JP3109778B2 - Voice rule synthesizer - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech rule synthesizing apparatus for generating synthesized speech.

[0002]

2. Description of the Related Art In a conventional speech synthesizer for synthesizing speech according to rules, phonemes, syllables, V
CV (vowel / consonant / vowel) concatenation, CVC (consonant / vowel /
Consonants) Set units that take into account the connection with phonemes, such as concatenation and words, and set units that take into account articulation, store speech synthesis parameter values created by analyzing natural speech, and synthesize speech in units corresponding to the input character strings. Speech was synthesized by editing, combining, and transforming parameters (hereinafter, referred to as synthesis units).

[0003]

However, in the above-mentioned conventional speech synthesizer, the sound collected and uttered for each unit with the same phoneme or syllable is considerably different from the sound appearing in the sentence. There was a problem that it lacked.

For example, when a sentence is synthesized using a result of analyzing a natural voice that utters a single syllable or the like, the synthesized sound has an impression that each sound is clearly pronounced. The tendency increases as the speed of the synthesized sound increases.

[0005] In addition, a large amount of natural speech previously uttered in units longer than the synthesis unit, such as sentences and words, is provided.
When the optimal segment is selected and used as a synthetic segment, the articulatory connection is already expressed and the naturalness is improved, but a rule for selecting the optimal segment has not yet been found.

In particular, in order to select a segment in consideration of the spectral distortion at the connection of the synthesized segments in order to reduce the distortion due to the connection of the synthesized segments, it is necessary to calculate the inter-spectral distance between the segments. There is a problem that a large amount of calculation is required due to the large number of combinations of segments.

An object of the present invention is to select a segment having a relatively small amount of computation from a large amount of speech data and having a small spectral distortion at a segment connection portion, thereby outputting a synthesized speech having high clarity and naturalness. It is an object of the present invention to provide a voice rule synthesizing device capable of performing the above.

[0008]

SUMMARY OF THE INVENTION It is an object of the present invention to analyze natural speech and store speech synthesis parameters labeled for each phoneme, and to divide the speech into appropriate synthesis units for assembling output speech. Divided into prosodic information and synthesis units
Setting means for setting the front and rear of the phoneme sequence that is based on information from the phoneme sequence and prosodic information before and after that is set as in setting unit, a spectrum of connection phoneme center of connection part <br / > Calculating means for calculating the target spectrum, and calculating means
Target spectrum and spectrum calculated by
The composite element having the minimum distance is stored in the storage means.
And Ruoto voice synthesis parameter or al selection selecting means is achieved by in which the speech synthesis by rule system and connection means for connecting the case Narumotohen selected by the selection means.

[0009]

In the speech rule synthesizing apparatus of the present invention, the storage means comprises:
Speech synthesis labeled for each phoneme by analyzing natural speech
Store the parameters, setting means divides the appropriate synthesis unit to assemble the output speech, prosody information and the synthesis unit
And the phonological sequence before and after being divided into
From the before and after phoneme sequences and prosodic information set by the setting means
Based on the information, the spectrum of the connecting phonemes center of connection part
Calculating a target spectrum is, selecting means, target spectrum and space calculated by the calculating means
The composite element with the minimum vector distance is stored in the storage means.
From the speech synthesis parameters that have been established , and the connection means:
The synthesis unit selected by the selection unit is connected.

[0010]

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram showing a speech rule synthesizing apparatus according to an embodiment of the present invention;

FIG. 1 is a block diagram showing a configuration of an embodiment of a speech rule synthesizing apparatus according to the present invention.

The speech rule synthesizing apparatus shown in FIG. 1 is connected to a text input terminal 11, performs morphological analysis, kanji-kana conversion, accent processing, etc. based on a text to be converted input from the text input terminal, and outputs the result. A text analysis unit 12 connected to the text analysis unit 12 and generating prosody information for generating and outputting a pitch pattern, a time length pattern for each phoneme, and an amplitude pattern based on the analysis information output from the text analysis unit 12 A synthesizing unit setting unit 14 which is connected to the text analyzing unit 12 and is a setting unit which divides the synthesized speech into synthesizing units based on the analysis information output from the text analyzing unit 12 and outputs the synthesized speech. It is connected to the information generating unit 13 and the synthesizing unit setting unit 14. A target spectrum calculating unit 15 which is a calculating means for calculating and outputting an appropriate target spectrum based on the information from the unit 13; connected to the target spectrum calculating unit 15; a sound necessary for synthesis based on a large amount of voice data; analysis parameters, the speech parameter file 16 is a storage unit for outputting created, the target spectrum calculating unit 15 and the speech parameter file 1
6, a synthesis unit selection unit 17 and a synthesis unit selection unit which are selection means for selecting and outputting, from the speech parameter file 16, a spectrum whose connection section of the synthesis unit is closest to the target spectrum. a connecting means for coupling and outputting the connected and selected by the segment between the 17 synthesis fragment connection unit 1 8, synthesis fragment is connected to the prosody information generating unit 13 and the synthesis fragment connection 18 This is achieved by a synthesized speech generation unit 19 that generates a synthesized speech based on the synthesized speech sequence obtained by the connection unit 18 and the prosody information obtained by the prosody information generation unit 13 and outputs the synthesized speech to the output terminal 20.

