A text-to-speech voice was created for him using hours of audio recordings from when he could still speak.