An interaction method, system, terminal device and storage medium

By acquiring and processing real-time environmental and audio information, and combining anchor point filtering and cross-modal fusion feature vector recognition to identify target intent, the problem of ambiguous intent determination in traditional intelligent interactive devices is solved, improving the accuracy and experience of user interaction.

CN122392524APending Publication Date: 2026-07-14DONGGUAN EDIFIER TECH
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
DONGGUAN EDIFIER TECH
Filing Date
2026-04-30
Publication Date
2026-07-14

AI Technical Summary

Technical Problem

Traditional smart interactive devices rely on single voice modality data for intent recognition, lacking the ability to perceive real-world interaction scenarios. This leads to errors in voice-to-text conversion and highly ambiguous intent determination, impacting the user's interactive experience.

Method used

The system acquires real-time environmental information and target user audio information, converts them into semantic information, performs anchor point filtering, determines cross-modal fusion feature vectors, and identifies target intent through adaptive weight information.

Benefits of technology

It improves the accuracy of intent recognition and user interaction experience by filtering irrelevant information and fusing multimodal data to provide more accurate responses.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122392524A_ABST
    Figure CN122392524A_ABST
Patent Text Reader

Abstract

The application relates to the technical field of voice interaction, and discloses an interaction method, an interaction system, a terminal device and a storage medium. The interaction method comprises the following steps: acquiring real-time environment information and real-time audio information, and converting the real-time audio information into real-time semantic information; performing anchor point screening on the real-time audio information, the real-time environment information and the real-time semantic information to obtain audio anchor point features, text anchor point features and environment anchor point features; determining a cross-modal fusion feature vector based on the audio anchor point features, the text anchor point features and the environment anchor point features; and determining a target intention of a target user based on the cross-modal fusion feature vector to interact with the target user. The above method can filter irrelevant information, extract a more comprehensive cross-modal fusion feature vector, improve the accuracy of intention recognition, and then give a correct response according to the accurate intention, thereby improving the user interaction experience.
Need to check novelty before this filing date? Find Prior Art