JPH1152987A - Speech synthesis device with speaker adaptive function - Google Patents

Abstract

PROBLEM TO BE SOLVED: To make it possible to imitate a speech of a specific person by having a parameter adding part for determining a parameter expressing a feature of user's speech, and synthesizing speech by using the determined parameter. SOLUTION: A character string input device 1 inputs a character string and forwards it to a memory 10. Next, a synthetic parameter calculation part 2 calculates a parameter expressing a feature of the input character string. A parameter adding part 3 reads out a parameter difference value accumulated in the memory 10 or in a parameter difference valve accumulation device 9 and a synthetic parameter calculated by a synthetic parameter calculation means 2 from the memory 10, and sums up the difference value and the synthetic parameter and forwards the added parameter to the memory 10. And, a speech synthesis part 4 operates speech synthesis from the parameter outputted from a synthetic parameter calculation part 2 and forwards the speech waveform to the memory 10. Moreover, a speech input device 6 inputs user's speech and forwards it to the memory 10.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech synthesizer capable of changing an output mode of a speech synthesized from a character string into an output mode of a speaker's voice input separately.

[0002]

2. Description of the Related Art A conventional speech synthesizer uses a method of synthesizing a speech from a character string using synthesis parameters calculated from a sound source model representing a prosodic feature and a vocal tract model representing a phonemic feature. Was common. The synthesis parameters representing the characteristics of the speech include a fundamental frequency, a phoneme duration,
There are power, spectral parameters, etc.

In the case of using the speech synthesizer, in order to simulate a voice uttered by a human, a sound source model for generating a parameter representing a characteristic of a prosody and a vocal tract model for simulating a transfer characteristic of a vocal tract are precisely formed. Must be configured. The sound source model and the vocal tract model are information that is responsible for the speaker's personality, and by bringing both of them closer to the real voice as much as possible, the sound source model and the vocal tract model can be closer to the voice of the speaker. Among the sound source models, models that specify the fundamental frequency include, for example, a document “Fujisaki, Sudo:“ Basic frequency pattern of Japanese word accent and its generation mechanism ”, Journal of the Acoustical Society of Japan 27, pp. 445.
-453, 1971 ". On the other hand, the spectral information of speech or the energy distribution of a specific frequency on a frequency axis called a formant is a parameter of a vocal tract model that determines phonological properties.

[0004]

In order to synthesize speech of a specific individual using the speech synthesizer, it is necessary to construct a model using parameters analyzed and extracted from the speech of the individual. .

However, since the sound source model and the vocal tract model in the conventional speech synthesizer are designed based on the reference speech of the system, it is not possible to realize a synthesized speech of an unspecified speaker only from a character string. It was impossible.

Furthermore, since a sound source model such as a fundamental frequency of a voice approximates a fundamental frequency of a real voice by a generation rule, there is a problem that a fine pitch fluctuation which is not an object of modeling may not be expressed. .

Therefore, in the present invention, in order to solve the above-mentioned problem, the reference synthesis parameter of the system calculated for the character string and the analysis parameter obtained by analyzing the voice uttered by the individual are analyzed. An object of the present invention is to provide a speech synthesizer capable of simulating a specific individual's voice by obtaining a difference value and transmitting the difference value together with a character string.

[0008]

In order to achieve the above object, a voice synthesizing apparatus according to the present invention comprises a means for inputting a user's voice, a parameter analyzer for analyzing the input voice, and a synthesis parameter for a character string. A parameter calculation unit for calculating, a parameter comparison unit for comparing a parameter obtained from the input voice with a parameter calculated from a unique model of the system, and a parameter difference value storage for storing a difference value calculated by the parameter comparison unit Means for dividing a user's voice into a reference synthesis parameter of the system and a difference value. Further, the speech synthesis apparatus of the present invention has a parameter addition unit that adds a parameter difference value and a reference synthesis parameter calculated from a model of the system to a character string to obtain a parameter representing a feature of a user's voice. Then, by performing voice synthesis using the parameters obtained by the parameter adding unit, a voice having characteristics of the user's voice can be synthesized.

