CN-122019803-A - Education virtual person generating method, related equipment and computer program product

CN122019803ACN 122019803 ACN122019803 ACN 122019803ACN-122019803-A

Abstract

The application discloses an educational virtual person generating method, relative equipment and computer program product, when receiving the instruction of virtual person, the application analyzes the called target role, selecting first type features and second type features matched with target roles from a feature library, carrying out feature solidification on the selected features and a virtual person basic model, generating an educational virtual person instance in the current educational scene, and deploying the educational virtual person instance into an application scene. According to the application, the character style and background knowledge of the target character are stably branded in the education virtual person example by means of the characteristic plug-in and the real-time solidification mode of the virtual person basic model, so that the problem that the character of the universal large model is forgotten is avoided, and the identity consistency and knowledge accuracy of the education virtual person are improved. In addition, the virtual person basic model can still be realized based on a powerful basic model, so that the flexibility of processing the openness problem can be reserved.

Inventors

YAN CHENXI
LIU YANG
JIANG MANMAN
Lv xing
LIU CHANG

Assignees

科大讯飞股份有限公司

Dates

Publication Date: 20260512
Application Date: 20260130

Claims (11)

1. A method of educational virtual person generation comprising: receiving a virtual person calling instruction in a current education scene, and analyzing a called target role from the virtual person calling instruction; According to the target role, inquiring a first type of characteristics and a second type of characteristics matched with the target role in a configured characteristic library, wherein the first type of characteristics comprise the personality style and knowledge characteristics of the target role, and the second type of characteristics comprise tone characteristics; And solidifying the first type of features and the second type of features matched with the target roles with the virtual person basic model, generating an education virtual person instance under the current education scene, and deploying the education virtual person instance into an application scene.
2. The method of claim 1, wherein the feature library comprises knowledge base and personality style data corresponding to more than one character label, and tone features corresponding to more than one tone label; then, a process of querying the first type of features and the second type of features matched with the target role in the configured feature library comprises the following steps: Inquiring a knowledge base and personality style data corresponding to the target role in a configured feature library to serve as the first type of features; and calling a large model to indicate the large model to select a target tone color label matched with the target role from the tone color labels based on a knowledge base and personality style data corresponding to the target role, wherein tone color characteristics corresponding to the target tone color label are used as the second type of characteristics.
3. The method of claim 1, wherein the step of solidifying the first and second types of features matched by the target character with the virtual person base model to generate the educational virtual person instance in the current educational scenario comprises: Binding a system prompt containing the target character human set description with a first type of characteristics matched with the target character, and then solidifying the system prompt into a prefix context when the virtual human basic model is inferred; Binding the second type of features with the virtual person basic model, and taking the second type of features as control parameters when the text generated by the virtual person basic model is subjected to voice/video synthesis.
4. The method of claim 1, wherein the second class of features further comprises character dynamics graphs.
5. The method of claim 1, wherein the target character invoked by the virtual person invocation instruction in the current educational scenario is a target teacher; The first type of characteristics matched with the target teacher comprise the personalized teaching style and knowledge characteristics of the target teacher, and the knowledge characteristics comprise teaching assets of the target teacher; The second type of features matched with the target teacher comprise tone features of the target teacher; the generated education virtual person instance under the current education scene is the education virtual person instance corresponding to the target teacher.
6. The method of claim 5, wherein the target teacher's personalized teaching style feature is obtained by: Acquiring teaching assets of the target teacher, wherein the teaching assets comprise at least one of video recording of history teaching, teaching courseware, classroom notes and online answering records; And calling the large model to indicate the large model to generate personalized teaching style characteristics of the target teacher based on the teaching assets of the target teacher.
7. The method of claim 1, wherein the virtual person invocation instruction in the current educational scenario is an educational virtual person invoking two or more target characters, and wherein the generated educational virtual person instance in the current educational scenario comprises educational virtual person instances corresponding to the two or more target characters.
8. The method of any of claims 1-7, wherein the virtual person invocation instruction in the current educational scenario further comprises task content describing a task that the invoked target character needs to perform, the method further comprising: And calling the generated education virtual person instance in the current education scene, and executing the task content.
9. An electronic device is characterized by comprising a memory and a processor; The memory is used for storing programs; the processor is configured to execute the program to implement the steps of the educational virtual person generating method according to any one of claims 1 to 8.
10. A readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the educational virtual person generating method according to any one of claims 1 to 8.
11. A computer program product comprising a computer program which, when executed by a processor, implements the steps of the educational virtual person generating method according to any of claims 1 to 8.

Description

Education virtual person generating method, related equipment and computer program product Technical Field The application relates to the technical field of virtual digital people, in particular to an educational virtual person generating method, related equipment and a computer program product. Background With the rapid development of artificial intelligence and online education, education virtual persons are taken as a novel interactive teaching tool, and have great application potential. They can play a variety of roles including history personas, science directors, language partners, etc., providing a personalized, immersive learning experience for the learner. One common virtual human generation scheme is a "role playing" model based on a generic Large Language Model (LLM). The solution takes advantage of the powerful generation capabilities of large models, and guides the model to play a role through specific Prompt words (promts). Its advantage is high flexibility and high effect on opening. However, the disadvantages are also very pronounced: The identity inconsistency is that the general large model is easy to have the problem of 'character forgetting' or 'person setting collapse' in multiple rounds of conversations, so that the virtual person identity is unstable. Behavioral instability and hallucination problems-the model may generate content that does not conform to the knowledge background or facts of the character (i.e., a "hallucination"), which is unacceptable in serious educational scenarios. Disclosure of Invention In view of the foregoing, the present application has been developed to provide an educational virtual person generating method, related apparatus, and computer program product that promote identity consistency and knowledge accuracy of an educational virtual person. The specific scheme is as follows: in a first aspect, there is provided an educational virtual person generating method, comprising: receiving a virtual person calling instruction in a current education scene, and analyzing a called target role from the virtual person calling instruction; According to the target role, inquiring a first type of characteristics and a second type of characteristics matched with the target role in a configured characteristic library, wherein the first type of characteristics comprise the personality style and knowledge characteristics of the target role, and the second type of characteristics comprise tone characteristics; And solidifying the first type of features and the second type of features matched with the target roles with the virtual person basic model, generating an education virtual person instance under the current education scene, and deploying the education virtual person instance into an application scene. In another implementation manner of the first aspect of the embodiment of the present application, the feature library includes a knowledge base and personality style data corresponding to more than one character label, and tone features corresponding to more than one tone label; then, a process of querying the first type of features and the second type of features matched with the target role in the configured feature library comprises the following steps: Inquiring a knowledge base and personality style data corresponding to the target role in a configured feature library to serve as the first type of features; and calling a large model to indicate the large model to select a target tone color label matched with the target role from the tone color labels based on a knowledge base and personality style data corresponding to the target role, wherein tone color characteristics corresponding to the target tone color label are used as the second type of characteristics. In another implementation manner of the first aspect of the embodiment of the present application, the process of solidifying the first class of features and the second class of features matched with the target character with the virtual person basic model to generate the educational virtual person instance in the current educational scene includes: Binding a system prompt containing the target character human set description with a first type of characteristics matched with the target character, and then solidifying the system prompt into a prefix context when the virtual human basic model is inferred; Binding the second type of features with the virtual person basic model, and taking the second type of features as control parameters when the text generated by the virtual person basic model is subjected to voice/video synthesis. In a possible design, in another implementation manner of the first aspect of the embodiment of the present application, the second type of features further includes a character dynamic graph. In another implementation manner of the first aspect of the embodiment of the present application, the target role called by the virtual person calling instruction in the current educational