Robot anthropomorphic interaction method based on multi-modal emotion recognition and customized portrait generation

Abstract

The invention discloses a robot anthropomorphic interaction method based on multi-modal emotion recognition and customized portrait generation. The method comprises the following steps: S1, dynamically fusing multi-modal emotions; the method comprises the following steps: S1, synchronously acquiring voice, visual and text signals through a multi-source heterogeneous sensor, capturing a user voice stream by a high-fidelity microphone array, and extracting acoustic characteristics such as intonation and speed, S2, performing cross-modal reasoning; s3, synchronously generating contents; step S4: style migration; step S5, anthropomorphic voice and expression generation; according to the method, man-machine interaction emotion is analyzed and generated by utilizing a large language model and multi-modal information fusion, the singleness of interaction emotion and the deficiency of emotional sharing ability are avoided, a strong emotion interaction characteristic is achieved, the image of the robot is obtained through a generative technology and can be migrated to any image, the limitation that a specific image is independently made is broken through, and the interaction effect of the robot is improved. The advantage that one robot can be suitable for different scenes is achieved.

Images