WO2021196644A1

WO2021196644A1 - Method, apparatus and device for driving interactive object, and storage medium

Info

Publication number: WO2021196644A1
Application number: PCT/CN2020/129793
Authority: WO
Inventors: 吴文岩; 吴潜溢; 钱晨; 白晨
Original assignee: 北京市商汤科技开发有限公司
Priority date: 2020-03-31
Filing date: 2020-11-18
Publication date: 2021-10-07
Also published as: CN111460785A; SG11202111909QA; CN111460785B; TW202138992A; KR20210124307A; JP2022530935A

Abstract

Disclosed are a method, apparatus and device for driving an interactive object, and a storage medium. The method comprises: acquiring a phoneme sequence corresponding to text data; acquiring a control parameter value of at least one local region of an interactive object that matches the phoneme sequence; and controlling the posture of the interactive object according to the acquired control parameter value.

Description

Driving method, device, equipment and storage medium of interactive object

Related cross references

This application is based on a Chinese patent application with an application number of 2020102458024 and an application date of March 31, 2020, and claims the priority of the Chinese patent application. The entire content of the Chinese patent application is hereby incorporated by reference into this application.

Technical field

The present disclosure relates to the field of computer technology, and in particular to a method, device, device, and storage medium for driving interactive objects.

Background technique

The way of human-computer interaction is mostly based on keystrokes, touches, and voice input, and responds by presenting images, texts or virtual characters on the display screen. At present, virtual characters are mostly improved on the basis of voice assistants.

Summary of the invention

The embodiments of the present disclosure provide a driving solution for interactive objects.

According to an aspect of the present disclosure, there is provided a method for driving an interactive object, the method comprising: obtaining a phoneme sequence corresponding to text data; obtaining a control parameter value of at least one partial region of an interactive object matching the phoneme sequence; The acquired control parameter value controls the posture of the interactive object.

With reference to any of the embodiments provided in the present disclosure, the method further includes: controlling the display device displaying the interactive object to display text according to the text data, and/or controlling the display device according to the phoneme sequence corresponding to the text data Output speech.

With reference to any one of the embodiments provided in the present disclosure, the control parameter of the local area of the interactive object includes the posture control vector of the local area, and the control parameter value of at least one local area of the interactive object matching the phoneme sequence is obtained, The method includes: performing feature coding on the phoneme sequence to obtain a first coding sequence corresponding to the phoneme sequence; obtaining a feature code corresponding to at least one phoneme according to the first coding sequence; obtaining the interaction corresponding to the feature code The attitude control vector of at least one local area of the object.

In conjunction with any one of the embodiments provided in the present disclosure, performing feature encoding on the phoneme sequence to obtain the first encoding sequence corresponding to the phoneme sequence includes: generating for each of the multiple phonemes contained in the phoneme sequence The sub-coding sequence corresponding to the phoneme; and obtaining the first coding sequence corresponding to the phoneme sequence according to the sub-coding sequences corresponding to the multiple phonemes respectively.

With reference to any one of the embodiments provided in the present disclosure, for each of the multiple phonemes included in the phoneme sequence, generating a sub-coding sequence corresponding to the phoneme includes: detecting whether the phoneme corresponds to each time point; By setting the coding value at the time point when the phoneme is present to the first value, and setting the coding value at the time point when the phoneme is not present to the second value, the sub-coding sequence corresponding to the phoneme is obtained.

With reference to any one of the embodiments provided in the present disclosure, the method further includes: for the sub-coding sequence corresponding to each phoneme of the multiple phonemes, using a Gaussian filter to perform continuous values of the phonemes in time Gaussian convolution operation.

In conjunction with any one of the embodiments provided in the present disclosure, controlling the posture of the interactive object according to the obtained control parameter value includes: obtaining a sequence of posture control vectors corresponding to the second coding sequence; and according to the posture control vector The sequence of controls the gesture of the interactive object.

With reference to any one of the embodiments provided in the present disclosure, the method further includes: in the case that the time interval between the phonemes in the phoneme sequence is greater than a set threshold, controlling the parameter value according to the setting of the local area To control the posture of the interactive object.

In conjunction with any one of the embodiments provided in the present disclosure, acquiring the attitude control vector of at least one local area of the interactive object corresponding to the feature code includes: inputting the feature code into a pre-trained recurrent neural network to obtain the The attitude control vector of at least one local area of the interactive object corresponding to the feature code.

With reference to any one of the embodiments provided in the present disclosure, the recurrent neural network is obtained through feature coding sample training; the method further includes: obtaining a video segment of a character's voice, and obtaining a plurality of video segments containing the character according to the video segment Extract the corresponding voice segment from the video segment, obtain a sample phoneme sequence according to the voice segment, and perform feature encoding on the sample phoneme sequence; obtain at least the corresponding voice segment corresponding to the first image frame A feature encoding of a phoneme; converting the first image frame into a second image frame containing the interactive object, and obtaining the attitude control vector value of at least one local area corresponding to the second image frame; controlling according to the attitude The vector value is used to annotate the feature code corresponding to the first image frame to obtain the feature code sample.

With reference to any one of the embodiments provided in the present disclosure, the method further includes: training the initial recurrent neural network according to the characteristic coding samples, and training to obtain the recurrent neural network after the change of the network loss satisfies the convergence condition, wherein The network loss includes the difference between the attitude control vector value of the at least one local area predicted by the recurrent neural network and the marked attitude control vector value.

