Wondering if most of the AI agents use real time apis or transcription apis.. anyone had experience with building voice agents can comment ?