JPH0216681A - Winking signal generating system for face animation picture synthesizing - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed description of the invention] [Table of contents] Overview Industrial field of application Prior art (Figure 19) Means for solving the problem to be solved by the invention (Figure 1) Effect (Figure 1) ) Embodiment Description of the first embodiment (Figs. 2 to 12) Description of the second embodiment (Fig. 13) Description of the third embodiment (Figs. 14 to 18) Effects of the invention [Summary] At the time of initialization By using a small amount of initialization data that is transmitted, it is possible to generate a blink signal for facial dynamic image synthesis in a device that synthesizes and displays facial dynamic images on the receiving side according to audio information transmitted during communication. Regarding the method of blinking, the aim is to make it possible to change the way the blinking occurs depending on whether you are talking or not, and to make it possible to synthesize more natural video images. , 1 The average value and standard deviation of the time interval of each blink during utterance and non-utterance are transmitted as initialization data, and during communication, the average value and standard deviation of the time interval of each blink during utterance and non-utterance are transmitted. at time intervals that follow a normal distribution according to .

A pulse train signal for vocalization and a pulse train signal for non-vocalizing are respectively generated, and the receiving side outputs the pulse train signal for vocalization as a blink signal during vocalization according to the detection signal detected by the audio input detection section. , during non-utterance, the pulse train signal for non-utterance is output as a blink signal.

[Industrial Application Field] The present invention uses a small amount of initialization data transmitted at the time of initialization to synthesize a moving image of a face on the receiving side according to audio information transmitted during communication. The present invention relates to a method of generating a blink signal for facial moving image synthesis in a display.

The goal is to eventually adopt a transmission method that uses public telephone lines for television (TV) telephones, TV conferences, etc., and for this reason, it is necessary to compress the obtained image information as much as possible. It is requested.

[Prior Art] Images transmitted in video telephones etc. are usually moving images of people, but such moving image information.

As shown in FIG. 19, in the conventional method 4, the input image is transmitted independently of the audio information, that is, the input image is generated as an analog image signal by the TV camera 61 on the transmitting side, and the input image is The signal is converted into a digital signal by an image encoding device 62, encoded, compressed, and sent to the receiving side. On the receiving side, an image decoding device 63 decodes the received image into an original signal and displays it on a display 64 as an output image.

In addition, the input voice is obtained as voice information by the micro 5 on the transmitting side, and is then compressed by voice-specific encoding in the voice encoding device 66. Then, on the receiving side, it is decoded by the voice decoding device 67, and then it is transmitted to the speaker. 68 as output audio.

However, since the amount of information contained in moving images is large, this conventional video transmission method that has been commonly used cannot use low bit rate communication lines.
There were problems in that the cost was high and that it was far from being applicable to TV telephones using public telephone lines.

Therefore, the sending side sends, for example, still image information of the face in advance, and the receiving side deforms only the mouth part to match the audio information sent from the sending side. It is also possible to play back images.

However, this has the problem that the eyelids, which play an important role in facial expressions, do not move at all, which increases the unnaturalness.

Therefore, in addition to the deformation of the mouth area, an image transmission method has also been developed that makes it possible to further compress the information in the original video while making the facial expressions look less unnatural by making the eyes blink randomly. Proposed.

[Problem to be solved by the invention] However, with the conventional means of deforming the mouth and blinking randomly, the occurrence of blinking is completely random, so it is difficult to talk. No matter when the eyes are blinking or not, the way the eyes blink remains the same, and it still feels unnatural, so some kind of improvement is desired.

The present invention was devised under these circumstances, and allows for the synthesis of more natural moving images by making it possible to change the way the eyes blink depending on whether they are talking or not. made possible.

The purpose of this invention is to provide a blink signal generation method for facial dynamic image synthesis.

[Means to solve the problem] FIG. 1 is a block diagram of the principle of the present invention.

In FIG. 1, reference numeral 28 denotes a blink signal generation section for facial moving image synthesis;

Standard normal random number table 281. 1st. Second random number converter 282, 283. 1st. Second pulse generator 284,2
85. Audio input detection section 286. It is configured to include a pulse train selection section 287.

