This repo is the official release of HQ-VoxCeleb dataset, which is proposed in Speech Fusion to Face: Bridging the Gap Between Human's Vocal Characteristics and Facial Imaging.
The main paper of this work is published on ACMMM 2022 as a full paper [arxiv]. Please refer to our supplementary material for more information about HQ-VoxCeleb dataset.
HQ-VoxCeleb is now open for download at: Google Drive Link.
The file structure of the above provided HQ-VoxCeleb is illustrated in the figure below. Face data of identities from VoxCeleb1 is stored in vox1/, and face data of identities from VoxCeleb2 are stored in vox2/. Under each partition, face images whose backgrounds are masked by image segmentation algorithm is stored in masked_faces/, and the original face images are stored in origin_faces/.
We provide several sampled data in our repo, please refer to samples.