According to an aspect of the present disclosure, there is provided a driving device for an interactive object, the device including: a first acquiring unit for acquiring a phoneme sequence corresponding to text data; a second acquiring unit for acquiring a phoneme sequence matching the phoneme sequence The control parameter value of at least one partial area of the interactive object; the driving unit is used to control the posture of the interactive object according to the acquired control parameter value.

According to an aspect of the present disclosure, an electronic device is provided, the device includes a memory and a processor, the memory is used to store computer instructions that can be run on the processor, and the processor is used to execute the computer instructions when the computer instructions are executed. The method for driving interactive objects described in any of the embodiments provided in the present disclosure is implemented.

According to an aspect of the present disclosure, there is provided a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the method for driving an interactive object according to any one of the embodiments provided in the present disclosure is realized.

The driving method, device, device, and computer readable storage medium of an interactive object according to one or more embodiments of the present disclosure obtain a phoneme sequence corresponding to text data, and obtain at least one partial region of an interactive object matching the phoneme sequence The control parameter value of to control the posture of the interactive object, so that the interactive object can make a posture that matches the phoneme corresponding to the text data. The posture includes facial posture and body posture, so that the target object generates that the interactive object is speaking The sense of text content enhances the interactive experience between the target object and the interactive object.

Description of the drawings

In order to more clearly explain one or more embodiments of this specification or the technical solutions in the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, in the following description The drawings are only some of the embodiments described in one or more embodiments of this specification. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative work.

FIG. 1 is a schematic diagram of a display device in a method for driving interactive objects proposed by at least one embodiment of the present disclosure;

2 is a flowchart of a method for driving interactive objects proposed by at least one embodiment of the present disclosure;

FIG. 3 is a schematic diagram of a process of feature encoding for a phoneme sequence proposed by at least one embodiment of the present disclosure;

4 is a schematic structural diagram of a driving device for interactive objects proposed in at least one embodiment of the present disclosure;

FIG. 5 is a schematic structural diagram of an electronic device proposed in at least one embodiment of the present disclosure.

Detailed ways

The exemplary embodiments will be described in detail here, and examples thereof are shown in the accompanying drawings. When the following description refers to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementation manners described in the following exemplary embodiments do not represent all implementation manners consistent with the present disclosure. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

The term "and/or" in this article is only an association relationship describing the associated objects, which means that there can be three relationships, for example, A and/or B, which can mean: A alone exists, A and B exist at the same time, exist alone B these three situations. In addition, the term "at least one" in this document means any one of a plurality of or any combination of at least two of the plurality, for example, including at least one of A, B, and C, may mean including A, Any one or more elements selected in the set formed by B and C.

At least one embodiment of the present disclosure provides a method for driving interactive objects. The driving method may be executed by electronic devices such as a terminal device or a server. The terminal device may be a fixed terminal or a mobile terminal, such as a mobile phone, a tablet, or a game. The server includes a local server or a cloud server, etc., and the method can also be implemented by a processor calling computer-readable instructions stored in a memory.

In the embodiments of the present disclosure, the interaction object may be any virtual image capable of interacting with the target object. In an embodiment, the interactive object may be a virtual character, or may also be a virtual animal, virtual item, cartoon image, or other virtual images capable of implementing interactive functions. The display form of the interactive object may be 2D or 3D, which is not limited in the present disclosure. The target object may be a user, a robot, or other smart devices. The interaction manner between the interaction object and the target object may be an active interaction manner or a passive interaction manner. In an example, the target object can make a demand by making gestures or body movements, and trigger the interactive object to interact with it by means of active interaction. In another example, the interactive object may actively greet the target object, prompt the target object to make an action, etc., so that the target object interacts with the interactive object in a passive manner.

The interactive objects may be displayed through terminal devices, which may be televisions, all-in-one machines with display functions, projectors, virtual reality (VR) devices, and augmented reality (AR) devices Etc., the present disclosure does not limit the specific form of the terminal device.

Fig. 1 shows a display device proposed by at least one embodiment of the present disclosure. As shown in Figure 1, the display device has a transparent display screen, and a stereoscopic picture can be displayed on the transparent display screen to present a virtual scene and interactive objects with a stereoscopic effect. For example, the interactive objects displayed on the transparent display screen in FIG. 1 include virtual cartoon characters. In some embodiments, the terminal device described in the present disclosure may also be the above-mentioned display device with a transparent display screen. The display device is configured with a memory and a processor, and the memory is used to store computer instructions that can run on the processor. The processor is used to implement the method for driving the interactive object provided in the present disclosure when the computer instruction is executed, so as to drive the interactive object displayed on the transparent display screen to communicate or respond to the target object.

In some embodiments, in response to the sound-driven data used to drive the interactive object to output voice, the interactive object may emit a specified voice to the target object. The terminal device can generate sound-driven data according to the actions, expressions, identities, preferences, etc. of the target object around the terminal device to drive the interactive object to respond by issuing a specified voice, thereby providing anthropomorphic services for the target object. It should be noted that the sound-driven data can also be generated in other ways, for example, generated by the server and sent to the terminal device.

During the interaction between the interactive object and the target object, when the interactive object is driven to make a specified voice according to the sound driving data, the interactive object may not be able to drive the interactive object to make facial movements synchronized with the specified voice, making the interactive object dull when uttering the voice , Unnatural, affecting the interactive experience between the target object and the interactive object. Based on this, at least one embodiment of the present disclosure proposes a method for driving an interactive object, so as to improve the interaction experience between the target object and the interactive object.