Here, the standard normal random number table 281 is a random number series Ui (i=1.2,
3. ..., nj; n is a sufficiently large integer).

The first random number conversion unit 282 receives the average value m and the standard deviation σ of the time interval of blinking during utterance at the time of initialization, and when communication is started, the standard normal random number table 282
The random value Ui is read from 1 of 1, and it is converted as shown in equation (1) to convert it into a random value X that follows a normal distribution with a mean value m and standard deviation σ1.Similarly, the second The random number conversion unit 283 also receives the average value m2 and standard deviation σ2 of the blink time during non-utterance at the time of initialization, and when communication starts, it converts the random number Ui from 1 in the standard normal random number table 281. This is read out and subjected to conversion as shown in equation (2) to convert it into a random value X that follows a normal distribution with an average value m2 and a standard deviation σ2.

X=UiX a1+m1 (X>O) --(1)X
= UiXcr, +m, (where X > 0) -- (2) When the first pulse generator 284 receives the random value When it becomes equal to the value of the random number value X, a pulse is generated, and then a control signal is generated to the first random number conversion unit 282 to input the next value of the random number value X, and the same process is repeated.

Similarly, when the second pulse generator 285 receives the random number X from the second random number converter 283, it counts the clock, and the count value becomes the random number X. When it becomes equal to the value of This outputs the following.

The audio input detection unit 286 samples the energy of the transmitted audio at regular time intervals.

If the energy is larger than a preset threshold value, it is turned on, and if it is smaller, it is turned off, thereby detecting whether vocalization is occurring or not.

The pulse train selection section 287 outputs the pulse P□ from the first pulse generation section 284 while the voice input detection section 286 detects that the voice is being uttered.
While non-voice is detected in step 6, the second pulse generator 285 switches to output the pulse P2.

[Function] With this configuration, at the time of initialization, the average value mi and the standard deviation σ1 of the time intervals of blinking during vocalization are transmitted to the first random number conversion unit 282 as initialization data, and the non-vocalizing The average value m of the blink time interval and the standard deviation σ2 are determined by the second random number conversion unit 283
transmitted to.

Then, during communication, the average value of the time interval of each blink during utterance and non-utterance is determined by the average value m×1m2 and the standard deviation σ1.
．． The first . The second pulse generating sections 284 and 285 generate a pulse train signal P1 for vocalization and a pulse train signal P2 for non-vocalizing, respectively.

Furthermore, on the reception side, the pulse train selection section 287 switches according to the detection signal detected by the audio input detection section 286, so that during vocalization, the pulse train signal P1 for vocalization is selected.
is output as a blink signal, and during non-utterance, a non-utterance pulse train signal P2 is output as a blink signal.

This allows you to blink your eyes differently depending on whether you are talking or not.

[Example] Hereinafter, the present invention will be described in detail with reference to the drawings.

(a) Description of First Embodiment FIG. 2 is a block diagram showing a first embodiment of the present invention. In this first embodiment, a transmitting section 10 and a receiving section 20 are provided, and the transmitting section 10 includes: It includes an image processing section 11 that performs image processing on facial image input, and an audio encoding section 12 that encodes audio input.

The receiving unit 20 also includes a background image memory 19. Audio decoding section 21. Speech recognition unit 22. Code book 23A, mouth shape model transformation section (mouth shape model image storage means) 24A, control point coordinate memory (table) 23B, shadow model transformation section (eyelid shape model image storage means) 24B9 synthesis section 25. Interpolation point calculation unit 27. Blink signal generation unit 28 for facial dynamic image synthesis. It has a coordinate table control section 29.

Here, the background image memory 19 stores still image data of one frame of a face image sent from the transmitting side at the time of initialization.

Further, the speech decoding section 21 decodes the speech code encoded by the transmission section 10, and the speech recognition section 22 speech recognizes the speech signal output from the speech decoding section 21. 23A sequentially selects a set of mouth shape parameter values from phoneme codes (consisting of vowels or consonants, etc., which are the basic constituent units of speech) output one after another from the speech recognition section 22; The model image storage means 24A transforms the mouth shape model image according to a set of mouth shape parameter values successively selected in the codebook 23.