Further, the data composed of the difference value and the character string calculated by the speech synthesizer of the present invention is transferred between the speech synthesizers existing in different time and space due to the electric storage medium and the communication means. Is provided. With this feature, it is possible to easily synthesize a speech adapted to personal information by the difference value.

[0010]

Embodiments of the present invention will be described below with reference to the drawings.

First, an embodiment of the present invention will be described with reference to FIGS. FIG. 1 is a block diagram showing the configuration of claim 1 of the present invention. In FIG. 1, reference numeral 1 denotes a character string input device for inputting a character string and transferring the character string to a memory 10;
Is a synthesis parameter calculation unit that calculates a parameter representing a feature of a voice from the character string and transfers the parameter to the memory 10, 3 is a parameter difference value stored in the memory 10 or the parameter difference value storage device 9, The calculated synthesis parameters are read from the memory 10,
The parameter addition unit 4 adds the difference value and the synthesis parameter and transfers the added parameter to the memory 10. The parameter addition unit 4 synthesizes speech from the parameter output from the synthesis parameter calculation unit 2 and transfers the speech waveform to the memory 10. A voice synthesizing unit, 5 is a voice output device for outputting a voice synthesized by the voice synthesizing unit 4, and 6 is a memory for inputting a user's voice.
0 is a voice input device for transferring to 0, 7 is a voice input device 6
A voice parameter analysis unit which reads voice data input from the memory 10 from the memory 10 and analyzes the parameters; a memory parameter analysis unit 8 which reads the synthesis parameters calculated by the synthesis parameter calculation means 2 and the voice parameters calculated by the voice parameter analysis means 7 from the memory 10 It reads out, calculates a parameter difference value, and stores the parameter difference value storage device 9 or the memory 1
A parameter comparison unit 9 for transferring the parameter value to 0 is a parameter difference value storage device for storing the difference value calculated by the parameter comparison unit 8. FIG. 2 is a flowchart illustrating a procedure in the synthesis parameter calculation unit. FIG. 7 shows the data structure of the memory 10.