FIG. 2 shows a flowchart of a method for driving an interactive object according to at least one embodiment of the present disclosure. As shown in FIG. 2, the method includes steps 201 to 203.

Step 201: Obtain a phoneme sequence corresponding to the text data.

The text data may be driving data used to drive the interactive object. The drive data can be drive data generated by the server or terminal device according to the actions, expressions, identity, preferences, etc. of the target object interacting with the interactive object, or drive data called by the terminal device from the internal memory. The present disclosure does not limit the method of obtaining the text data.

In the embodiment of the present disclosure, the phoneme corresponding to the morpheme can be obtained according to the morphemes contained in the text, so as to obtain the phoneme sequence corresponding to the text. Among them, the phoneme is the smallest phonetic unit divided according to the natural attributes of the speech, and a pronunciation action of a real person can form a phoneme.

In an embodiment, in response to the text being a Chinese text, the Chinese text can be converted into pinyin, a phoneme sequence can be generated using pinyin, and a timestamp for each phoneme can be generated.

Step 202: Obtain a control parameter value of at least one partial region of an interactive object that matches the phoneme sequence.

The local area is obtained by dividing the whole (including face and/or body) of the interactive object. The control of one or more local areas of the face may correspond to a series of facial expressions or actions of the interactive object. For example, the control of the eye area may correspond to the facial actions of the interactive object such as opening, closing, blinking, and changing the perspective; The control of the mouth area can correspond to facial actions such as closing the mouth of the interactive object and opening the mouth to different degrees. The control of one or more local areas of the body may correspond to a series of physical actions of the interactive object. For example, the control of the leg area may correspond to the actions of the interactive object such as walking, jumping, and kicking.

The control parameter of the local area of the interactive object includes the posture control vector of the local area. The attitude control vector of each local area is used to drive the local area of the interactive object to perform actions. Different posture control vector values correspond to different motions or motion ranges. For example, for the posture control vector of the mouth area, one set of posture control vector values can make the mouth of the interactive object slightly open, and another set of posture control vector values can make the mouth of the interactive object open wider. By controlling the vector values with different postures to drive the interactive objects, the corresponding local areas can make different actions or actions with different amplitudes.

The local area can be selected according to the action of the interactive object that needs to be controlled. For example, when the face and limbs of the interactive object need to be controlled to perform actions at the same time, the posture control vector values of all the local areas can be obtained; when the expression of the interactive object needs to be controlled At this time, the posture control vector value of the local area corresponding to the face can be obtained.

In the embodiment of the present disclosure, by performing feature encoding on the phoneme sequence, the control parameter value corresponding to the feature code can be determined, thereby determining the control parameter value corresponding to the phoneme sequence. Different encoding methods can reflect different characteristics of the phoneme sequence. The present disclosure does not limit the specific encoding method.

In the embodiment of the present disclosure, the corresponding relationship between the feature code of the phoneme sequence corresponding to the text data and the control parameter value of the interactive object can be established in advance, so that the corresponding control parameter value can be obtained through the text data. The specific method for obtaining the control parameter value matching the feature code of the phoneme sequence of the text data will be described in detail later.

Step 203: Control the posture of the interactive object according to the acquired control parameter value.

Wherein, the control parameter value, such as the posture control vector value, matches the phoneme sequence contained in the text data. For example, when the display device that displays the interactive object is controlled to display text according to the text data, and/or the display device is controlled to output speech according to the phoneme sequence corresponding to the text data, the gesture made by the interactive object is different from that of the interactive object. The output voice and/or the displayed text are synchronized, thereby giving the target object a feeling that the interactive object is speaking.

In the embodiment of the present disclosure, by obtaining the phoneme sequence corresponding to the text data, and obtaining the control parameter value of at least one local area of the interactive object matching the phoneme sequence, the posture of the interactive object can be controlled, so that the interactive object A gesture that matches the phoneme corresponding to the text data is made, and the gesture includes facial gestures and body gestures, so that the target object feels that the interactive object is speaking the text content, and the interactive experience of the target object is improved.

In some embodiments, the method is applied to a server, including a local server or a cloud server. The server processes text data, generates control parameter values of the interactive objects, and uses three-dimensional rendering according to the control parameter values. The engine performs rendering to obtain the animation of the interactive object. The server may send the animation to the terminal for display to communicate or respond to the target object, and may also send the animation to the cloud, so that the terminal can obtain the animation from the cloud to communicate or respond to the target object . After the server generates the control parameter value of the interactive object, the control parameter value may also be sent to the terminal, so that the terminal completes the process of rendering, generating animation, and performing display.

In some embodiments, the method is applied to a terminal, and the terminal processes text data, generates control parameter values of the interactive object, and renders the interactive object using a three-dimensional rendering engine according to the control parameter value to obtain the interactive The animation of the object, the terminal can display the animation to communicate or respond to the target object.

In some embodiments, the display device displaying the interactive object may be controlled to display text according to the text data, and/or the display device may be controlled to output speech according to the phoneme sequence corresponding to the text data.

In the embodiment of the present disclosure, since the control parameter value matches the phoneme sequence of the text data, the voice and/or text output according to the text data is different from controlling the gesture of the interactive object according to the control parameter value. In the case of synchronization, the gesture made by the interactive object is synchronized with the output voice and/or displayed text, giving the target object the feeling that the interactive object is speaking.

In some embodiments, the control parameter of at least one local area of the interactive object includes a posture control vector, and the posture control vector can be obtained in the following manner.