By the way, as shown in FIG. 4, the codebook 23A includes parameters ■ (for example, the width of the mouth),
(For example, the thickness of the lips), ... tn (For example, the vertical width of the mouth) A table is stored in advance as the personal information. Here, for example, phoneme 1. FIG. 6(a), (
b) and (c).

In addition, the mouth shape model deformation unit 24A obtains one screen (one frame) worth of mouth image data of the specific speaker in advance as the personal information via the background image memory 19, and uses this as the skeleton of the geometric shape of the mouth. The mapped patch model is stored as a mouth shape model image. In this way, even if one screen worth of images of the eyes is first sent from the transmitter 10, the codebook 23A needs to be created in advance.

The interpolation point calculation unit 27 receives the coordinate data of all vertices P, ~P, of the eyelid shape model (see Fig. 7) corresponding to the still image data at the time of initialization, and calculates the coordinate data at each frame point from the start to the end of blinking. Control points p,,p,.

The coordinates of P4 are calculated by linear interpolation, and the data is sent to the control point coordinate memory 23B.

In other words, this eyelid shape model is as shown in FIG.
It is composed of eight vertices P1 to P (each point has a two-dimensional coordinate value of Xt'j) and six triangular patches T1 to Tll formed by connecting these vertices P1 to P. , this eyelid shape model synthesizes the blinking action, so p
,,p,,p are control points (points that change the x, y coordinates)
The other five points are fixed (fixed points).

Then, in this interpolation point calculation unit 27, in addition to the coordinates of the eight vertices P1 to P, p, p, p,
, p, indicating the lowest point of , p.

The coordinate values of the three points P4' are also given, and from the pre-given number N of frames per blink, P2→P2'
→P,,P□→P,'→p,,p4→P.

→ Each section of P4 is linearly interpolated.

The control point coordinate memory 23B stores the blinking motion of the eyelids based on the eyelid parameters of the shadow model image. Specifically, three control points p, , p, at each frame time point from the start to the end of blinking are interpolated and calculated by the interpolation point calculation unit 27.

The coordinates of P4 are stored in the control point coordinate memory 23B in the form of a table.
It is stored in the storage area. An example of the structure of this control point coordinate table is shown in FIG.

The blink signal generator 28 generates a blink signal (pulse signal), and as shown in FIG. Second random number converter 282, 283° 1st. Second pulse generator 284, 285. Audio input detection section 286. It is configured to include a pulse train selection section 287.

Here, the random number generator 280 generates a pointer value ill 12 that sets a random number start position according to a signal input at the time of initialization.
(1≦11t'12≦n) is randomly generated.

The standard normal random number table 281 includes a random number sequence Ui (i
==1, 2, 3.・a and nun are sufficiently large integers)
This is a table (memory) that stores the values of .

At the time of initialization, the first random number conversion unit 282 uses the average value m of the time interval of blinking during utterance, @ standard deviation σ1 and a pointer value 1 for setting the random number start position from the random number generator 280.
1, and communication is started, the random number Ui corresponding to this address iL is read from address 11 of the standard normal random number table 281, and the above-mentioned formula (1) (see below) is read out from address 11 of the standard normal random number table 281.
By performing the following transformation, the random value X is converted into a random value X that follows a normal distribution with an average value m and a standard deviation σ1 as shown in FIG.

x:=tJix a, +m, (X>0)−−(
1) Then, this first random number conversion section 282 waits for a control signal from a first pulse generation section 284, which will be described later, and converts 18 into 1.
Increase by increments and repeat the same process.

Such processing is shown in FIG. 11, that is, first step a1
Then, set the initial values m, σ□p xz, and proceed to step a2.
Then, U corresponding to standard normal random number table 281 to 11
Read i in step a3.