Next, the operation of the speech rule synthesizing apparatus shown in FIG. 1 will be described.

When a text to be converted into speech is input from the text input terminal 11, the text analysis unit 12 performs syntax analysis such as dependency, morphological analysis such as part-of-speech analysis, kanji-kana conversion, and accent processing. Necessary analysis information is sent to the synthesis unit setting unit 14 and the prosody information generation unit 13. The analysis information includes, for the synthesis unit setting unit 14, a symbol string indicating the distinction of phonemes, and for the prosody information generation unit 13, the number of mora in an exhalation paragraph, accent form, utterance speed, and the like.

The prosody information generation unit 13 generates a pitch pattern, a time length pattern for each phoneme, and an amplitude pattern based on these information according to rules.

The synthesis unit setting unit 14 divides the input phoneme symbol string into synthesis units suitable for assembling output speech such as syllables and VCV phoneme sequences, and divides the divided phoneme sequences into target spectrum calculation units 15. Output to

The target spectrum calculation unit 15 includes a phonological sequence before and after divided into synthesis units, and a prosody information generation unit 1.
The optimum target spectrum is calculated based on the information from Step 3.

The voice parameter file 16 is created in advance by offline processing based on a large amount of voice data. For example, phonetic labeling is performed on audio data for several hours, such as words and sentences by one announcer, by inspection using digital sonargrams, and acoustic parameters required for synthesis are analyzed in advance.

The synthesis unit selection unit 17 selects, from the speech parameter file 16, the one whose connection unit spectrum is closest to the target spectrum.

The synthesized segment connecting unit 18 combines the selected segments and sends the result to the synthesized waveform generating unit 19.

The synthesized speech generating unit 19 includes the synthesized speech unit connecting unit 1.
A synthesized speech is generated based on the synthesized speech sequence obtained in step 8 and the prosody information obtained in the prosody information generation unit 13, and the generated voice is output to the output terminal 10.

In the above-described configuration, the text analysis unit 12 is provided. When the text analysis is performed in advance and the analysis information is input to the apparatus, the text analysis unit 1 is used.
2 can be omitted.

Similarly, when a prosody pattern is generated in advance and input to the apparatus, the prosody information generation unit 13 can be omitted.

There are no particular restrictions on the synthesizer for generating the acoustic parameters and the synthesized speech used here, and the present invention can be applied to all.

Next, referring to the flowchart of FIG.
The operation of the target spectrum calculator 15 will be described in detail.

FIG. 2 shows an example of calculating the target of / N / when synthesizing / oNsei /.

First, the preceding and succeeding synthesis units and prosody information are input (step S1), a phoneme sequence is set centering on the phoneme of the connection part (step S2), and the speech parameters including the phoneme sequence are read from the speech parameter file 16. Search (step S3).

If no candidate is found, the search is performed while sequentially deleting the search phoneme sequence from both sides. For example, if there is no voice parameter including / oNse /,
oNs / → / Ns / → / N /

Next, the pitch condition of the connection portion is set from the prosody information (step S4), and candidates are narrowed down (step S5). This pitch condition is, for example, ± 5 of the pitch.
%. If there is no such condition, the pitch condition is expanded to ± 10%, 15%.

Next, a candidate closest to the continuation length of the connected phoneme is selected from the candidates (step S6), and the spectrum of the center of the connected phoneme is calculated from the selected speech parameters (step S7). The spectrum is set (step S8).

With the above processing, the target spectrum of the connection part is calculated.

Next, the speech rule synthesizing process by the speech rule synthesizing apparatus shown in FIG. 1 will be specifically described.

For example, when the word “voice” is input to the text input terminal 11, the text analysis unit 12 outputs / oN
A phonological sequence “sei /” and prosodic information are generated. Assuming that the synthesis unit is VCV, the synthesis unit setting unit 14 sets / So /, / oN /, / Nse /, / ei /, / iS /
Is divided into five synthesis units. However, / S / represents silence. Next, we select segments for each synthesis unit,
An example in the case of / Nse / is shown below.

First, the target spectrum calculator 15 calculates the target spectrum at the connection between / oN / and / Nse /. In this case, the voice parameter of the phoneme sequence of / oNse / is searched from the voice parameter file 16, and
A spectrum that is the temporal center of / N / of the speech parameter selected by narrowing down from the prosody information is set as a target spectrum SP1.

Similarly, the target spectrum SP2 at the connection between / Nse / and / ei / is calculated.

Next, the synthesis unit selection unit 17 outputs / Nse /
The speech parameter having the phoneme sequence of is searched from the speech parameter file 16. Next, the target spectrum S
The minimum value of the spectral distance of the / N / and / e / portions is calculated for each of the candidates searched for P1 and SP2, and the candidate having the smallest sum of the minimum values is selected as the synthesis unit. After selecting the segments for each synthesis unit in this way, the synthesis segments are connected at the position where the distance from the target spectrum is the minimum, and a synthesized waveform is generated.