First, feature encoding is performed on the phoneme sequence to obtain the encoding sequence corresponding to the phoneme sequence. In order to distinguish it from the coding sequence mentioned later, the coding sequence corresponding to the phoneme sequence of the text data is called the first coding sequence, that is, the first coding sequence is obtained by performing feature coding on the phoneme sequence.

For multiple phonemes included in the phoneme sequence, a sub-coding sequence corresponding to each phoneme is generated.

In an example, it is detected whether there is a first phoneme corresponding to each time point, and the first phoneme is any one of the multiple phonemes; the encoding value at the time point where the first phoneme is present is set to The first value, the coding value at the time point without the first phoneme is set to the second value, and the coding sequence corresponding to the first phoneme can be obtained after assigning the coding value at each time point. For example, the code value at the time point when the first phoneme is present may be set to 1, and the code value at the time point when the first phoneme is not present may be set to 0. That is, for each phoneme of the multiple phonemes included in the phoneme sequence, it is detected whether the phoneme corresponds to the phoneme at each time point; the encoding value at the time point where the phoneme is present is set to the first value, and there is no phoneme. The coding value at the time point of the phoneme is set to the second value, and the coding sequence corresponding to the phoneme can be obtained after assigning the coding value at each time point. Those skilled in the art should understand that the above setting of the encoding value is only an example, and the encoding value can also be set to other values, which is not limited in the present disclosure.

The first coding sequence corresponding to the phoneme sequence is obtained according to the respective sub-coding sequences corresponding to the multiple phonemes.

In an example, for the sub-coding sequence corresponding to the first phoneme, a Gaussian filter may be used to perform a Gaussian convolution operation on the continuous values of the first phoneme in time, so as to filter and smooth the matrix corresponding to the feature encoding. The transition of the mouth area when each phoneme is converted.

FIG. 3 shows a schematic diagram of a driving method of interactive objects proposed by at least one embodiment of the present disclosure. As shown in FIG. 3, the phoneme sequence 310 contains phonemes j, i1, j, and ie4 (for brevity, only some phonemes are shown), and corresponding sub-coding sequences 321, 322, and 321 are respectively obtained for each phoneme j, i1, and ie4. 323. In each sub-coding sequence, the corresponding code value at the time point where the phoneme is present is set to a first value (for example, 1), and the corresponding code value at the time point without the phoneme is set to the second value ( For example, 0). Taking the sub-coding sequence 321 as an example, at the time point when there is phoneme j in the phoneme sequence 310, the value of the sub-coding sequence 321 is the first value 1, and at the time point when there is no phoneme j, the value of the sub-coding sequence 321 is the first value. The two value is 0. All the sub-coding sequences constitute the first coding sequence 320.

Next, according to the first coding sequence, a feature code corresponding to at least one phoneme is obtained.

According to the encoding values of the

sub-coding sequences

321, 322, and 323 corresponding to phonemes j, i1, and ie4, and the duration of the corresponding phonemes in the three sub-coding sequences, that is, the duration of j in the sub-coding sequence 321, The duration of i1 in the sub-coding sequence 322 and the duration of ie4 in the sub-coding sequence 323 can obtain the characteristic information of the

sub-coding sequences

321, 322, and 323.

In an example, a Gaussian filter may be used to perform Gaussian convolution operations on the consecutive values of phonemes j, i1, and ie4 in the

sub-encoding sequences

321, 322, and 323, respectively, to smooth the feature encoding to obtain the smoothed的第一coding sequence 330. That is, the Gaussian convolution operation is performed on the continuous value of the phoneme in time through the Gaussian filter, so that the code value in each code sequence changes from the second value to the first value or from the first value to the second value. smooth. For example, in addition to 0 and 1, the values of the coding sequence also show intermediate values, such as 0.2, 0.3, etc., and the posture control vector obtained according to the values of these intermediate states makes the interaction characters excessively move and change their expressions more smoothly , Naturally, improve the interactive experience of the target object.

In some embodiments, the feature code corresponding to at least one phoneme may be obtained by performing a sliding window on the first code sequence. Wherein, the first coding sequence may be a coding sequence after a Gaussian convolution operation.

A sliding window is performed on the coding sequence with a time window of a set length and a set step size, and the feature code in the time window is used as the feature code of the corresponding at least one phoneme. After the sliding window is completed, according to With multiple feature codes of, the second code sequence can be obtained. Since the duration of each phoneme is different, and the duration of each phoneme is different in proportion to the length of the time window, the number of phonemes corresponding to the feature code in the time window may be 1, 2 or even more depending on the position of the time window. As shown in FIG. 3, by sliding a time window of a set length on the first coding sequence 320 or the smoothed first coding sequence 330, feature code 1, feature code 2, feature code 3 are obtained respectively, and so on, After traversing the first code sequence, feature code 1, feature code 2, feature code 3,..., Feature code M are obtained, thereby obtaining a second code sequence 340. Among them, M is a positive integer, and its value is determined according to the length of the first coding sequence, the length of the time window, and the sliding step of the time window.

Finally, the attitude control vector of at least one partial region of the interactive object corresponding to the feature code is acquired.

According to feature code 1, feature code 2, feature code 3,..., feature code M, the corresponding attitude control vector 1, attitude control vector 2, attitude control vector 3,..., attitude control vector M can be obtained respectively, thereby obtaining attitude control 350 of the sequence of vectors.

