Dialogue control method and device based on full-duplex voice model, equipment and medium
By constructing a full-duplex speech model and combining text roles and speech example information, the problems of single roles and uncustomizable timbre in existing models are solved, enabling efficient and personalized voice interaction in financial and health business scenarios, and improving the professionalism and naturalness of the service.
CN122347950APending Publication Date: 2026-07-07平安科技(上海)有限公司
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- 平安科技(上海)有限公司
- Filing Date
- 2026-04-14
- Publication Date
- 2026-07-07
Smart Images
Figure CN122347950A_ABST
Abstract
The application relates to the technical field of artificial intelligence, and discloses a dialogue control method and device based on a full-duplex voice model, equipment and a medium, which comprises the following steps: constructing a full-duplex voice model, obtaining historical training data, training the full-duplex voice model; constructing a mixed system prompt module based on text role information and voice example information; inputting the mixed system prompt module into the trained full-duplex voice model to complete context pre-filling configuration of the full-duplex voice model; inputting real-time voice input by a user into the full-duplex voice model; performing joint modeling processing on the real-time voice through the full-duplex voice model to output agent text conforming to role constraints and agent voice matching a target voice timbre, and completing full-duplex voice dialogue. The application can be applied to voice interaction business scenarios such as financial technology and medical health, efficiently recognizes a role voice timbre, generates a voice reply in a specified speaker voice timbre, and realizes voice timbre role double controllability.
Need to check novelty before this filing date? Find Prior Art