First, a user inputs characters using the character string input device 1. The character string input device 1 may be a portable recording medium reading device, a character string transmission unit via a communication line, a voice recognition device, or the like, as long as the device can input a character string in addition to the keyboard. Here, it is assumed that a character string “Today's topic” is input as an example.
The character string is stored in the memory 1000 at 1000. Next, the synthesis parameter calculation unit reads a character string from the memory 1000 and divides the character string into phonemes (step S101). A method of dividing a character string into phonemes is described in, for example, the method of Miyazaki et al. (“Language processing system for Japanese sentence speech output” Transactions of Information Processing Society of Japan, Vol. 27, No. 1).
1, pp. 1053-1061, 1986). Of course, the calculation method is an example, and a method of dividing another phoneme may be used. Thus, the character string “today ’s topic” is “ky / o / o / n / o / w / a”.
/ D / a / i ”, and the phoneme division data is stored in 1001 of the memory 10. Here, "k
“y”, “w”, “a”, etc. are symbols indicating phonemes. Of course, the phoneme symbol data is an example, and other phoneme symbol expressions may be used. The unit to be obtained is not limited to a phoneme unit, but may be a unit obtained by dividing a phoneme into two or a syllable. Next, the phoneme divided data divided into phonemes is read from 1001 of the memory 10, the duration of each phoneme is calculated (step S102), and the duration data is transferred to 1002 of the memory 10. The method of calculating the duration time for each phoneme is described in, for example, the method of Sakasaka et al. (“Phonological time length control for speech synthesis by rules”, IEICE Transactions, Vol. J67-A, N
o. 7, pp. 629-636, 1984). Of course, the calculation method is an example, and another calculation method of the phoneme duration may be used. In this way, the phoneme division data “ky / o / o / n / o / w / a / d /
From “a / i”, for example, the duration time data “50/150/120/40/150/20/200” in millisecond units
/ 40/180/170 "is obtained as the calculation result,
It is stored in 1002 of the memory 10. Of course, the duration is an example, and duration data in seconds or the like may be used. Next, from 1000 and 1001 of the memory 10,
The character string and the phoneme division data are read, an accent for designating the pitch of the sound is given (step S103), and the accent data is transferred to 1003 of the memory 10. The method of adding an accent to a character string is described, for example, by the method of Sakasaka et al. (“Accent Rules for Japanese Word Chains” IEICE Transactions, Vol. J66-D, No. 7, pp. 849-
856, 1983). Of course, the calculation method is an example, and another accenting method may be used. In this way, the character string “today's topic” is converted into accent data “kyo'ono, wadai.” And stored in 1003 of the memory 10. Here, syllables with “′” are syllables with accent nuclei, “,” is a phrase delimiter, and “.” Is a symbol indicating the end of a sentence. Of course, this symbol is an example, and other symbols may be used. Next, the phoneme divided data and the accent data are read from 1001 and 1002 of the memory 10 and the fundamental frequency is calculated (step S104).
To 1004. The calculation method of the fundamental frequency is described in, for example, the method of Fujisaki et al.
445-453, 1971). Of course, the calculation method is an example, and another calculation method of the fundamental frequency may be used. In this way, the string "today's topic"
The data is converted into basic frequency data (F0, F1,..., Fi,..., Fp) and stored in the memory 1004. Here, Fi is a numerical value representing the fundamental frequency, for example, a value obtained in units of 10 milliseconds. P is the number of values representing the fundamental frequency. Of course, the method of representing the fundamental frequency is merely an example, and if the method is capable of finding the value of the fundamental frequency, the method of finding the fundamental frequency in a time unit other than a fixed time interval, and the parameters of the model for finding the fundamental frequency May be used.
Next, the phoneme duration data is read from the memory 1002 in the memory 10 and the power is calculated (step S105).
The power data is transferred to 1005 of the memory 10. The method of power calculation is described in, for example, the method of Mimura et al. (“Analysis and Control of Speech Power Using Statistical Method”, Journal of the Acoustical Society of Japan, 4
9 pp. 253-259, 1993).
Of course, the calculation method is an example, and another power calculation method may be used. In this way, the character string "today's topic" is (P0, P1, ..., Pi, ..., Pq)
1005 of the memory 10
Is stored in Here, Pi is a value representing power and d
Stored as B value. q is the number of power data. Of course, the method of expressing the power value is an example, and another power value expressing method may be used. Next, the phoneme division data is read from the memory 1001, the cepstrum is calculated (Step S 106), and the cepstrum data is transferred to the memory 1006. As a method for obtaining cepstrum data, for example, a method using a result of cepstrum analysis of a representative waveform for each phoneme is used. Of course, the calculation method is an example, and another cepstrum analysis method may be used. In this way, the string "today's topic"
It is converted into cepstrum data (C0, C1,..., Ci,... Cr) and stored in 1006 of the memory 10. Here, r is the number of cepstrum data, Ci
Is vector data representing cepstrum coefficients, (c
1,. . . , Ci,. . . , Cs). s indicates the order of the cepstrum coefficient. Of course, the method of representing cepstrum data is merely an example, and other expressions may be used.

In this way, the phoneme duration data, the fundamental frequency data, the power data, and the cepstrum data, which are parameters representing the characteristics of speech, are stored in the memory 1002, 1004, 1005, and 1006, respectively. In the present embodiment, phoneme duration data, fundamental frequency data, power data, and cepstrum data are used as parameters representing voice characteristics. However, if parameters represent voice characteristics, the formant frequency, linear prediction It may be a parameter such as a linear prediction coefficient obtained as a result of the analysis or a mel cepstrum converted to a mel scale.

