Text-to-speech engine: speak(text, rate) with pause, resume, stop, live rate change and word-boundary events. No UI dependency.