Calculate random value X,=UiXσ□+m0, step a4
Then, determine whether X>0, and if YES, step a
In step 5, a random value X is input, and in step a6, it is determined whether or not a control signal has been input from the first pulse generator 284. If a control signal has been input, step a
7, as i,=il+1, in the first step a8, 1
Determine whether 1≦n. Such processing is i,=n+1
When i,=n+1, in step a9, the process is initialized to 1i=1 and the same process is repeated.

Note that if the random value X becomes a negative value in step a4, step a5. a6 jumps.

Further, in step a6, the process does not proceed to the first step until the control signal is input.

Similarly, the second random number conversion unit 283 also calculates the average value m2 of the blink time during non-utterance at the time of initialization. When communication is started upon receiving the standard deviation σ and the pointer value 12 for setting the random number start position from the random number generator 280, this address 12 is selected from address 12 of the standard normal random number table 281.
Read out the random value Ui corresponding to , and apply the above (2) to this
The random value X is converted into a random value X that follows a normal distribution with an average value m2 and a standard deviation σ2, which is substantially similar to that shown in FIG.

X=UiX a2+m2 (where X>O) --(2) Then, this second random number conversion section 283 also waits for a control signal from the second pulse generation section 285, which will be described later, and increases 12 by 1 and performs the same process. repeat.

Note that the processing flow in this second random number conversion section 283 is also the same as that shown in FIG.

The first pulse generator 284 includes a counter 284a that counts clocks, a comparator 284b that compares the count value from the counter 284a and the random value X from the first random number converter 282, and a match from the comparator 284b. It is equipped with a pulse generator 284C that outputs a pulse when a pulse is generated, and when a random number value X is input from the first random number converter 282, the clock is counted and the count value is equal to the random number value X. When the value becomes equal to the value, a pulse is generated, and then a control signal is generated to the first random number converter 282 to input the value of the 9th order random number X, and the same process is repeated. ) A pulse train signal P as shown in
Similarly, the second pulse generator 28 outputs □.
5 also includes a counter 285a that counts clocks, a count value from this counter 285a, and a second random number converter 28.
3, and a pulse generator 285C that outputs a pulse when a coincidence pulse is output from the comparator 285b. When the numerical value X is input, the clock is counted, and when the count value becomes equal to the value of the random numerical value By inputting the value of numerical value X and repeating the same process, Figure 12 (b
) outputs a pulse train signal P2 as shown in FIG.

The audio input detection unit 286 samples the energy of the transmitted audio at regular time intervals, and turns on if the energy is greater than a preset threshold.
If it is smaller, it is turned off [see FIG. 12(c)], thereby detecting whether vocalization is occurring or not.

The pulse train selection unit 287 outputs the pulse P0 from the first pulse generation unit 284 as a blink start signal while the audio input detection unit 286 detects that the voice is being uttered, and the audio input detection unit 286 outputs the pulse P0 as a blink start signal. While it is detected that utterance is in progress, the pulse P2 from the second pulse generator 285 is switched to be output as the blink start signal, and a multiplexer is used.

Therefore, the output pulse train from the pulse train selection section 287 is as shown in FIG. This allows you to change the way your eyes blink.

By the way, the coordinate table control unit 29 in FIG. 2 sequentially reads out all vertex data in the coordinate table of the control point coordinate memory 23B from the time when it receives the blink start signal from the blink signal generation unit 28, and creates a shadow model for each frame. It is transferred to the deforming section 24B.

The shadow model transformation unit 24B stores a warming model image defined by shadow parameters indicating the geometrical shape of the eyelid portion of the face.The shadow model transformation unit 24B stores the eyelid parameters from the control point coordinate memory 23B The model image is then transformed based on the eyelid parameters.

Specifically, by the action of the coordinate table control unit 29, the eyelid parameters sequentially sent from the control point coordinate memory 23B are taken in, and the dewarming model image is transformed based on the eyelid parameters.

Here, the deformation of this temperature removal model image is schematically shown in FIGS. 8(a) to 8(c).

The synthesis unit 25 combines the self-portrait generated from the mouth shape model transformation unit 24A and the eyelid image generated from the shadow model transformation unit 24B with an image other than the eyes and eyelids of the still face image stored in the background image memory 19. It is something that is synthesized.

