datasets: CREMA-D / RAVDESS / TESS / JL Corpus models: NVIDIA A2F - 3D, LAM audio2 expression, check the HF page for links