Nyx avatar (Gaussian-splat head, ARKit-52 blendshape rig) driven by a surprise clip's blendshape labels derived from this dataset.
14,082 emotional-speech clips, each annotated with two parallel 52-channel ARKit blendshape sequences (NVIDIA Audio2Face-3D-v2.3.1-James and LAM_Audio2Expression) plus a 26-dimensional NVIDIA Audio2Emotion conditioning vector.
Reference-only dataset — the original audio is not shipped. Each row contains a… See the full description on the dataset page:
https://huggingface.co/datasets/myned-ai/audio2face-emotion-arkit-teacher.