Next, the operation of this first embodiment will be explained.

The audio input is encoded by the audio encoder 12 and sent to the receiver 20.
This audio code is decoded by the audio decoding section 21 and output as audio.

On the other hand, this voice output is sent to the voice recognition unit 22, and its phoneme codes are sequentially extracted and stored in the codebook 22.
Sent to 3A. In the codebook 23A, based on the input phoneme code, a set of parameter values 1. II...
・, select n.

Then, based on the selected set of parameter values, the mouth shape model deforming section 24A generates a self-portrait by deforming the mouth shape model image stored in advance. As a result, the correspondence between the generated self-image and the phonemes extracted by the voice recognition section 22 is as shown in FIGS. 6(a), 6(b), and 6(Q), for example.

On the other hand, the blink signal generator 28 generates a blink start signal at different random time intervals depending on whether the phone is busy or not.

That is, at the time of initialization, the average value m of the time interval of blinking during one utterance and Itl standard deviation σ1 are used as initialization data.
The pointer value 11 for setting the random number start position is transmitted to the first random number converter 282 shown in FIG.
Average value of blink time interval during non-utterance, m, 4
! The 111 standard deviation σ2 and pointer values i and I for setting the random number start position are transmitted to the second random number conversion unit 283.

During communication, the average value m1 of the time interval of each blink during one utterance and during non-utterance is determined. m2 and standard deviation σ1
．． The first . The second pulse generators 284 and 285 generate a pulse train signal P for vocalization and a pulse train signal P2 for non-vocal, respectively.

Furthermore, on the reception side, the pulse train selection section 287 switches according to the detection signal detected by the audio input detection section 286, so that during vocalization, the pulse train signal P for vocalization as shown in FIG. During non-speech, a non-speech pulse train signal P2 as shown in FIG. 12(b) is output as a blink start signal.

As a result, different pulse train signals are output depending on whether one episode is being made or not [Figure 12(c)]
, see (d)].

When the pulse train signal is output from the blink signal generating section 28 in this way, the coordinate table control section 29 starts the control point coordinate memory 23 from the time when this blink start signal is received.
All vertex data in the coordinate table of B is read out and transferred to the shadow model transformation unit 24B for each frame. Such transfer ends when the number of frames per unit blink has elapsed since the blink start signal was generated. Then, the shadow model deformation unit 24B generates an eyelid image by deforming the pre-stored warming model image according to the above vertex data.

The self-portrait and eyelid image deformed and generated in this way are combined with the image other than the mouth and eyelids of the still face image stored in the background image memory 19 in the compositing unit 25 to create a moving image of the entire face. This will be output as .

This makes it possible to further compress the information in the original video, significantly reducing the amount of information.As a result, it is possible to realize a low-cost image transmission method that uses a low bit rate line, and also to During a conversation, the person blinks at different intervals depending on whether they are talking or not, making their facial expressions more natural.

The method used to transform the mouth shape model image in the mouth shape model transformation section 24A and the shadow model image in the shadow model transformation section 24B is described in IEICE technical report IE87-2.
．． 87, No. 19, 1987.

(b) Description of Second Embodiment FIG. 13 is a block diagram showing a second embodiment of the present invention. The difference from the first embodiment shown in FIG. 13, and the transmitting side separates the phoneme code and other information (intonation, pitch, etc.) and sends it to the receiver 20. The receiver 20 uses the phoneme code as it is in the codebook 23A, and also separates the phoneme code and intonation, etc. information is synthesized by the speech synthesis section 26 to generate speech output. The other configurations and operations (including the configuration and operations of the blink signal generator) are the same as those in FIGS. 2 and 3. Therefore, in this second embodiment as well, the same effects and advantages as in the first embodiment described above can be obtained.

(c) Description of the third embodiment By the way, in each of the first or more embodiments, since the pre-stored codebook 23A is unique to a predetermined speaker, mouth images of an unspecified number of people should be transmitted. Then,
To rewrite the full mouth form code stored in the codebook to adapt it to the speaker each time the speaker changes, or to record all codebook information for registered speakers. A huge memory area must be prepared for the codebook.

