JP2703253B2 - Speech synthesizer - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Object of the Invention] (Industrial application field) The present invention relates to a speech synthesis device and a speech synthesis method for generating a synthesized speech with high naturalness.

(Prior Art) Conventionally, various speech synthesis techniques have been proposed, and one of them is rule synthesis. This rule synthesis analyzes an arbitrary input character string, obtains its phonological information and prosody information, and synthesizes and outputs the voice indicated by the input character string based on a predetermined rule according to the information. It is. According to this rule synthesis, a synthesized speech of an arbitrary word or phrase can be easily generated. The synthesized speech generated by this rule synthesis is of high quality in terms of intelligibility.
However, "ki, ku, shi, su, chi, tsu, pi, pu, hi, fu"
In order to obtain good synthesized speech even when speech or words containing vowels to be devoiced are synthesized, the voice power of the speech unit parameter file of the vowel to be devoiced must be set to an appropriate value. . However, this setting is difficult, and considerable effort is required to create the speech unit parameter file in consideration of adjustment of the energy balance of speech and intelligibility. In a rule synthesizing apparatus that only outputs a female voice synthesized voice, a trial listening experiment may be repeated to create a voice unit parameter file of unvoiced vowels. However, in a speech synthesizer that includes a plurality of types of speech unit parameter files at the same time and can output a plurality of types (for example, three types of female voice, male voice, and child voice), only the type of synthesized voice is used. It is necessary to create a speech unit parameter file of the unvoiced vowel, which requires a great deal of labor.

(Problems to be Solved by the Invention) As described above, in order to adopt a conventional rule synthesis method and develop a speech synthesizer capable of outputting a plurality of types of synthesized speech,
Speech unit parameter files must be created for each type. However, it is difficult to create these speech unit parameter files due to problems with audio power settings and intelligibility.In particular, vowel syllables to be unvoiced must be repeatedly created through trial listening experiments. There was a problem that required great effort. Therefore, the present invention eliminates the above-described disadvantages, and can synthesize a plurality of types of voices without impairing the smoothness and audibility of voices that are rule-synthesized, and can produce speech units of vowel syllables to be unvoiced. An object of the present invention is to provide a speech synthesis device and a speech synthesis method that can reduce the labor for creating a parameter file.

[Constitution of the Invention] (Means for Solving the Problems) The present invention provides a character string analyzing means for analyzing an input character string to obtain a phoneme symbol string and prosody information, and at least one type of synthesized voice de-voicing. Speech unit parameter storage means for storing speech unit parameters of vowels and for storing speech unit parameters other than a plurality of types of unvoiced vowels according to the type of synthesized speech, and speech obtained by the character string analysis unit A voice parameter sequence generating unit for generating a voice parameter sequence by referring to the voice unit parameter storage unit based on information indicating a type of a synthesized voice from a symbol sequence; and prosody information and synthesis obtained by the character string analyzing unit. Based on the basic pitch and sound source rules determined based on information indicating the type of sound,
It has a configuration in which a prosody parameter sequence generation means for generating a prosody parameter sequence, and speech synthesis means for outputting a synthesized speech based on the speech parameter sequence and the prosody parameter sequence are provided.

Further, the present invention analyzes the input character string to obtain a phonological symbol string and prosodic information, respectively, and when a syllable included in the obtained phonetic symbol string is an unvoiced vowel, a predetermined specific type Extract voice parameters, and if the syllables included in the voice symbol string determined by the character string analysis means is other than a non-voicing vowel, extract a voice unit parameter of a specified type from a plurality of types of synthesized voice, Generate a voice parameter sequence, according to the obtained prosodic information and the basic pitch and sound source rules based on the type of the specified synthesized voice, generate a prosody parameter sequence, based on the voice parameter sequence and the prosody parameter sequence, It consists of processing to output synthesized speech.

(Operation) In the speech synthesis device and the speech synthesis method of the present invention, the speech unit parameter file contains speech corresponding to each type of speech to be synthesized such as a male voice, a female voice, or a child's voice for syllables other than the unvoiced vowel in advance. Register the unit parameters and
As the syllable of the unvoiced vowel, one type of speech unit parameter is registered. The voice parameter sequence generating means generates a voice parameter sequence by referring to the voice unit parameters of one type of unvoiced vowel registered in the voice unit parameter file irrespective of the type of voice to be synthesized. .

