JPH0439698A - speech synthesizer - Google Patents

Abstract

PURPOSE:To obtain various and highly natural synthesized voices by controlling synthesis parameters in accordance with a part of speach in a word including an accent in a sentence to be synthesized. CONSTITUTION:The voice synthesizer has a text analyzing means 2 for analyzing an inputted text, a synthesis parameter forming means 6 for forming a synthesis parameter in accordance with the result of the means 2 and a speech synthesizing means 9 for synthesizing a voice from the synthesis parameter. The means 6 controls the synthesis parameter based upon the speach part information of a word including an accent and amplitude information. Consequently, various and highly natural synthesized voices can be obtained.

Description

[Detailed Description of the Invention] Industrial Field of Application The present invention relates to a speech synthesizer. 4. Prior Art A conventional speech synthesizer is ζ. Furui: Digital Speech Processing p. 146 (Tokai University Press 1985 )
1 is a block diagram showing the configuration of this conventional speech synthesis device and the first embodiment of the present invention. 1 is a character string input terminal where a sentence containing kanji and kana is input, and 2 is a text analysis unit. So, the entered kanji/kana mixed sentences are in Dictionary 3.
are divided into words using the pronunciation, accent type,
4 is a prosodic control unit that determines the accent type of a clause based on the accent type of each word/adjunct word, pause position, intonation of the entire sentence, etc., based on rules. 5 is the phonological control unit, which processes the pronunciation obtained by the text analysis unit, devoices and nasalizes vowels, processes cases where the kana notation and phonetic notation are different (the ``ha'' in ``Washiwa'' is
6 is the synthesis parameter creation section, which obtains the synthesis parameter string necessary for synthesis from the prosodic information and phonetic notation information.The synthesis parameter is the pitch of the sound. Fundamental frequency to be determined Return line to determine the loudness of the voice Vocal tract descriptive parameters (PAR) to determine the state of the vocal tract
COR coefficient, formant frequency, etc.), voiced/unvoiced determination flags that determine the state of the vocal cords, etc., and are stored in parameter table 7.Here, we will refer to both the conventional example and the example.Fujisaki model/k synthesizer is used to give the fundamental frequency. Regarding the Fujisaki model (for example, Fujisaki et al.: Journal of the Acoustical Society of Japan vol. 1.27), using a serial/parallel formant synthesizer.
no, 9 pp, 445-pp456 (1971
) 2, 4 Also, regarding the serial/parallel formant synthesizer, please refer to the example ζL by Allen et al. Alien
From Text to 5peec
h: The MITalk system Chapter 12 (
Cambridge University Pres.
s 1987)) 8 is a parameter interpolation unit, which interpolates the parameters for each phoneme obtained in the parameter table 7 to obtain a synthesis parameter sequence at a fixed time interval. 9 is a speech synthesis unit, 8 Actual speech is synthesized from the synthesis parameter sequence obtained in , and a speech waveform is obtained at the synthesized speech output terminal 10.

Problems to be Solved by the Invention The speech rule synthesis device is used in a variety of fields such as WP text reading and public information broadcasting.
It can be said that the kami is at a practical level.It must be said that the level is still low from the viewpoint of quality such as naturalness.Naturality is low 1. X.

