SLF Evaluation Dataset

This dataset was constructed from the test set split of the VoxCeleb 2 dataset (VoxCeleb). The VoxCeleb 2 test set contains 118 speakers each in several different videos. To develop this dataset, only one video per speaker was selected. A face image was also extracted from the video, as well as, a low resolution face image (8x8). Age, gender and ethnicity of the person in the face image were determined using the “DeepFace” library, a face recognition and facial attribute analysis library. This dataset can be used to evaluate speech2face, speech conditioned face generation and speech conditioned face super-resolution systems.