3D digital human dynamic speaking face generation method and system based on tensor radiation field

The invention relates to a 3D digital human dynamic speaking face generation method and system based on a tensor radiation field, and the method comprises the steps: inputting a source video which comprises a video content in which a person speaks; performing feature extraction on the source video to obtain a posture feature, an audio feature and an expression feature corresponding to the character; combining the posture features, the audio features and the expression features into total features; generating three-dimensional coordinate information corresponding to the figure according to the source video; inputting the total features and the three-dimensional coordinate information into a dynamic tensor radiation field for tensor decomposition; and performing tensor reconstruction, and outputting a dynamic speaking face corresponding to the character. According to the invention, a 3D speaking face generation model based on dynamic tensor radiation field rapid training is built by using TensoRF, a high-qualit