In other words, one of the reasons why the voice becomes very mechanical is that the synthesized voice is very monotonous, and because it is synthesized with a certain prosody, it has a mechanical quality with little variation. . The present invention has been made in view of the problems of the prior art, and aims to improve the synthesis quality by making various changes to the synthesis rules.For example, "That's a cow.""That's a man." k
When we consider the accent nucleus (the syllable position where the accent is placed) in the second clause of the two sentences, ``Ushide'su'' and ``Otoko'da'' are both four-molar clauses with an accent on the third mora. With conventional rule synthesizers, it is difficult to pronounce words with exactly the same intonation.
However, in the former sentence, the word with the accent nucleus is the adjunct word ``desu'', and in the latter sentence, the accent nucleus exists in the independent word ζyo ``man'', even though they do not have the same intonation. The invention as claimed in claim 1 has been made in view of the problems of the prior art, and by controlling the synthesis parameters according to the part of speech of the word in which the accent core of the sentence to be synthesized exists, it is possible to achieve a variety of naturalness. Although the purpose of the rule synthesizer is to provide synthesized speech with a high
10 m5 ec to 30 m5 ec, which is fixed by the synthesizer) is 41. The power to perform the synthesis process assuming that it is stationary (actual speech includes fluctuations in sound. For example, 2
When synthesizing the sound "a" at a height of 50Hz, the pitch was fixed at 250Hz and the sound was synthesized, so the sound would be like a three-sided buzzer and the naturalness would be lost, but we added some fluctuation to improve the naturalness. The invention as claimed in claim 2 has been made in view of the problem of the prior art, and includes a means for imparting a fluctuation, which produces an unnatural synthesized sound depending on the amplitude of the synthesis parameter. By controlling the magnitude of the fluctuation component, the purpose is to provide a highly natural synthesized voice with a rich variety of variations, and it also serves as a standard for creating synthesis parameters regarding the duration and pitch (intonation) of each phoneme. Although speed and reference fundamental frequency 1 are variable, it is cumbersome to set reference values for each sentence or clause, so the invention according to claim 3 does not apply even if the same value is used for one sentence or paragraph. This was done in view of the problems with the prior art, and the aim is to automatically change reference values such as speaking speed when creating synthesis parameters, thereby providing a highly natural synthesized sound with a rich variety. Means for Solving the Problems (1) A text input means for inputting a character string or a symbol string (hereinafter referred to as text), a text analysis means for analyzing the text input from the input means, and a text analysis means for reference by the text analysis means. , a dictionary information storage means for creating a synthesis parameter according to the result of the text analysis means, a synthesis parameter creation means for creating a synthesis parameter according to the result of the text analysis means, and a speech synthesis means for synthesizing speech from the synthesis parameter. A speech synthesis device that controls synthesis parameters according to part-of-speech information of a word having an accent core is configured.(2) Text input means for inputting a character string or symbol string (hereinafter referred to as text); a dictionary information storage means referred to by the text analysis means, a composition parameter creation means for creating a composition parameter according to the result of the text analysis means, and a fluctuation imparting that gives fluctuation to the composition parameter. and a speech synthesis device for synthesizing speech from the synthesis parameters, and the fluctuation imparting means controls the fluctuation component according to the amplitude information obtained by the synthesis parameter creation means. (3) Characters A text input means for inputting a string or a symbol string (hereinafter referred to as text), a text analysis means for analyzing the text input from the input means, a dictionary information storage means for reference by the text analysis means, and the text analysis means. The speech synthesis device comprises a synthesis parameter creation means for creating a synthesis parameter according to the result of the synthesis parameter, and a speech synthesis means for synthesizing speech from the synthesis parameter, and a means for selecting a reference value from the synthesis parameter creation means ζ. Effects With the above configuration, synthesis parameters and fluctuation components are controlled using part-of-speech information and amplitude information of words with accent cores, and by controlling reference values for synthesis parameter creation, it is possible to create synthesized speech that is rich in variation and highly natural. Embodiment 1 FIG. 1 is a block diagram showing the configuration of a speech synthesizer in an embodiment of the invention as claimed in claim 1. Since this is common to the conventional example, explanations regarding each block will be omitted. I will omit the details of each process using actual examples.
This is speech synthesis. '' is input. The text analysis section 2 divides the input sentence into the following parts and obtains information on accent and pronunciation (
Input text) "This is speech synthesis." (word division)
This/is/speech/synthesis/.

