MAI-Voice-1 is a natural speech generation model that the company claims can generate a minute of audio in under a second while running on a single GPU. It is already being used in Microsoft’s Copilot ...