JPH02205896A - Voice synthesizing device - Google Patents

Abstract

PURPOSE:To generate a synthesized voice with a tone quality fit for the purpose of use by providing a voice element piece file for proper noun and a voice element piece file for sentence and selectively switching and using them in accordance with contents of voice synthesis. CONSTITUTION:Phonetic information obtained in a character information analyzing part 1 is inputted to a voice parameter generating part 2, and a voice element piece file 3 for proper noun is referred to generate a voice parameter corresponding to inputted phonetic information at the time of synthesizing a voice of only a proper noun, and a voice element piece file 4 for sentence is referred to generate a voice parameter corresponding to inputted phonetic information at the time of synthesizing a voice of a sentence. Metrical information obtained in the character information analyzing part 1 is inputted to a metrical parameter generating part 5, and a metrical parameter corresponding to inputted metrical information is generated. A voice synthesizing part 6 generates and outputs a synthesized voice based on prescribed synthesizing rules in accordance with inputted voice parameters and metrical parameters. Thus, the synthesized voice is generated with a tone quality fit for the purpose of use.

Description

DETAILED DESCRIPTION OF THE INVENTION [Object of the Invention] (Field of Industrial Application) The present invention relates to a speech synthesis device that generates speech from arbitrary character information input by a rule synthesis method.

(Prior Art) Various speech synthesis techniques have been proposed in the past, and one of the techniques is a speech rule synthesis method.

The rule synthesis method analyzes arbitrary text input manually to obtain its phonological and prosodic information, and outputs synthesized speech corresponding to the input text information based on predetermined rules. It is. According to such a rule synthesis method, synthesized speech of any word or phrase can be easily generated.

By the way, when outputting only proper nouns such as first and last names or company names, it is desirable that the rule-synthesized speech has high clarity so that the content can be clearly understood. However, when outputting a weather forecast or a message, for example, it is desired that the synthesized sound be smooth and not stuttered, rather than high intelligibility.

In this way, for example, even though the required sound quality is different when outputting only proper nouns and when outputting sentences, the synthesized sound that can be output is either one of them, and it is not possible to output both. It was not possible to output selectively.

(Problems to be Solved by the Invention) As described above, the present invention is capable of outputting even though the required sound quality is different depending on, for example, when outputting only proper nouns and when outputting sentences as voice. This was done to solve the problem that synthesized sounds can be either one of these, and it is not possible to selectively output both.For example, the sound quality can be adjusted depending on whether only proper nouns or sentences are output. The purpose of the present invention is to provide a speech synthesizer that can be changed.

[Structure of the Invention] (Means for Solving the Problems) The present invention obtains phonological information and prosody information by analyzing input character information, and generates a speech segment file in which speech segment parameters are stored. is configured to generate speech parameters corresponding to the obtained phonetic information and prosodic parameters corresponding to the obtained prosodic information, respectively, and to synthesize speech according to the generated speech parameters and prosodic parameters. The speech synthesis device is equipped with a plurality of the speech segment files, and is configured to selectively switch and use speech segment files to be referred to for generating speech parameters according to the content of speech to be synthesized and output. It is characterized by what it did.

(Function) For example, by having both speech segment files for proper nouns and speech segment files for sentences, and selectively switching and using them depending on the content to be synthesized, the synthesized speech can only be produced by, for example, proper nouns. The speech segment file to be referred to for generating speech parameters can be switched depending on whether the speech segment file is a text or a text. Therefore, for example, in the case of only proper nouns, a synthesized voice with high clarity can be obtained, and in the case of sentences, a synthesized voice with a smooth and highly natural sound can be obtained, and a synthesized voice with a sound quality suitable for each purpose of use can be generated. can.

(Example) Hereinafter, an example of the present invention will be described with reference to the drawings.

In FIG. 1, reference numeral 1 denotes a character information analysis unit that analyzes input character information to obtain its phonological information and prosodic information. The phonetic information obtained by the character information analysis section 1 is input to the speech parameter generation section 2. When performing speech synthesis of only proper nouns, the speech parameter generation unit 2 generates speech parameters corresponding to the input phonetic information by referring to the speech segment file 3 for proper nouns, and converts the sentence into speech. In the case of synthesis, speech parameters corresponding to the input phonetic information are generated by referring to the speech segment file 4 for sentences.