Embodiment An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing an embodiment of a speech synthesis apparatus according to the present invention. Reference numeral 1 denotes a character string analysis unit that analyzes the input character string 100 and obtains a phoneme symbol string 200, prosody information 300, and selection information. The phoneme symbol string 200 obtained by the character string analysis unit 1 is input to the speech parameter string generation device 2 together with the selection information, and converted into a speech parameter string 400 by referring to the speech unit parameter file 3. Is done.
On the other hand, the prosody information 300 obtained by the character string
Is supplied to the prosody parameter sequence generation device 4 together with the selection information, and is converted and generated into a prosody parameter sequence 500 here. The voice parameter sequence 400 converted and generated by the voice parameter sequence generation device 2 and the prosody parameter sequence 500 converted and generated by the prosody parameter sequence generation device are input to the voice synthesizer 5, and according to the input parameter sequences 400 and 500, and Based on a predetermined synthesis rule, a synthesized speech 600 corresponding to the input character string is generated and output.

Next, the operation of this embodiment will be described. Female voice, male voice,
In a speech synthesizer capable of outputting a plurality of types of synthesized speech such as a child's voice, a selection signal for selecting which type of synthesized speech is to be generated is required in addition to the input character string 100. In this example, for example, α, β, γ
Is a selection signal input sequence for designating a female voice, a male voice, and a child voice, respectively. FIG. 2 is a block diagram showing a detailed example of the character string analysis unit 1 shown in FIG. The discriminator 11 discriminates the type of the selection signal and sends selection information indicating which type of synthesized sound output, male, female, or child, is selected to the voice parameter sequence generator 2 and the prosody parameter sequence generator 4. . The speech parameter sequence generator 2 determines which of the male, female, and child voice segment parameter files of the speech segment parameter file 3 to select from the selection information. decide male voice, female voice, in any of the child's voice or also the sound source rule to set the basic pitch f ₀ from. Further, the discriminator 11 calculates the above α,
Input signal (input character string) other than selection input signal of β and γ 10
Regarding 0 as Japanese text for speech synthesis,
These input signals are passed to the text analysis unit 12. The text analysis unit 12 analyzes the text of the input signal using the language dictionary 13 to obtain a phoneme symbol string 200 and prosody information 300. The phoneme symbol sequence 200 and the prosody information 300 obtained in this manner are used as the speech parameter sequence generation device 2 and the prosody parameter sequence generation device 4.
Given to. The speech parameter sequence generation device 2 extracts a speech segment parameter corresponding to the inputted phoneme symbol string from the speech segment parameter file 3 based on the selection information previously inputted from the discriminator 11, and performs interpolation combination. To generate a voice parameter sequence 400. The prosody parameter sequence generation device 4 generates a prosody parameter sequence 500 based on the sound source generation rules from the basic pitch f ₀ determined by the selection information previously input from the discriminator 11 and the sound source rules and the input prosody information. I do.

Here, it is assumed that the selection signal and the character string input to the character string analysis unit 1 are, for example, “β smooth sound can be obtained. Select the α function. Do you want to display the γ screen?” . In this case, the voice parameter sequence generator 2 first selects a male parameter file in the voice unit parameter file 3 by the selection signal β, and the prosodic parameter sequence generator 4 selects the male voice basic pitch f ₀ and the sound source generation rule are selected. On the other hand, in the text analysis unit 12,
"Smooth speech is obtained" is analyzed using the language dictionary 13, and a phoneme symbol string 200 and a prosody information string 300 corresponding to the character string are obtained. Therefore, the voice parameter sequence generation device 2 and the prosody parameter sequence generation device 4 generate the voice parameter sequence 400 and the prosody parameter sequence 500 corresponding to the character string.
Are generated, and these parameter strings are input to the speech synthesizer 5. For this reason, the speech synthesizer 5 outputs “smooth speech is obtained” by the ruled synthetic voice of the male voice. Similarly, the voice synthesizer 5 outputs "Please select a function" with the ruled synthetic voice of the female voice, and then outputs "Do you want to display the image?"

FIG. 3 is a schematic diagram showing a configuration example of the speech unit parameter file 3. As speech unit parameters 3, cepstrum parameters, LPC parameters, PARCOR
There are parameters, LSP parameters, formant parameters, etc., and those shown in FIG. 3 (A) are a female voice segment parameter file 31, a male voice segment parameter file 32, a child voice segment parameter file 33 and a devoiced vowel element. It is composed of one parameter 50, and is used for a voice rule synthesizer capable of outputting female voice, male voice, and child voice. In this voice unit parameter file 3, syllables other than unvoiced vowels (Japanese hundred syllables and foreign syllables) are registered by analyzing the syllables of female, male, and child voices from natural speech and parameterizing them. Have been. Vowels of unvoicing vowels As for, a natural voice of a female announcer created in consideration of energy balance and intelligibility of voice is registered in the voice unit parameter file 3 as a voice unit parameter 50 of a common unvoiced voice. Note that the speech segment parameters of the silenced speech may be created using a male announcer or a child's voice, but in this example, only one type is registered.

