Microsoft claims text-to-speech model achieves parity with human readers, needs as little as 3 seconds to clone someone's voice. - Microsoft