Subsequently, the user uses the voice input device 6 to say "today's topic" and capture the voice into the device.
For example, audio is captured as monaural audio with a sampling frequency of 16 kHz and a quantization bit number of 16 bits. Next, the voice input device 6 stores the voice data in the memory 10
Transfer to 7. Next, the voice parameter analysis means 7 reads out the voice data from the memory 1007 of the memory 10, analyzes the fundamental frequency, and obtains the fundamental frequency data (F'0, F ').
1,. . . , F'i,. . . , F′k) to the memory 1009. Here, F′i is the value of the fundamental frequency, and k is the number of values of the fundamental frequency data. As a method of analyzing the fundamental frequency, for example, a method of obtaining the fundamental frequency from the autocorrelation function of the residual signal of the LPC analysis is used.
Of course, the calculation method is an example, and another fundamental frequency analysis method may be used. Here, the fundamental frequency analysis interval is set equal to the basic unit when the fundamental frequency data is obtained from the character string “today's topic”. For example, the fundamental frequency data is analyzed in units of 10 milliseconds. Of course, the analysis method is merely an example, and any other fundamental frequency analysis method and data structure of the fundamental frequency may be used as long as the method can calculate a difference value of the fundamental frequency described later. In this way, the fundamental frequency data of the voice "today's topic" is stored in the memory 1
0 is stored in 1009. Next, 100 of the memory 10
7, the voice data is read out, the phoneme duration is analyzed, and the phoneme duration data is transferred to 1008 of the memory 10. The analysis of the phoneme duration is performed, for example, according to the method of Nakagawa et al. (“Segmentation of continuous speech using HMM method and Bayesian probability” Transactions of the Institute of Electronics, Information and Communication Engineers, Vol. J.
72-D-II pp. 1-10, 1989). Of course, the calculation method is an example, and another analysis method of the phoneme duration may be used. In this way, the phoneme duration data “80/140” of the voice “Today's topic”
/ 150/30/150/50/180/50/180
/ 180 ”is stored in 1008 of the memory 10. Here, the unit of the time length data is milliseconds as in the phoneme duration calculated from the character string “today's topic”. However, the unit is, of course, an example, and the difference of the phoneme duration is described later. Other phoneme duration calculation methods and data representation methods may be used as long as the value can be calculated. Next, audio data is read from the memory 1007 of the memory 10, the power is analyzed, and the power value is stored in the memory 10.
010. The power analysis uses, for example, the zero-order term of the cepstrum parameter obtained as a result of the cepstrum analysis. Of course, the calculation method is an example, and another power data calculation method may be used. In this way,
Power data of voice "Today's topic"(P'0, P '
1,. . . , P'i,. . . , P′j) are stored in 1010 of the memory 10. Here, P′i is power data, and j is the number of power data. Of course, the expression method is merely an example, and any other power data expression method may be used as long as a difference value of power data described later can be calculated. Next, voice data is read from 1007 of the memory 10, cepstrum analysis is performed, and cepstrum parameter data is transferred to 1011 of the memory 10. The unit time of the cepstrum analysis is, for example, equal to the cepstrum parameter stored in 1006 of the memory 10. In this way, the audio "Today's topic"
Cepstrum data (C′0, C′1,..., C ′)
i,. . . C′m) and 1006 of the memory 10
Is stored in Here, m is the number of cepstrum data,
C′i is vector data representing a cepstrum coefficient,
(C'1, ..., c'i, ..., c'n). n indicates the order of the cepstrum coefficient. Of course, the method of representing cepstrum data is merely an example, and other expression methods may be used as long as a difference value of cepstrum described later can be obtained.

According to the above procedure, the phoneme duration, fundamental frequency, power, and cepstrum data are stored in the memory 10.
08, 1009, 1010, and 1011.

Subsequently, in the parameter comparing means 8, the synthesized parameters calculated by the synthesized parameter calculating means 2 and the analysis parameters analyzed by the voice parameter analyzing means 7 are stored in the memory 10 in 1002, 1004, 1005, 1006.
And 1008, 1009, 1010, and 1011, and the corresponding parameters are compared. Here, the corresponding parameters are the fundamental frequency in the synthesis parameter and the fundamental frequency in the analysis parameter, the phoneme duration in the synthesis parameter and the phoneme duration in the analysis parameter, the power data in the synthesis parameter and the power data in the analysis parameter, and the synthesis. It is a cepstrum data in a parameter and a cepstrum data in an analysis parameter.

