 (b) are diagrams schematically showing a process of generating a voice parameter sequence in the embodiment apparatus. 1 ... Character string analysis device, 2 ... Speech parameter string generation device, 3 ... Parameter file, 3a, 3b, to 3h ... Speech element file, 4 ... Prosodic parameter generation device, 5 ... Speech synthesizer .

Claims (1)

[Claims] 1. A character string analysis means for analyzing an input character string to obtain a phonological symbol string and prosodic information corresponding to the input character string, and a speech segment for the beginning of a word in which independently existing speech segments are stored. It consists of a file, a speech segment file that stores the speech segment determined according to the immediately preceding vowel, and a speech segment file that stores the speech segment determined according to the unvoiced immediately preceding vowel. A parameter file defined only by a combination (CV) of consonants (C) and vowels (V), and each phoneme for the phoneme symbol string obtained by the character string analysis means with reference to the parameter file. , And a voice parameter for generating a voice parameter sequence that approximates the vocal tract characteristics of the voice by selectively using the voice segment file according to whether the immediately preceding vowel is voiced or unvoiced. String generation means, prosody parameter string generation means for generating a prosody parameter string in accordance with the prosody information calculated by the character string analysis means, speech parameter string and prosody parameter string generation calculated by the speech parameter string generation means A voice synthesizing device comprising: a voice synthesizing unit for regularly synthesizing a voice according to the prosody parameter sequence obtained by the means.