On the other hand, the prosodic level determined by the character information analysis section 1 is input to the prosodic parameter generation section 5.

The prosodic parameter generation unit 5 generates prosodic parameters corresponding to the input prosodic information.

Audio parameters generated by the audio parameter generation unit 2,
The prosodic parameters generated by the prosodic parameter generator and the prosodic parameter generator are respectively input to the speech synthesizer 6. The speech synthesis unit 6 generates and outputs synthesized speech based on predetermined synthesis rules in accordance with input speech parameters and prosody parameters.

Note that whether to refer to the speech segment file 3 for proper nouns or the speech segment file 4 for sentences is selected by a software switch means, for example, at the time of initialization of this device. Switching is performed depending on the method and symbol input (for example, when the symbol $ is input, the phonetic segment file 3 for proper nouns is selected, and when the symbol # is input, the phonetic segment file 4 for sentences is selected, etc.) There are many possible methods.

Here, to explain the case where only proper nouns are synthesized into speech, for example, an address in a database,
A device that reads names and phone numbers one after another can be considered. In this case, it is important to output a clear synthesized sound.

On the other hand, when synthesizing sentences such as weather forecasts and messages, it is important to generate synthesized sounds that are not only clear but also smooth.

Therefore, the required sound quality is different when synthesizing only proper nouns and when synthesizing sentences, but by providing speech segment files for both in one device, it is possible to perform speech synthesis for both purposes. can be used.

Now, an example of a method for creating speech segment files at the proper noun level and sentence level will be described. At the proper noun level, the clarity of syllables such as ``ja'' and ``chiya'' is also important, so a phoneme file of the syllable is created from isolated utterances of the syllable.

It also emphasizes the high frequencies of the audio and performs parameter analysis.

On the other hand, since it is important to generate smooth synthesized sounds at the sentence level, syllable speech segment files are created using syllables extracted from the natural speech of two-, three-, and four-syllable words. . For example, when extracting the elemental segment parameter of a syllable / ka /, utter the natural sound of a word that includes this syllable, such as "nakama", and extract the corresponding part / ka / from the phoneme /nakama/. , extract the parameters, and create a speech segment file. The high frequencies of speech are also analyzed without emphasizing them as much as those at the proper noun level. In this way, for sentences, speech synthesis is performed using speech segment files obtained by analyzing continuously uttered natural speech, resulting in smooth synthesized speech.

Note that the speech unit parameters include cepstrum parameters, LPG parameters, PARCOR parameters,
Possible options include formant parameters.

Further, in the above embodiment, for proper nouns, a speech segment file that can generate a synthesized sound with high clarity is referred to, and for sentences, a speech segment file that can generate a smooth and highly natural synthesized speech is referred to. However, depending on the user's needs, by inputting the switching symbol of the phoneme file, the user can select the part for which he/she wants to produce a clear synthesized sound, and then create the phoneme for the proper noun. You can refer to a single file and make only that part a clearer synthesized sound.

[Effects of the Invention] As detailed above, according to the speech synthesis device of the present invention, it is provided with both speech segment files for proper nouns and speech segment files for sentences, and can be selectively synthesized depending on the content to be speech synthesized. By switching to , depending on whether the synthesized speech is only proper nouns or sentences,
It is possible to switch the speech segment file referred to for generating speech parameters. Therefore, for example, in the case of only proper nouns, a synthesized voice with high clarity can be obtained, and in the case of sentences, a synthesized voice with a smooth and highly natural sound can be obtained, and a synthesized voice with a sound quality suitable for each purpose of use can be generated. can.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing one embodiment of the present invention. 1... Character information analysis section, 2... Speech parameter generation section, 3... Speech segment file for proper noun, 4... Speech segment file for sentence, 5... Prosodic parameter generation section, 6...Speech synthesis section. Applicant's agent Patent attorney Takehiko Suzue

Claims (1)

[Claims] Phonological information and prosody information are obtained by analyzing input character information, and the obtained phonological information is obtained by referring to a speech segment file in which speech segment parameters are stored. A speech synthesis device configured to generate corresponding speech parameters and prosodic parameters corresponding to the obtained prosodic information, and to synthesize speech according to the generated speech parameters and prosodic parameters, the speech synthesis device comprising: 1. A speech synthesis device comprising a plurality of segment files and selectively switching and using speech segment files to be referred to for generating speech parameters according to the content of speech to be synthesized and output.