Here, the speech synthesizer shown in FIG. Is output as a male voice synthesis voice, the voice parameter sequence generation device 2 outputs the male voice segment parameter of the voice segment parameter file 3 shown in FIG.
Take out 32, Extracts the unvoiced vowelization parameters 50, which are common data of unvoiced vowels, and performs intercept coupling to generate a corresponding voice parameter sequence 400. The same applies to the case of generating a synthetic voice of a female voice and a synthetic voice of a child's voice. FIG. 3 (B) shows an example in which voice unit parameters of unvoiced vowels are provided in the voice unit parameter file 3 for male, female, and child voices in the voice unit file, respectively. Since the same parameter is registered in one parameter, the parameter is substantially the same as that shown in FIG. 3 (A). However, when the unvoiced vowel segment parameter file 50 obtained by voice analysis of the unvoiced vowel uttered by the female announcer is used in common as shown in FIG. Need not be stored in the memory, and the memory can be saved accordingly. By the way, among the unvoiced vowels, Are difficult to create in terms of clarity and energy balance. In this example, Is created as common data in the speech unit parameters 50 of the unvoiced vowels, and the remaining unvoiced vowels Has registered each natural voice in each parameter file.

According to the present embodiment, the unvoiced vowel segment parameter 50 registered in the speech segment parameter 3 is only one type created from, for example, a natural voice of a woman. There is no need to create the unvoiced vowel speech unit parameters, and the labor for creating the unvoiced vowel speech unit parameters can be greatly reduced. In addition, since the one type of unvoiced vowel speech unit parameter 50 is used as common data when creating a synthetic voice for a female voice, as well as when creating a synthetic voice for a male voice and a child voice, any of the synthetic voices has smoothness, It is possible to regularly synthesize high-quality items having high naturalness without impairing the audibility and the like.

[Effects of the Invention] As described above, according to the speech synthesizing apparatus and the speech synthesizing method of the present invention, a plurality of types of speech can be synthesized without impairing the smoothness and audibility of the speech that is regularly synthesized. This has the effect of reducing the labor required to create a speech unit parameter file of the vowel syllable to be unvoiced.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an embodiment of a speech synthesizer according to the present invention, FIG. 2 is a block diagram showing a detailed example of a character string analyzing section shown in FIG. 1, and FIG. 3 is a schematic diagram showing a configuration example of a speech unit parameter file shown in FIG. DESCRIPTION OF SYMBOLS 1 ... Character analysis part 2 ... Speech parameter string generation device 3 ... Speech unit parameter file 4 ... Prosody parameter string generation device 5 ... Speech synthesizer, 11 ... Discriminator 12 ... Text analysis part, 13 ...... Language dictionary

Claims (2)

(57) [Claims] 1. A character string analyzing means for analyzing an input character string to obtain a phoneme symbol string and prosody information, and storing speech unit parameters of at least one type of unvoiced vowel of a type of synthesized speech, and A speech unit parameter storage unit for storing speech unit parameters other than a plurality of types of unvoiced vowels according to the type of speech, and a speech symbol string obtained by the character string analysis unit, based on information indicating the type of synthesized speech. A voice parameter sequence generating unit that generates a voice parameter sequence by referring to the voice unit parameter storing unit; a basic pitch determined based on prosody information obtained by the character string analyzing unit and information indicating a type of synthesized voice. And a prosody parameter sequence generating means for generating a prosody parameter sequence based on the sound source rules, and a synthesized speech based on the speech parameter sequence and the prosody parameter sequence. A voice synthesizing means for outputting voice. 2. An input character string is analyzed to obtain a phoneme symbol string and prosody information. If a syllable included in the obtained phonetic symbol string is an unvoiced vowel, a predetermined specific type of speech If the syllables included in the phonetic symbol sequence obtained by the character string analysis means are other than unvoicing vowels, the speech unit parameters of the specified type are extracted from a plurality of synthesized speech types, and Generating a parameter sequence, generating a prosody parameter sequence in accordance with the determined prosody information and a basic pitch and sound source rule based on the specified type of synthesized speech, and synthesizing based on the speech parameter sequence and the prosody parameter sequence. A speech synthesis method characterized by outputting speech.