That sounds like a fun challenge: make a voice-to-text widget that also generates phonetic or semantic annotations...