Khin Mi Mi Aung

Forget brute-force audio attacks: this new method aligns generative model latent spaces with speaker embeddings, letting you impersonate voices with 10x fewer queries to speaker recognition systems.