a voice agent (STT → LLM → TTS full loop) where the specific technical differentiator/challenge is robustness under low-bitrate/degraded audio conditions