Therefore, in the third embodiment shown below, the codebook can be used to suit unspecified speakers.

That is, as shown in Fig. 14, a standard codebook is created by measuring each parameter value of the mouth shape model for the mouth shape of a character that pronounces all standard human phonemes, and each parameter value in this codebook is A predetermined basic phoneme code (for example, silence code) is normalized (divided) by the parameter value to create a normalized codebook after the parameters (see FIG. 15).

Then, as shown in Fig. 16, a set of parameters are measured from the individual's mouth image corresponding to the basic phoneme code, and the total phoneme code of the normalized codebook is obtained for each parameter as shown in Fig. 15. By multiplying each parameter for , a personal codebook can be created.

That is, 1. For example, if the obtained set of personal mouth image parameters is byo, ~b, n, then in FIG. 15, the phoneme code ■
Then, the normalized code a21/a1□ of parameter ■ is multiplied by the above parameter b1□ (as□/a z) and converted to the code bll, and similarly for parameter l, parameter b11 is completely The phoneme code will be multiplied.

FIG. 17 is a block diagram showing a third embodiment that is provided with an initialization device 30 for creating such a personal codebook, and this initialization device 30 initializes the codebook 23A for personal use. By doing so, dynamic images of an unspecified number of speakers are reproduced.

The specific configuration of this initialization device 30 is shown in FIG. When the mouth image is sent, the feature point extraction unit 31 of the initialization device 30 extracts the feature points of the mouth image. Then, a parameter calculation unit 32 calculates a set of parameters based on the distance between feature points and the like. This set of parameters is
As shown in FIG. 5, the multiplier 34 performs multiplication for each parameter of the normalized codebook prepared in advance in the normalized codebook memory 33 to create a personal codebook memory 35. Store.

From now on, it will be referenced when transmitting that individual's 0 image.

In this way, since the initialization device 30 is provided so that the prepared codebook can be updated for each speaker, it becomes possible to easily deal with an unspecified number of speakers.

Note that this initialization device 30 is similarly applied to the embodiment shown in FIG.

[Effects of the Invention] As described above, according to the blink signal generation method for face dynamic image synthesis of the present invention, the frequency of blink signal generation can be changed depending on whether the person is talking or not. This allows you to change the way your eyes blink.
This has the advantage of being able to synthesize more natural moving images.

[Brief explanation of the drawing]

Fig. 1 is a block diagram of the principle of the present invention, Fig. 2 is a block diagram showing a first embodiment of the invention, Fig. 3 is a block diagram of the blink signal generator, Fig. 4 is a block diagram of the codebook, Figure 5 is a configuration diagram of the control point coordinate table, Figure 6 (a),
(b) and (c) are diagrams showing mouth images corresponding to phoneme codes, Figure 7 is a diagram showing the shape model configuration of the eyelid region, and Figure 8 (a
), (b), and (c) are diagrams explaining the concept of deformation of a shadow model image. Figure 9 shows an average of 0. Figure 10 shows a normal distribution with a standard deviation of 1. FIG. 11 is a flowchart showing the procedure for calculating random numbers. FIG. 12 is a diagram showing waveforms of various parts in the blink signal generation section. FIG. 13 is a second embodiment of the present invention. FIG. 14 is a diagram showing the procedure for creating a normalization codebook in the third embodiment of the present invention, FIG. 15 is a block diagram of the normalization codebook, and FIG. Figure 17 is a block diagram showing the third embodiment of the present invention; Figure 18 is a block diagram of the initialization device; Figure 19 is a conventional general image. FIG. 2 is a system diagram showing a transmission method. In the figure, 10 is a transmitting section, 11 is an image processing section, 12 is a speech encoding section, and 13 is a speech recognition section. 19 is a background image memory, and 20 is a receiving section. 21 is a voice decoding unit, 22 is a voice recognition unit, 23A is a food book, 23B is a control point coordinate memory (table), and 24A is a mouth shape model transformation unit (mouth shape model image storage means). 24B is a shadow model transformation unit (eyelid shape model image storage means)
, 25 is a synthesis section, 26 is a speech synthesis section, 27 is an interpolation point calculation section, 28 is a blink signal generation section, and 29 is a coordinate table control section. 30 is an initialization device, 31 is a feature point extraction unit, and 32 is a parameter calculation unit. 33 is a normalization codebook memory, 34 is a multiplication unit, 35 is a personal codebook memory, 280 is a random number generator, and 281 is a standard normal random number table. 282 is a first random number converter, 283 is a second random number converter, 284 is a first pulse generator, and 284a is a counter. 284b is a comparator, 284c is a pulse generator, 285 is a second pulse generator, 285a is a counter, 285b is a comparator, 285c is a pulse generator, 286 is an audio input detector, and 287 is a pulse train selector. Two Dobu 1 Wame Kyouho Concave Figure 4 Stab fJP Massacre. Resurrection, 11 O water Ypres / LA 6 support] Kyosha Se area /
l-shape 7-degree Makihiro-Ko-su Diagram 7 Diagram Phoneme Technique Phoneme ■ Sound #: II (b) (C) Fuyuumamoto 1; Length size Ororo Yamai family! Hosu sword figure 6 average m+, mark Tan I ding left σ ring regular 7 minutes 卆 beg small whistle 1° figure S! - Hitoshi oJ story (Kurai 1st normal distribution begging diagram d) several % calculation 91 shell showing flow hand diagram, positive Inumi Ihi code 7 Gui cho ~ Ki P1 Kai 7 Su Gu 14th
Figure Pure Silk, Code Book Mewa Iff Drawing Figure 15 A Cord for Solid Person ° BU 7 Gui Ti Might Ill Shellfish Begging Hosulo Figure 16 Size SL Iro

