Voice-Driven Virtual Population Type Method Based on Attention Mechanism

Based on the voice-driven virtual population algori thm, the audio synchronization accuracy of the character's mout h shape and audio is low. In order to improve the accuracy of the character's mouth shape and audio synchronization in the video. In this paper, the CBAM attention mechanism is added to the face encoding and decoding and speech encoding modules to solve t he problem of information loss in the long sequence information t ransmission of the CNN network and improve the accuracy of lip shape and audio synchronization. The experimental results show that, compared with the mainstream voice-driven virtual population algorithm, the improved model improves the accuracy of the mouth shape and audio synchronization in the video, and is close r to the corresponding real video character mouth shape visually.