Cascading type voice interaction method for moxibustion equipment

The voice interaction technology for moxibustion devices, which utilizes a two-level heterogeneous processing architecture and a lightweight large language model, solves the balance between low power consumption and high intelligence in moxibustion devices. It achieves high wake-up rate and high recognition rate in complex noise environments, ensuring the safety and intelligence of the devices.

CN122290586APending Publication Date: 2026-06-26WUHAN INST OF TECH
0 Cites 0 Cited by

Patent Information

Application Number
CN202610304456.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-03-13
Publication Date
2026-06-26

AI Technical Summary

Technical Problem

Existing voice interaction technology for moxibustion devices cannot achieve a balance between low power consumption and high intelligence. It cannot accurately identify complex consultation intentions, has a low wake-up rate in complex noise environments, cannot provide efficient intelligent interaction in offline environments, and lacks security.

Method used

It adopts a two-level heterogeneous processing architecture, including a low-power first-level speech processing unit and a high-performance second-level speech processing unit. Combined with a microphone array and a lightweight large language model, it achieves real-time response and complex semantic understanding, and ensures device security through a multi-level security redundancy mechanism.

Benefits of technology

It achieves high-intelligence interaction with low power consumption, improves wake-up rate, increases recognition accuracy, enhances user satisfaction, strengthens security, and still provides an efficient intelligent interaction experience in environments without network coverage.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention relates to the field of voice interaction of intelligent equipment, and discloses a cascaded voice interaction method for moxibustion equipment, and the method comprises the steps: collecting user voice, and carrying out the preprocessing of the user voice, and obtaining an effective voice frame and an acoustic feature pattern; the feature map is input into a primary voice processing unit, a low-power-consumption wake-up word detection and short instruction recognition model is operated, and a control instruction is directly output when a short instruction is recognized; if not, calculating a complex intention tendency score, and if a threshold value is exceeded, waking up a secondary voice processing unit; the secondary unit runs a locally deployed lightweight large language model, performs end-to-end speech recognition and semantic understanding, and generates an interaction instruction in combination with a moxibustion knowledge base; and finally, executing the instruction and broadcasting a result. According to the invention, low power consumption and high intelligence are both considered through a two-stage cascade architecture, offline complex semantic understanding is realized, and voice interaction experience in a moxibustion scene is improved.
Need to check novelty before this filing date? Find Prior Art