There’s a difference between being able to technically do this, and deploying it as a product with legal checks and scaled up infrastructure.
Since the article was about generative voice models, I assumed the ad part was also about that, but you’re right it might just be using voice recognition.