First, a method of calculating the parameter difference value relating to the phoneme duration in the parameter comparing means 8 of FIG. 1 will be described with reference to FIG. FIG. 4 is a specific example implemented for the character string “Today's topic (Kyo-Onawadai)”. Reference numeral 401 denotes synthesis parameters obtained as a result of the phoneme duration calculation procedure 103 in FIG. The synthesis parameter 401 is a parameter value unique to the system, and is stored in the format of 401. Synthesis parameter 4
01 is defined by a combination of a phoneme symbol 402 and a duration 403, and has a different value for each phoneme symbol. The unit of the duration 403 is millisecond. For example, as shown at 404, the phoneme “o” has a continuation length of 150 milliseconds. In this way, "Today's topic"
Is obtained for the character string.
Reference numeral 405 denotes an analysis parameter table indicating a result of analyzing a voice uttered by the user with respect to the content of the character string “today's topic”. The phoneme symbol 406 and the duration 407 are the same as the synthesis parameter 401. “O” 408 in the analysis parameter corresponding to “O” in 404 indicates a value different from that of the synthesis parameter 401. This means that the utterance of the user is different from the synthesis parameter assumed in the system, and can be interpreted as the personality of the user. 421 is an analysis parameter obtained by the above procedure, and 422 is a synthesis parameter. 42
Reference numeral 3 denotes a differentiator for calculating a difference value 424 between the analysis parameter 421 and the synthesis parameter 422. In this embodiment, the difference unit 423 calculates the difference between the continuation lengths of the corresponding items of the phoneme symbol 402 and the phoneme symbol 406 and stores the difference as a parameter difference value table 409. For example, the value of 404 is subtracted from the value of 408 for the first “o”. As a result, the difference value “30” becomes 4 as the difference continuation length.
12 is stored. The parameter difference value 409 is obtained using the above method. Thus, "30 / -10 /
30 / −10 / 0/30 / −20 / 10/0/10 ”is stored in 1012 of the memory 10.

Hereinafter, similarly, the fundamental frequency, power,
Compute the difference value for the cepstrum parameter.

In this manner, the difference between the phoneme duration, the fundamental frequency, the power, and the cepstrum is stored in the memory 10 as 1
012, 1013, 1014, and 1015.
Here, the difference value stored in the memory 10 can be stored and stored in a storage medium that is the parameter difference value accumulating device 9, or the difference value can be used by a method described in an embodiment described later.

Here, the principle of obtaining the synthesis parameter from the difference value data will be described with reference to FIG. 301 is FIG.
The voice data 302 of the user input by the voice input device 6 is a voice parameter analysis unit that analyzes the voice data 301 and outputs an analysis parameter representing a feature of the voice.
Reference numeral 03 denotes character string data input by the character string input device 1 of FIG. 307 is a parameter addition unit that calculates a synthesis parameter from the synthesis parameter calculated from the character string data 303 and the parameter difference value 306, and 308 synthesizes speech from the synthesis parameter output by the parameter addition unit 307. This is a speech synthesis unit that performs Here, the parameter comparison unit 305 and the parameter addition unit 307 use the same combined parameter,
Different parameters may be used. Using the above principle, the voice 30 having the voice characteristics of the user's voice 301
9 can be synthesized.

Using the above principle, a composite parameter is calculated from the parameter difference value. In the present embodiment, the synthesis parameters are calculated by a method described below. Suppose that the parameter difference value d is obtained from the synthesis parameter p and the synthesis parameter q.
Let f = d (p, q), where fd is the calculation to obtain. Here, the parameter addition method fs for obtaining the synthesis parameter p from the difference value d uses an operation fs that satisfies the following relationship. That is, fs that satisfies p = fs (fd (p, q)) is adopted. For example, in the above example, since the parameter difference is used for the parameter comparison method, the relationship can be satisfied by using the sum of the parameters for the parameter addition method. Of course, the relationship is merely an example, and a parameter addition method that can calculate a parameter representing the feature of the speaker's voice from the difference value may be used. The synthesis parameters calculated in this way are stored in the memory 10 at 1002, 100
4, 1005, and 1006. Subsequently, the voice synthesizing unit 4 stores the data in the memory 1002, 1004, 100
5, a synthesis parameter is read out from 1006, and voice synthesis is performed by driving a synthesis filter to generate voice data. As the speech synthesis method, for example, a method described in the document “Furui:“ Digital Speech Processing ”, p. 22, Tokai University Press, 1985” is used. Of course,
The speech synthesis method is an example, and another speech synthesis method may be used. The audio data generated in this way is stored in the memory 1016. Next, the audio data stored in 1016 of the memory 10 is output through the audio output device 5. According to the above procedure, a desired voice is synthesized.

