Sub-50ms Voice2Voice Latency: Bare-Metal H100 Architecture

How VOCALIS achieves <50ms latency in voice2voice: bare-metal H100 architecture, streaming chunks, complete ASR/LLM/TTS benchmark.