We introduce GaussianSpeech, a novel approach that synthesizes high-fidelity animation sequences of photo-realistic, personalized 3D human head avatars from spoken audio. To capture the expressive, detailed nature of human heads, including skin furrowing and finer-scale facial movements, we propose to couple speech signal with 3D Gaussian splatting to create realistic, temporally coherent motion sequences. We propose a compact and efficient 3DGS-based avatar representation that generates expression-dependent color and leverages wrinkle- and perceptually-based losses to synthesize facial details, including wrinkles that occur with different expressions. To enable sequence modeling of 3D Gaussian splats with audio, we devise an audio-conditioned transformer model capable of extracting lip and expression features directly from audio input. Due to the absence of high-quality datasets of talking humans in correspondence with audio, we captured a new large-scale multi-view dataset of audio-visual sequences of talking humans with native English accents and diverse facial geometry. GaussianSpeech consistently achieves state-of-the-art performance with visually natural motion at real time rendering rates, while encompassing diverse facial expressions and styles.
@article{arxiv.2411.18675,
title = {GaussianSpeech: Audio-Driven Gaussian Avatars},
author = {Shivangi Aneja and Artem Sevastopolsky and Tobias Kirschstein and Justus Thies and Angela Dai and Matthias Nießner},
journal= {arXiv preprint arXiv:2411.18675},
year = {2024}
}
Comments
Paper Video: https://youtu.be/2VqYoFlYcwQ Project Page: https://shivangi-aneja.github.io/projects/gaussianspeech