As described above, by setting the optimum target spectrum at the connection portion and connecting the synthesis segments having the spectrums closest to the optimum target spectrum, a synthesized speech with little connection distortion can be obtained.

In the conventional method of performing synthesis with small connection distortion without setting a target spectrum, a large amount of calculation is required due to the large number of combinations between connected synthesis segments. Can greatly reduce the amount of calculation.

Further, as a method of reducing the amount of calculation and the memory, a target spectrum is calculated in advance for each phoneme sequence and prosody information, and a synthesis unit optimal for the target spectrum is registered in a table.

For example, by combining VCV units,
If / is also considered as a vowel, the connections are / a, i, u, e,
o and N /.

Considering the minimum hardware configuration, the stationary part of a single vowel uttered at a normal pitch is analyzed in advance,
Let each spectrum be a target spectrum.

Next, an optimum synthesis unit for the target spectrum is selected by the same algorithm as that of the synthesis unit selection unit 17 and registered in a table. At the time of synthesis, the synthesis unit is selected by referring to the table. In this case, it is possible to construct a table in which one type of synthesized element corresponds to each VCV.

In this method, although there is a possibility that the quality of synthesized speech may be lower than in the method of performing search processing at the time of synthesis, it is only necessary to store only the synthesis segments described in the table in the voice parameter file. Since search processing is not performed at the time of synthesis, the amount of calculation and memory can be significantly reduced.

Further, if a slightly larger hardware configuration is possible, it is possible to target a stationary part of a single vowel uttered at a plurality of pitches, or from a phoneme sequence (for example, unvoiced or nasalized) which is strongly affected by articulation. The quality can be further improved by creating a target and creating a synthesis unit table.

[0045]

The speech rule synthesizing apparatus of the present invention analyzes natural speech and stores speech synthesis parameters labeled for each phoneme, and divides the output speech into appropriate synthesis units to assemble the output speech. Divided into prosodic information and synthesis units
Setting means for setting the front and rear of the phoneme sequence that, based on information from the phoneme sequence and prosodic information before and after that is set as in setting unit, a spectrum of connection phoneme center of connection part data <br /> and calculating means for calculating a target over the spectrum, the calculation means
Therefore, the calculated target spectrum and spectral distance
The smallest synthesis fragments of the release, that is stored in the storage means
Selection means for either et selection speech synthesis parameters, since a connecting means for connecting the case Narumotohen selected by the selecting means, advance to accumulate large amount of speech parameters, for the synthesis of speech Then, an output speech is synthesized by extracting and connecting an appropriate synthesis unit. As a result, a voice with high clarity and good naturalness can be obtained with a small amount of calculation.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of an embodiment of a speech rule synthesis device according to the present invention.

FIG. 2 is a flowchart for explaining a target spectrum calculation process performed by the speech rule synthesis device of FIG. 1;

[Explanation of symbols]

 Reference Signs List 11 text input terminal 12 text analysis unit 13 prosody information generation unit 14 synthesis unit setting unit 15 target spectrum calculation unit 16 voice parameter file 17 synthesis unit selection unit 18 synthesis unit connection unit 19 synthesized voice generation unit 20 voice output terminal

Continuation of front page (56) References JP-A-3-119394 (JP, A) JP-A 1-284898 (JP, A) JP-A 4-281495 (JP, A) JP-A 5-94199 (JP) , A) JP-A-5-53595 (JP, A) IEICE Technical Report [Voice], Vol. 89, No. 318, SP89-65, Tetsuya Nomura et al. "Phoneme segment connection method considering continuity of low-order cepstrum coefficients", p. 9-16 (published November 27, 1989) Proceedings of the Acoustical Society of Japan, Fall Meeting, 1989, I-P, 3-P-4, Tetsuya Nomura, et al. Segment connection method ”, pp. 283-284 (October 4, 1989) (58) Fields investigated (Int. Cl. ⁷ , DB name) G10L 11/00-13/08 G10L 19/00-21 / 06 JICST file (JOIS)

Claims (2)

(57) [Claims] 1. A storage means for analyzing natural speech and storing speech synthesis parameters labeled for each phoneme, and dividing the speech into appropriate synthesis units for assembling an output speech.
Setting means for setting the phoneme series of front and rear divided into information and the synthesis unit, phonological system before and after set by the setting means
Based on the information from the column and the prosody information, connecting section of the connection tone
Synthesis element comprising a calculation means for calculating a target spectrum is the spectrum of the rhyme center, a minimum of te <br/> target spectrum and spectral distance calculated by the calculating means
Pieces, and selection means for Ruoto voice synthesis parameters or al selection is stored in the storage means, selected by said selection means
That a connecting means for connecting the coupling Narumotohen speech rule synthesizing apparatus according to claim. 2. The method according to claim 1, wherein the calculating means calculates the prosody information and the synthesis unit.
Input the phonological sequence before and after the division and before and after connecting
From speech parameter candidates including the phoneme sequence of the synthesis unit,
Select the speech parameter closest to the duration of the phoneme of the connection
The pitch condition of the connection part is set from the
Narrow down meter candidates and check if
Calculating a target spectrum from the
The speech rule synthesis device according to claim 1.