SEED is a first diffusion-based embedding enhancement framework designed to improve speaker representation robustness under adverse acoustic conditions. It leverages a powerful diffusion network ...