The sequence 350 of the attitude control vector and the second coding sequence 340 are aligned in time. Since each feature code in the second coding sequence is obtained according to at least one phoneme in the phoneme sequence, the sequence of the attitude control vector Each control vector in 350 is also obtained based on at least one phoneme in the phoneme sequence. While playing the phoneme sequence corresponding to the text data, the interactive object is driven to make an action according to the sequence of the posture control vector, that is, the interactive object can be driven to emit the sound corresponding to the text content while making synchronization with the sound The action gives the target object the feeling that the interactive object is speaking, which improves the interactive experience of the target object.

Assuming that the feature code starts to be output at the set time of the first time window, the attitude control vector value before the set time can be set to the default value, that is, when the phoneme sequence is just started to be played, the interactive object A default action is made, and after the set time, the interactive object is driven to make an action using the sequence of the posture control vector obtained according to the first coding sequence. Taking Fig. 3 as an example, the feature code 1 starts to be output at time t0, which corresponds to the default attitude control vector before time t0.

The length of the time window is related to the amount of information contained in the feature code. In the case where the amount of information contained in the time window is relatively large, the cyclic neural network processing will output a relatively uniform result. If the length of the time window is too large, the expression of the interactive object may not correspond to part of the text; if the length of the time window is too small, the expression of the interactive object may appear rigid when speaking. Therefore, the duration of the time window needs to be determined according to the minimum duration of the phoneme corresponding to the text data, so that the actions taken by driving the interactive object have a stronger correlation with the sound.

The sliding step length of the time window is related to the time interval (frequency) of obtaining the attitude control vector, that is, it is related to the frequency of driving the interactive object to make an action. The length and step length of the time window can be set according to the actual interactive scene, so that the expressions and actions made by the interactive object are more closely related to the sound, and are more vivid and natural.

In some embodiments, when the time interval between phonemes in the phoneme sequence is greater than a set threshold, the interactive object is driven to take actions according to the set posture control vector of the local area. That is, when the interactive character pauses for a long time, the interactive object is driven to make a set action. For example, when the output voice pauses for a long time, the interactive object can be made to make a smiling expression or slightly swing the body to avoid the interactive object standing upright without expression when the pause is long, thereby making the interactive object speak more Natural and smooth, it improves the interaction between the target object and the interactive object.

In some embodiments, the feature code may be input to a pre-trained recurrent neural network, and the recurrent neural network outputs at least one of the interactive objects corresponding to the feature code according to the first coding sequence. The attitude control vector of the local area. Since the recurrent neural network is a time recurrent neural network, it can learn the historical information of the input feature code, and output the attitude control vector of the at least one local area according to the feature code sequence. Wherein, the characteristic coding sequence includes a first coding sequence and a second coding sequence. The recurrent neural network may be, for example, a long short-term memory network (Long Short-Term Memory, LSTM).

In the embodiment of the present disclosure, a pre-trained recurrent neural network is used to obtain the posture control vector of at least one local area of the interactive object corresponding to the feature code, and the historical feature information of the feature code and the current feature information are merged, thereby The historical attitude control vector has an impact on the current attitude control vector change, making the expression changes and body movements of the interactive characters more smooth and natural.

In some embodiments, the recurrent neural network can be trained in the following manner.

First, a feature coded sample is obtained, the feature coded sample is annotated with a true value, and the true value is a posture control vector value of at least one local area of the interactive object.

After obtaining the feature coding samples, the initial recurrent neural network is trained according to the feature coding samples, and the recurrent neural network is trained after the change of the network loss satisfies the convergence condition, wherein the network loss includes the recurrent neural network The difference between the attitude control vector value of the at least one local area and the real value obtained by network prediction.

In some embodiments, feature code samples can be obtained by the following method.

First, obtain a video segment of a character's voice, and obtain a plurality of first image frames containing the character according to the video segment. For example, a video segment in which a real person is speaking can be obtained.

Next, extract a corresponding voice segment from the video segment, obtain a sample phoneme sequence according to the voice segment, and perform feature encoding on the sample phoneme sequence. Wherein, the manner of encoding the sample phoneme sequence is the same as the encoding manner of the phoneme sequence corresponding to the text data described above.

According to the sample code sequence obtained by performing feature coding on the sample phoneme sequence, the feature code of at least one phoneme corresponding to the first image frame is obtained. Wherein, the at least one phoneme may be a phoneme within a set range of the appearance time of the first image frame.

Then, the first image frame is converted into a second image frame containing the interactive object, and the attitude control vector value of at least one local area corresponding to the second image frame is obtained. Wherein, the attitude control vector value may include the attitude control vector value of all the local areas, and may also include the attitude control vector value of some of the local areas.

Taking the first image frame as an image frame containing a real person as an example, the image frame of the real person can be converted into a second image frame containing the image represented by the interactive object, and the local area of the real person The posture control vector corresponds to the posture control vector of each local area of the interactive object, so that the posture control vector of each local area of the interactive object in the second image frame can be obtained.

Finally, the feature code of at least one phoneme corresponding to the first image frame obtained above is annotated according to the attitude control vector value to obtain feature code samples.

In the embodiment of the present disclosure, the video segment of a character is split into a plurality of corresponding first image frames and voice segments, and the first image frame containing the real person is converted into the second image containing the interactive object. Frames are used to obtain the attitude control vector corresponding to the feature code of the phoneme, so that the feature code has a better correspondence with the attitude control vector, so as to obtain high-quality feature code samples, so that the actions of the interactive objects are closer to the real actions of the corresponding characters.