Claims (2)

[Claims] (1) By using a small amount of initialization data transmitted during initialization, depending on the audio information transmitted during communication,
In a device that synthesizes and displays moving images of faces on the receiving side, at the time of initialization, the average value (m_1,
m_2) and the standard deviation (σ_1, σ_2), and during communication, the average value (m_1, m_2) of the time interval of each blink during vocalization and non-vocalization and the standard deviation (σ_
A pulse train signal for vocalization (P_1) and a pulse train signal for non-vocalization (P_2) are generated at time intervals according to a normal distribution according to According to the detected detection signal (S), during vocalization, the pulse train signal for vocalization (P_1) is output as a blink signal, and during non-vocalization, the pulse train signal for non-vocalization (P_2) is output as a blink signal. A blink signal generation method for facial dynamic image synthesis, which is characterized by outputting. (2) Random number sequence Ui (i=1, 2
, 3, . When communication is started, a random value Ui is read from the standard normal random number table (281), and the required conversion is performed on it to make it follow a normal distribution of mean value (m_1) and standard deviation (σ_1). Communication starts upon receiving the first random number conversion unit (282) that converts into a random number value (X), and the average value (m_2) and standard deviation (σ_2) of the time interval of blinking during non-utterance at the time of initialization. Then, the random value Ui is read from the standard normal random number table (281), and the required conversion is performed on it to convert it into a random value (X) that follows a normal distribution of the mean value (m_2) and standard deviation (σ_2). When a random number (X) is input from the second random number converter (283) and the first random number converter (282), the clock is counted and the count value is equal to the value of the random number (X). , it generates a pulse, and then
generating a control signal to the first random number converter (282);
A first pulse generator (284) that inputs the next random number value (X) and repeats the same process, and a second random number converter (
When a random number value (X) is input from the second random number converter (
A second pulse generator (283) generates a control signal, inputs the next random number value (X), and repeats the same process.
85), the energy of the transmitted voice is sampled at fixed time intervals, and if the energy is greater than a preset threshold, it is turned on, and if it is smaller, it is turned off. a voice input detection unit (286) that detects the voice input detection unit (286);
While it is detected that vocalization is being performed, the pulse (P1) from the first pulse generating section (284) is output, and the voice input detecting section (286) detects that no vocalization is being performed. a pulse train selection section (287) that switches to output the pulse (P_2) from the second pulse generation section (285) while blink signal generation method.