Voice response model online learning method and device based on user touch feedback

CN121938367APending Publication Date: 2026-04-28HONG KONG UNIV OF SCI & TECH (GUANGZHOU) +1
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
HONG KONG UNIV OF SCI & TECH (GUANGZHOU)
Filing Date
2026-03-09
Publication Date
2026-04-28

AI Technical Summary

Technical Problem

Existing voice intelligence systems struggle to adapt to different users' speaking speeds, pausing habits, and diverse dialogue scenarios. They also lack effective methods for obtaining implicit feedback, resulting in inflexible response timing and impacting the interactive experience.

Method used

By extracting acoustic feature sequences from real-time user voice signals, combining them with dilated temporal convolutional networks and Transformer modules to determine response timing, and simultaneously collecting touch signals to determine the operation type, online learning samples are constructed for model optimization to achieve continuous learning of personalized response timing.

Benefits of technology

Without disrupting natural voice interaction, it achieves continuous learning and optimization of users' personalized response timing preferences, thereby improving the accuracy and adaptability of response timing.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121938367A_ABST
    Figure CN121938367A_ABST
Patent Text Reader

Abstract

The invention discloses a voice response model online learning method and device based on user touch feedback, and belongs to the technical field of artificial intelligence and man-machine interaction, and the method comprises the steps: collecting a voice signal of a user in real time, and extracting an acoustic feature sequence of the voice signal; inputting the acoustic feature sequence into a voice response model for response opportunity judgment, and outputting a response triggering probability value; synchronously acquiring a touch signal of a user, and determining an operation type of the touch signal according to a comparison result between the response triggering probability value and a preset threshold value; constructing an online learning sample based on the acoustic feature sequence, and determining a sample label of the online learning sample according to the operation type; and continuously learning the voice response model according to the online learning sample to obtain the voice response model after parameter updating. Therefore, continuous learning and optimization of personalized response opportunity preference of the user can be realized on the premise that natural voice interaction is not damaged.
Need to check novelty before this filing date? Find Prior Art