FIG. 4 shows a schematic structural diagram of a driving device for interactive objects according to at least one embodiment of the present disclosure. As shown in FIG. 4, the device may include: a first obtaining unit 401, configured to obtain a phoneme sequence corresponding to text data; and second The acquiring unit 402 is configured to acquire a control parameter value of at least one partial region of an interactive object matching the phoneme sequence; the driving unit 403 is configured to control the posture of the interactive object according to the acquired control parameter value.

In some embodiments, the device further includes an output unit for controlling the display device displaying the interactive object to display text according to the text data, and/or controlling the display device according to the phoneme sequence corresponding to the text data Output speech.

In some embodiments, the second obtaining unit is specifically configured to: perform feature coding on the phoneme sequence to obtain a first coding sequence corresponding to the phoneme sequence; and obtain at least one phoneme corresponding to the phoneme sequence according to the first coding sequence. The feature code of; obtaining the attitude control vector of at least one local area of the interactive object corresponding to the feature code.

In some embodiments, when performing feature encoding on the phoneme sequence to obtain the first encoding sequence corresponding to the phoneme sequence, the second acquiring unit is specifically configured to: for multiple phonemes included in the phoneme sequence, A sub-coding sequence corresponding to each phoneme is generated; and the first coding sequence corresponding to the phoneme sequence is obtained according to the sub-coding sequences respectively corresponding to the multiple phonemes.

In some embodiments, when generating the sub-coding sequence corresponding to each phoneme for the multiple phonemes included in the phoneme sequence, the second acquiring unit is specifically configured to: detect whether there is a first phoneme corresponding to each time point. , The first phoneme is any one of the multiple phonemes; by setting the code value at the time point when the first phoneme is present to the first value, the time point when the first phoneme is not present is set to the first value. The encoding value of is set to a second value to obtain the sub-encoding sequence corresponding to the first phoneme.

In some embodiments, the device further includes a filtering unit for performing a Gaussian filter on the continuous value of the phoneme in time for the sub-coding sequence corresponding to each phoneme of the multiple phonemes. Gaussian convolution operation. In an embodiment, for the sub-coding sequence corresponding to the first phoneme, a Gaussian filter is used to perform a Gaussian convolution operation on the continuous values of the first phoneme in time, and the first phoneme is one of the multiple phonemes. Of any kind.

In some embodiments, when acquiring feature codes corresponding to at least one phoneme according to the first coding sequence, the second acquiring unit is specifically configured to: A sliding window is performed on the coding sequence, the feature code in the time window is used as the feature code of the corresponding at least one phoneme, and the second code sequence is obtained according to the multiple feature codes obtained by completing the sliding window.

In some embodiments, the driving unit is specifically configured to: obtain a sequence of a posture control vector corresponding to the second coding sequence; and control the posture of the interactive object according to the sequence of the posture control vector.

In some embodiments, the device further includes a pause drive unit, which is used to control the set control parameter value of the local area when the time interval between phonemes in the phoneme sequence is greater than a set threshold. The posture of the interactive object.

In some embodiments, when acquiring the attitude control vector of at least one partial region of the interactive object corresponding to the feature code, the second acquiring unit is specifically configured to: input the feature code into a pre-trained loop A neural network obtains a posture control vector of at least one local area of the interactive object corresponding to the feature code.

In some embodiments, the neural network is obtained through phoneme sequence sample training; the device further includes a sample acquisition unit for: acquiring a video segment of a character's voice, and acquiring a plurality of video segments containing the voice according to the video segment The first image frame of the character; extract the corresponding voice segment from the video segment, obtain a sample phoneme sequence according to the voice segment, and perform feature encoding on the sample phoneme sequence; obtain the corresponding voice segment of the first image frame Feature encoding of at least one phoneme; transforming the first image frame into a second image frame containing the interactive object, and obtaining the attitude control vector value of at least one local area corresponding to the second image frame; according to the attitude The control vector value is used to annotate the feature code corresponding to the first image frame to obtain a feature code sample.

In some embodiments, the device further includes a training unit for training the initial recurrent neural network according to the characteristic coding samples, and training to obtain the recurrent neural network after the change in network loss satisfies the convergence condition, wherein The network loss includes the difference between the attitude control vector value of the at least one local area and the labeled attitude control vector value predicted by the recurrent neural network.

At least one embodiment of this specification also provides an electronic device. As shown in FIG. 5, the device includes a memory and a processor. The memory is used to store computer instructions that can run on the processor. The method for driving interactive objects described in any embodiment of the present disclosure is realized by computer instructions.

At least one embodiment of this specification also provides a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, the method for driving an interactive object according to any embodiment of the present disclosure is realized.

Those skilled in the art should understand that one or more embodiments of this specification can be provided as a method, a system, or a computer program product. Therefore, one or more embodiments of this specification may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, one or more embodiments of this specification may adopt computer programs implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes. The form of the product.

The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, as for the data processing device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and for related parts, please refer to the part of the description of the method embodiment.

The foregoing describes specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims can be performed in a different order than in the embodiments and still achieve desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or sequential order shown in order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

The embodiments of the subject and functional operations described in this specification can be implemented in the following: digital electronic circuits, tangible computer software or firmware, computer hardware including the structures disclosed in this specification and structural equivalents thereof, or among them A combination of one or more. The embodiments of the subject matter described in this specification can be implemented as one or more computer programs, that is, one or one of the computer program instructions encoded on a tangible non-transitory program carrier to be executed by a data processing device or to control the operation of the data processing device Multiple modules. Alternatively or in addition, the program instructions may be encoded on artificially generated propagated signals, such as machine-generated electrical, optical or electromagnetic signals, which are generated to encode information and transmit it to a suitable receiver device for use by the data The processing device executes. The computer storage medium may be a machine-readable storage device, a machine-readable storage substrate, a random or serial access memory device, or a combination of one or more of them.