The above embodiment is an example in which a user inputs a character string "today's topic" and calculates a difference value when a user inputs a voice "today's topic". For a plurality of difference values obtained as a result of performing analysis on voices uttered for a plurality of character strings, an average difference value obtained by calculating an average of the difference values, or a value such as an average value and a variance value regarding the difference value, It may be used as a difference value. Further, the difference value may be any value that indicates a difference from the reference parameter, for example, by using a ratio with the reference parameter.

The voice synthesizer only needs to be able to output the difference value and the synthesized voice. For example, the voice synthesizer is stored in a computer device having a character string input device and a voice input device, and the CPU of the computer device is read and executed. It may be a program to be performed.

Next, another embodiment of the present invention will be described with reference to FIG. 50 is the system that sends the mail,
Reference numeral 60 denotes a mail receiving side system, 56 denotes a mail transmitting side system communication device, 66 denotes a mail receiving side system communication device, and 58 denotes a communication path through which the communication device 56 and the communication device 66 can communicate. When the sender of the mail uses the mail sender system 50 to send, for example, an e-mail with the content “Today's meeting has been postponed” to the e-mail recipient, the e-mail sender is First, CPU5
After instructing 5 to read and execute the mail transmission program from the memory 54, the character input device 51 inputs a character string "Today's meeting has been postponed." Here, it is assumed that the mail transmission program is transferred from a recording medium such as a fixed disk and stored in the memory 54 before being executed. Of course, other program storage methods may be used as long as they are stored in the memory 54 before execution. The input character string is displayed on the display device 53 after being transferred to the memory 54.
Subsequently, the user uses the voice input device 52 to input a voice saying "Today's meeting has been postponed."
The input voice is transferred to the memory 54. continue,
The mail transmission program reads the speech synthesis program having the above-described functions from the memory 54 and starts execution.
Here, it is assumed that the speech synthesis program is transferred from a recording medium such as a fixed disk and stored in the memory 54 before being executed. Of course, the memory 5
4, other program storage methods may be used. Next, the speech synthesis program reads the input character string from the memory 54, generates a reference synthesis parameter of the system for speech synthesis, and transfers the generated reference synthesis parameter to the memory 54. Next, the CPU 55 reads the input voice from the memory 54, generates an analysis parameter representing a feature of the voice, and transfers the analysis parameter to the memory 54. Next, the speech synthesis program compares the reference synthesis parameter stored in the memory 54 with the analysis parameter, calculates a parameter difference value, and transfers the parameter difference value to the memory 54. Next, the mail transmission program reads the character string “Today's meeting has been postponed.” And the difference value calculated by the voice synthesis program from the memory 54, and, for example, reads the difference After combining the data indicating that the value is included into one mail data, the mail data is transferred to the memory 54. FIG. 6 shows an example of the mail data. 507, mail data; 501, information indicating the destination of the mail data;
Is the information of the user who sends the mail, 503 is the title of the mail, and 504 is the content of the body of the mail. In this embodiment, the character string "Today's meeting has been postponed."
Reference numeral 505 denotes a difference value information presence / absence flag indicating that the difference value is attached to the mail, and 506 denotes a difference value calculated from the voice of the mail sender. The mail transmission program reads the mail data from the memory 54, and transmits the data to the communication device 66 using the communication device 56. As a result of the above,
E-mail data consisting of a character string “Today's meeting has been postponed”, a difference value bearing the characteristics of the sender's voice, and additional data indicating that difference value data has been added , Sent to the recipient system.