(Reading) Kore Tips Onse Couse (Accent) OD
1 0 b (Part of speech) Pronoun Subverb Name
The accent pattern given to ``koso'' and ``desu'' is listed in the NHK accent dictionary/commentary appendix (Japan Broadcast Publishing Co., Ltd., 1985), and is combined with an independent word. It also shows the accent type of each word (i.e., the syllable position where the accent is placed).Next, the prosody processing unit 5 calculates the pause position and the syllable position of each word. The accent of 瓢 determines the intonation of the entire sentence.
Pause) Onse-go°-se-desu'' 4 In other words, the first clause is 4-mora type 3 with an accent nucleus in the attached word, and the second clause is 10-mora type 5 with an accent nucleus in the independent word. Also, the acoustic control unit 5 obtains the actual phonetic notation, the parameter creation unit 6 obtains the actual synthesis parameters according to the prosodic and phonetic information obtained above, and the parameter interpolation unit 9 obtains the actual synthesis parameters of each phoneme thus obtained. Interpolate the parameter values and create a parameter string every 10 msec.Speech synthesis unit 1
0, the fundamental frequency (pitch) in the parameter creation section is synthesized using a formant type synthesizer.
It can also be found using the original model. The original model can be expressed using the following formula: In FO(t) = In Fm1
n ten Σ ApiGpi (t-TOi)
+ΣAaj(Gaj(t-tlj)-Gaj(t-t2
j) )Gpi(t) = 3.0 lines 3,0tex
p(-3,0t) for t >-0G
aj(t) = Min[1-(1-20t)ex
p (-20t), 0.9 for t > -0 api: amplitude of phrase component 0.43 (beginning of sentence), -0,50 (end of sentence) aaj: amplitude of accent component 0.40 (
undulation), 0.20 (flat plate) i: Number of phrase components j: Number of accent components toi: Phrase component rising position tii
: Accent component rising position t2i
: Accent component falling position Pugin :
Standard value (=80) Here, the amplitude of the accent component is 3 even if the undulations are planned as a flat plate.
This is because 0.40 is used because it is an undulating type as well as type 5. In this example, the amplitude value of the accent component is controlled by the part of speech of the word in which the accent nucleus exists. Since there is an accent nucleus in , the amplitude value is set to 0.30. According to this embodiment, by controlling the accent amplitude value according to the part-of-speech information in which the accent kernel exists, it is possible to provide synthesized speech that is rich in variation and highly natural. Figure 2 C is a block diagram showing the configuration of a speech synthesis device in an embodiment of the invention as claimed in claim 2.4 Note that elements common to those in embodiment 1 are given the same numbers. 1 is a character string input terminal. 2 is the text analysis section, and the input sentence containing kanji and kana is inputted in Dictionary 3.
are divided into words using the pronunciation, accent type,
4 is a prosodic control unit that determines the accent type of a clause based on the accent type of each word/adjunct word, pause position, intonation of the entire sentence, etc., based on rules. 5 is the phonological control unit, which processes the pronunciation obtained by the text analysis unit, devoices vowels, makes them nasal, etc., and processes cases where the kana notation and vocalization are different (the ``ha'' in ``watashi wa'' is
6 is the synthesis parameter creation section. The synthesis parameter sequence necessary for synthesis is obtained from the prosodic information and phonetic notation information. The synthesis parameters are stored in the parameter table 7. 8 is a parameter interpolation unit that interpolates the parameters for each phoneme obtained in 7 to obtain a synthetic parameter sequence for each fixed time. 8 is a fluctuation component adding unit that uses random numbers to calculate the fundamental frequency. 9 is a speech synthesis unit which synthesizes actual speech from the synthesis parameter sequence obtained in 8, obtains a speech waveform at the synthesized speech output terminal lO, and then outputs a fluctuation component adding unit 11. To explain this, it is assumed that the parameter interpolation unit 9 obtains the following parameter value as a fundamental frequency every 10 m5ec.