The processing and logic flow described in this specification can be executed by one or more programmable computers executing one or more computer programs to perform corresponding functions by operating according to input data and generating output. The processing and logic flow can also be executed by a dedicated logic circuit, such as FPGA (Field Programmable Gate Array) or ASIC (Application Specific Integrated Circuit), and the device can also be implemented as a dedicated logic circuit.

Computers suitable for executing computer programs include, for example, general-purpose and/or special-purpose microprocessors, or any other type of central processing unit. Generally, the central processing unit will receive instructions and data from a read-only memory and/or a random access memory. The basic components of a computer include a central processing unit for implementing or executing instructions and one or more memory devices for storing instructions and data. Generally, a computer will also include one or more mass storage devices for storing data, such as magnetic disks, magneto-optical disks, or optical disks, or the computer will be operatively coupled to this mass storage device to receive data from or send data to it. It transmits data, or both. However, the computer does not have to have such equipment. In addition, the computer can be embedded in another device, such as a mobile phone, a personal digital assistant (PDA), a mobile audio or video player, a game console, a global positioning system (GPS) receiver, or, for example, a universal serial bus (USB ) Flash drives are portable storage devices, just to name a few.

Computer-readable media suitable for storing computer program instructions and data include all forms of non-volatile memory, media, and memory devices, including, for example, semiconductor memory devices (such as EPROM, EEPROM, and flash memory devices), magnetic disks (such as internal hard disks or Removable disks), magneto-optical disks, CD ROM and DVD-ROM disks. The processor and the memory can be supplemented by or incorporated into a dedicated logic circuit.

Although this specification contains many specific implementation details, these should not be construed as limiting the scope of any invention or the scope of the claimed protection, but are mainly used to describe the features of specific embodiments of a particular invention. Certain features described in multiple embodiments in this specification can also be implemented in combination in a single embodiment. On the other hand, various features described in a single embodiment can also be implemented in multiple embodiments separately or in any suitable sub-combination. In addition, although features may work in certain combinations as described above and even initially claimed as such, one or more features from the claimed combination may in some cases be removed from the combination, and the claimed The combination of protection can be directed to a sub-combination or a variant of the sub-combination.

Similarly, although operations are depicted in a specific order in the drawings, this should not be construed as requiring these operations to be performed in the specific order shown or sequentially, or requiring all illustrated operations to be performed to achieve the desired result. In some cases, multitasking and parallel processing may be advantageous. In addition, the separation of various system modules and components in the above embodiments should not be understood as requiring such separation in all embodiments, and it should be understood that the described program components and systems can usually be integrated together in a single software product. In, or packaged into multiple software products.

Thus, specific embodiments of the subject matter have been described. Other embodiments are within the scope of the appended claims. In some cases, the actions recited in the claims can be performed in a different order and still achieve desired results. In addition, the processes depicted in the drawings are not necessarily in the specific order or sequential order shown in order to achieve the desired result. In some implementations, multitasking and parallel processing may be advantageous.

The above descriptions are only preferred embodiments of one or more embodiments of this specification, and are not intended to limit one or more embodiments of this specification. All within the spirit and principle of one or more embodiments of this specification, Any modification, equivalent replacement, improvement, etc. made should be included in the protection scope of one or more embodiments of this specification.

Claims

A driving method of interactive objects includes:

Obtaining the phoneme sequence corresponding to the text data;

Acquiring a control parameter value of at least one partial region of an interactive object matching the phoneme sequence;

Controlling the posture of the interactive object according to the acquired control parameter value.
The method according to claim 1, further comprising: controlling the display device displaying the interactive object to display text according to the text data, and/or controlling the display device to output voice according to the phoneme sequence corresponding to the text data.
The method according to claim 1 or 2, wherein the control parameter of the local area of the interactive object includes the attitude control vector of the local area;

Obtaining the control parameter value of at least one partial region of the interaction object matching the phoneme sequence includes:

Performing feature encoding on the phoneme sequence to obtain a first encoding sequence corresponding to the phoneme sequence;

Obtaining a feature code corresponding to at least one phoneme according to the first coding sequence;

Obtain a posture control vector of at least one local area of the interactive object corresponding to the feature code.
The method according to claim 3, wherein, performing feature encoding on the phoneme sequence to obtain the first encoding sequence corresponding to the phoneme sequence comprises:

For each phoneme of the multiple phonemes included in the phoneme sequence, generating a sub-coding sequence corresponding to the phoneme;

The first coding sequence corresponding to the phoneme sequence is obtained according to the respective sub-coding sequences corresponding to the multiple phonemes.
The method according to claim 4, wherein, for each phoneme of a plurality of phonemes included in the phoneme sequence, generating a sub-coding sequence corresponding to the phoneme comprises:

Detecting whether the phoneme corresponds to each time point;

By setting the code value at the time point with the phoneme to the first value, and the code value at the time point without the phoneme to the second value, the sub-coding sequence corresponding to the phoneme is obtained.
The method according to claim 5, further comprising:

For the sub-coding sequence corresponding to each phoneme of the plurality of phonemes, a Gaussian filter is used to perform a Gaussian convolution operation on the continuous values of the phonemes in time.
The method according to any one of claims 3 to 6, wherein, according to the first coding sequence, obtaining a feature code corresponding to at least one phoneme comprises:

Using a time window of a set length and a set step size, the first coding sequence is window-slid, the feature code in the time window is used as the feature code of the corresponding at least one phoneme, and according to the completion of the Obtaining a second coding sequence by using a plurality of the feature codes obtained by the sliding window;

Controlling the posture of the interactive object according to the acquired control parameter value includes:

Acquiring a sequence of attitude control vectors corresponding to the second coding sequence;

The posture of the interactive object is controlled according to the sequence of the posture control vector.
The method according to any one of claims 1 to 7, further comprising:

In the case that the time interval between the phonemes in the phoneme sequence is greater than a set threshold, the posture of the interactive object is controlled according to the set control parameter value of the local area.
The method according to claim 3, wherein acquiring a posture control vector of at least one local area of the interactive object corresponding to the feature code comprises:

The feature code is input to a pre-trained recurrent neural network, and the attitude control vector of at least one local area of the interactive object corresponding to the feature code is obtained.
The method according to claim 9, wherein the recurrent neural network is obtained by training of feature coding samples;

The method also includes:

Acquiring a video segment of a character uttering a voice, and acquiring a plurality of first image frames containing the character according to the video segment;

Extracting a corresponding speech segment from the video segment, obtaining a sample phoneme sequence according to the speech segment, and performing feature encoding on the sample phoneme sequence;

Acquiring a feature code of at least one phoneme corresponding to the first image frame;

Transforming the first image frame into a second image frame containing the interactive object, and obtaining a posture control vector value of at least one local area corresponding to the second image frame;

According to the attitude control vector value, the feature code corresponding to the first image frame is annotated to obtain the feature code sample.
The method according to claim 10, further comprising:

The initial recurrent neural network is trained according to the feature code samples, and the recurrent neural network is trained after the change of the network loss satisfies the convergence condition, wherein the network loss includes the at least one predicted by the recurrent neural network The difference between the attitude control vector value of the local area and the marked attitude control vector value.
A driving device for interactive objects, including:

The first acquiring unit is used to acquire the phoneme sequence corresponding to the text data;

The second acquiring unit is configured to acquire the control parameter value of at least one partial region of the interaction object that matches the phoneme sequence;

The driving unit is configured to control the posture of the interactive object according to the acquired control parameter value.
The apparatus according to claim 12, further comprising an output unit for controlling the display device displaying the interactive object to display text according to the text data, and/or controlling the display device according to the phoneme sequence corresponding to the text data Output speech.
The device according to claim 12 or 13, wherein the second obtaining unit is configured to:

Performing feature encoding on the phoneme sequence to obtain a first encoding sequence corresponding to the phoneme sequence;

Obtaining a feature code corresponding to at least one phoneme according to the first coding sequence;

Acquiring a posture control vector of at least one local area of the interactive object corresponding to the feature code;

Wherein, performing feature encoding on the phoneme sequence to obtain the first encoding sequence corresponding to the phoneme sequence includes:

For each phoneme of the multiple phonemes included in the phoneme sequence, generating a sub-coding sequence corresponding to the phoneme;

The first coding sequence corresponding to the phoneme sequence is obtained according to the respective sub-coding sequences corresponding to the multiple phonemes.
The apparatus according to claim 14, wherein when acquiring the feature code corresponding to at least one phoneme according to the first coding sequence, the second acquiring unit is configured to:

A sliding window is performed on the coding sequence with a time window of a set length and a set step size, the feature code in the time window is used as the feature code of the corresponding at least one phoneme, and the sliding is performed according to the completion of the sliding window. Multiple feature codes obtained by the window to obtain a second code sequence;

The driving unit is used for:

Acquiring a sequence of attitude control vectors corresponding to the second coding sequence;

The posture of the interactive object is controlled according to the sequence of the posture control vector.
The device according to any one of claims 12 to 15, further comprising:

The pause driving unit, when the time interval between the phonemes in the phoneme sequence is greater than a set threshold, controls the posture of the interactive object according to the set control parameter value of the local area.
The device according to claim 14, wherein, when acquiring the attitude control vector of at least one partial region of the interactive object corresponding to the feature code, the second acquiring unit is configured to: input the feature code to A pre-trained recurrent neural network obtains the attitude control vector of at least one local area of the interactive object corresponding to the feature code.
The device according to claim 17, further comprising a sample acquisition unit, configured to:

Acquiring a video segment of a character uttering a voice, and acquiring a plurality of first image frames containing the character according to the video segment;

Extracting a corresponding speech segment from the video segment, obtaining a sample phoneme sequence according to the speech segment, and performing feature encoding on the sample phoneme sequence;

Acquiring a feature code of at least one phoneme corresponding to the first image frame;

Transforming the first image frame into a second image frame containing the interactive object, and obtaining a posture control vector value of at least one local area corresponding to the second image frame;

Mark the feature code corresponding to the first image frame according to the attitude control vector value to obtain the feature code sample;

The device further includes a training unit for training the initial recurrent neural network according to the characteristic coding samples, and training to obtain the recurrent neural network after the change of the network loss satisfies the convergence condition, wherein the network loss includes the The difference between the attitude control vector value of the at least one local area and the marked attitude control vector value obtained by the cyclic neural network.
An electronic device, comprising a memory and a processor, the memory is used to store computer instructions that can run on the processor, and the processor is used to implement any one of claims 1 to 11 when the computer instructions are executed Methods.
A computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the method according to any one of claims 1 to 11 is realized.