On the other hand, the mail receiver reads the mail receiving program from the memory 64 to the CPU 65,
Give instructions to run. Here, it is assumed that the mail receiving program is transferred from a recording medium such as a fixed disk and stored in the memory 54 before being executed. Of course, other program storage methods may be used as long as they are stored in the memory 54 before execution. The mail receiving program checks the presence of mail data stored in the communication device 66, and reads the mail data into the memory 64 when mail has arrived. The mail receiving program reads out the mail data from the memory 64, and refers to the additional data indicating that the difference value data has been added, and in FIG. 6, the difference value information presence / absence flag 505 to check whether the difference value has been added. If it is added, a program for dividing the character string and the difference value is executed, the character string and the difference value are transferred to the memory 64, and the received character string is displayed on the display device 63. When the difference value data is added to the mail, the mail receiving program reads out from the memory 64 a voice synthesizing program having a function of synthesizing voice based on the difference value, and executes it. Here, it is assumed that the speech synthesis program is transferred from a recording medium such as a fixed disk and stored in the memory 54 before being executed. Of course, the memory 54
Otherwise, another program storage method may be used. For example, a semiconductor memory provided in the device,
Data may be read from an external optical disk, DVD, magnetic disk, or another system connected by communication means such as network computing. The communication means does not need to be wired, and may be a communication means such as wireless, light, or infrared. The speech synthesis program thus read out on the scale 54 reads out the character string and the difference value from the memory 64, and reads the character string "Today's meeting has been postponed." A parameter is calculated, a composite parameter is calculated by adding the reference parameter and the difference value, and the calculated parameter is transferred to the memory 64. Next, the voice synthesizing program reads out the synthesis parameters from the memory 64 to perform voice synthesis, and then transfers the voice data to the memory 64. Next, the mail receiving program reads the voice data from the memory 64 and outputs the voice from the voice output device 62.

In the above embodiment, the mail sending / receiving program and the speech synthesizing program are described as separate programs. However, the mail sending / receiving program and the speech synthesizing program may be configured as a speech synthesizing program which is a part of the mail sending / receiving program.

Here, the difference value between the character string input by the mail sender and the voice corresponding to the character string is calculated. However, if the mail sender has already used the mail system, the voice input is performed. Alternatively, a method may be employed in which the calculated difference value is searched from the memory and added to the character string. Further, even if the difference value is not added to the mail, the receiving system searches the sender information of the mail for the difference value information stored in the memory and sets the difference value information of the sender. By adopting the method, it is possible to synthesize a speech having characteristics of the sender's speech.

[0028]

As described above, if the apparatus of the present invention is used, the contents to be synthesized by voice can be adapted to the characteristic parameters of a specific individual, so that the specific personal voice is output. This is an excellent effect for a system where it is desirable, especially an electronic mail reading system.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of the present invention.

FIG. 2 is a diagram illustrating an example of a processing procedure of a synthesis parameter calculation unit according to the present invention.

FIG. 3 is a diagram showing the principle of the present invention.

FIG. 4 is a diagram illustrating a parameter difference value of a phoneme duration, showing a specific example of a parameter comparison unit according to the present invention.

FIG. 5 is a block diagram showing a configuration of the present invention.

FIG. 6 is an example of mail data in the present invention.

FIG. 7 is a diagram showing a data structure of a memory according to the present invention.

[Explanation of Signs] 1 ... Character string input device, 2 ... Synthesis parameter calculation unit, 3 ...
Parameter addition section, 4 ... Speech synthesis section, 5 ... Speech output device, 6 ... Speech input device, 7 ... Speech parameter analysis section, 8
... parameter comparison unit, 9 ... parameter difference value storage device,
Reference Signs List 10: memory, 50: transmitting system, 51: character input device, 52: voice input device, 53: display device, 54: memory, 55: CPU, 56: communication device, 58: communication channel,
60 receiving system, 61 character input device, 62 voice output device, 63 display device, 64 memory, 65 C
PU, 66: communication device, 301: user's voice, 302
... Voice parameter analysis unit, 303, character string, 304, synthesis parameter calculation unit, 305, parameter comparison unit, 30
6 parameter difference value, 307 parameter addition unit, 3
08: voice synthesis unit, 309: voice having the characteristics of the user's voice, 401: synthesis parameter table, 402: phoneme symbol, 403: duration, 404: duration for "o",
405: analysis parameter table, 406: phoneme symbol, 407
… Analysis continuation length, 408… continuation length for “o”, 409
... Parameter difference value table, 410: Phoneme symbol, 411, Difference duration, 412, Duration of “o”, 421, Analysis parameter, 422, Synthesis parameter, 423, Difference device, 424, Parameter difference value, 501 Destination, 50
2 ... sender information, 503 ... title, 504 ... mail text, 505 ... difference value information presence / absence flag, 506 ... difference value,
507 ... Mail data.