However, here, the sampling frequency of the synthesized sound is 10KHz. Frame number 1 2 3 Fundamental frequency 250 250 248 (Hz) Pitch 40 40 40 (X 1/10m
5ec) Pitch pulse sample No. ζ used in the synthesis unit 10 Sample frame No. 1 as follows
2 3 sun 7° le NOO, 40,
80, 120, 160, 200, 240, 280 However, the actual voice has a fundamental frequency of 250H2.
In this way, the pitch pulses are not regular, but fluctuate, and although this fluctuation improves the naturalness, if a random number is used to give a fluctuation of, for example, ±2 to this pitch pulse train, it will be as follows. Naa Frame No. 1
2 3 sun 7° le NO0,39,
80, 122, 161, 200, 238, 280 The width of this fluctuation is determined in relation to the magnitude of the amplitude parameter. The width of fluctuation is large in parts with small amplitude,
In the part where the amplitude is large, the width of the fluctuation is small (in this example, ζ).The width of the fluctuation is given by the following formula:
When lxamp=40, fluctuation is given by a random number of ±3, i
,, amp = 80 gives a fluctuation of ±1.By controlling the width of fluctuation according to the amplitude information, the monotony of regular synthesized sounds is reduced and a highly natural synthesis with a rich variety is achieved. Embodiment 3 Figure 3 1' can also provide audio! , is a block diagram showing the configuration of a speech synthesis device in an embodiment of the invention as claimed in claim 3.A. Elements common to embodiment 1 are given the same numbers, and kl is a kanji character at the character string input terminal. A mixed sentence is input, and 21o2 is a text analysis section.The input kanji-kana mixed sentence is divided into words using a dictionary 3, and each word is given its reading, accent type, part of speech, etc.4 is a prosodic control section, where each word is divided into words. Determining the accent type of a clause based on the accent type of words and adjuncts.The pause position and intonation of the entire sentence are determined based on rules. 5 is the phonological control unit 6 Based on the pronunciation obtained by the text analysis unit Vowel devoicing and nasalization processing unit Processing when the kana notation and vocalization are different (the ``ha'' in ``watashi wa'' is
12 is a reference value selection means, which determines the reference value to be used for creating the fundamental frequency parameter in the next parameter creation part. 6 is a synthesis parameter creation part, which determines the prosody. The synthesis parameters necessary for synthesis are obtained from the information and phonetic notation information.The synthesis parameters are stored in the parameter table 7, and 48 is a parameter interpolation unit that interpolates the parameters for each phoneme obtained in 7 and calculates them at a fixed time interval. 9 is the speech synthesis section to obtain the synthesis parameter string.
The actual speech is synthesized from the synthesis parameter sequence obtained in step 8, and the speech waveform is obtained at the synthesized speech output terminal 10.The fundamental frequency (pitch) in the parameter creation section is also found using the Fujisaki model.The Fujisaki model is expressed using the following equation. Wasaremo In FO(t) = In Fm1
n ten Σ ApiGpi (t-TOi)
+ΣAaj[Gaj(t-tlj)-Gaj(t-t2
j)) Gpi(t) -3,0 lines 3.0texp(-
3, Ot) for t >-0Gaj(t)
-Min[1-(1-20t)exp(-20t),
0.91for t ＞= Q api: Amplitude of phrase component 0.43 (beginning of sentence), -0,50 (end of sentence) aaj: Amplitude of accent component 0.40 (
undulation), 0.20 (flat plate) i: number of phrase components j: number of accent components toi; phrase component rising position tli
: Accent component rising position t2i
: Accent component falling position Fm1n :
Since 80 (Hz) is used as the reference value in the reference value example 1 (with this, if sentences with the same number of moras and the same accent type continue, the created fundamental frequency parameter will be the same -
Therefore, in the present invention, the random number (ran) is determined by the reference value selection means 6. The reference value Fm1n is set using the following formula: Fm1n = 80 + ran; Like this! By changing the reference value, the monotony of the synthesized sound can be reduced. Here, the fundamental frequency is used as the parameter to be controlled. As described above, according to this embodiment, it is possible to dynamically register sound quality information at the time of synthesis, and it is possible to provide a varied and effective synthesized speech. According to the authors, it is possible to reduce the monotony of synthesized sounds and provide highly effective synthesized sounds that are rich in variety and natural.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the configuration of a speech synthesizer according to a first embodiment of the present invention and a conventional example. FIG. 2 is a block diagram showing the configuration of a speech synthesizer according to a second embodiment of the present invention. FIG. 3 is a block diagram showing the configuration of a speech synthesis device according to a third embodiment of the present invention.
...Dictionary, 4...Prosody processing eye 5...Phonological processing section 6...Synthesis parameter creation eye 7...Parameter table number 8...Parameter interpolation i9...Speech synthesis ff1k 10...・Synthetic voice output ability 11... Fluctuation IK 12... Reference value selection

Claims (3)

[Claims] (1) a text input means for inputting a character string or a symbol string (hereinafter referred to as text); a text analysis means for analyzing the text input from said input means; a dictionary information storage means referred to by said text analysis means; It has a synthesis parameter creation means for creating a synthesis parameter according to the result of the text analysis means, and a speech synthesis means for synthesizing speech from the synthesis parameter, and the parameter creation means has an accent kernel obtained from the text analysis means. A speech synthesis device characterized by controlling synthesis parameters according to part-of-speech information of a word. (2) a text input means for inputting a character string or a symbol string (hereinafter referred to as text); a text analysis means for analyzing the text input from said input means; a dictionary information storage means referred to by said text analysis means; It has a synthesis parameter creation means for creating a synthesis parameter according to the result of the text analysis means, a fluctuation adding means for giving fluctuation to the synthesis parameter, and a speech synthesis means for synthesizing speech from the synthesis parameter, and the fluctuation giving means A speech synthesis device characterized in that a fluctuation component is controlled according to amplitude information obtained by a parameter creation means. (3) a text input means for inputting a character string or a symbol string (hereinafter referred to as text); a text analysis means for analyzing the text input from the input means; and a dictionary information storage means for reference by the text analysis means; The method further includes a synthesis parameter creation means for creating a synthesis parameter according to the result of the text analysis means, and a speech synthesis means for synthesizing speech from the synthesis parameter, and the synthesis parameter creation means includes means for selecting a reference value. Characteristic speech synthesizer.