During WWDC26 Keynote a second generation on-device model was announced with better speech generation capabilities. Is there a new API available for developers to generate speech?