 ──────────────────────────────────────────────────続 き Continued on the front page (72) Inventor Keiko Fujita 1-280 Higashi Koigakubo, Kokubunji-shi, Tokyo Inside the Hitachi, Ltd. Central Research Laboratory

Claims (6)

[Claims] 1. A character string input unit for inputting a character string, a synthesis parameter calculation unit for calculating a parameter representing a characteristic of voice from the character string, a voice synthesis unit for synthesizing voice from the parameter, and the voice synthesis unit A voice synthesizing device including a voice output unit that outputs a voice synthesized by the unit, a voice input device that inputs a voice, a voice parameter analysis unit that analyzes the input voice and calculates a parameter representing a characteristic of the voice,
A parameter comparison unit that compares the parameters output by the synthesis parameter calculation unit and the voice parameter analysis unit and outputs a difference value; and a parameter addition unit that adds the difference value and the parameter generated by the synthesis parameter calculation unit. And a voice synthesizing unit for synthesizing voice from the parameters output by the parameter adding unit. 2. A synthesizing parameter calculating section for calculating a first parameter representing a characteristic of a voice when the input character string data is output as a voice, and representing a characteristic of the voice data of the character string separately input. A voice parameter analysis unit for calculating a second parameter, the first parameter and the second
And a means for synthesizing the speech of the character string based on the difference. 3. A speech synthesis program comprising: a synthesis parameter calculation unit for calculating a parameter representing a feature of speech from a character string; and a speech synthesis unit for synthesizing speech from the parameter to generate speech data. A voice parameter analysis unit that calculates a parameter representing a feature of the voice, a parameter comparison unit that compares the parameters output by the synthesis parameter calculation unit and the voice parameter analysis unit and outputs a difference value, Voice synthesis, comprising: a parameter addition unit that adds a parameter generated by the synthesis parameter calculation unit; and a function of generating voice data by synthesizing voice from the parameter output by the parameter addition unit. A recording medium on which a program is recorded. 4. A computer-readable storage medium holding a program read and executed by a device for outputting text data by voice, wherein the storage medium analyzes the character string data and stores a standard data provided in advance. Obtaining a synthetic parameter at the time of typical voice output; obtaining a characteristic parameter necessary for resynthesizing the voice from the input voice read out of the character string; and a difference parameter between the synthetic parameter and the characteristic parameter. , A step of transmitting the difference parameter to another system, and a step of outputting a voice of a character string based on the received difference parameter. Storage medium. 5. A memory for storing a reference parameter of a voice set to a predetermined reference value, a voice input device for inputting a voice, and difference information indicating a difference between the parameter of the input voice and the reference parameter. Created and transmitted to the outside together with the text data, and, based on the difference information transmitted together with the text data received from the outside, the synthesized speech parameters obtained by changing the reference parameters stored in the memory, A mail transmitting / receiving device comprising: an arithmetic control unit for generating synthesized voice data of received text data; and a voice output device for outputting a synthesized voice in accordance with the synthesized voice data. 6. A memory for storing a reference parameter of a voice set to a reference value for predetermined text data, and difference information indicating a difference between an input voice parameter and the reference parameter are created. The synthesized speech parameter transmitted to the outside together with the text data, and obtained by changing the reference parameter stored in the memory based on the difference information transmitted together with the text data received from the outside, An e-mail transmitting / receiving apparatus comprising: an arithmetic control unit that generates synthesized voice data of the above; and a voice output device that outputs a synthesized voice according to the synthesized voice data.