I just released a research project I did last year as open source. It is a fully local speech-to-speech backend for LLM NPCs. So speech-to-text, local LLM, text-to-speech, no cloud needed. The main focus was NPCs talking to each other, not just answering the player, and my study looked at how players experience witnessing those NPC-to-NPC conversations and what it does for immersion. The NPCs can talk to each other, remember what they said, and later use that context when the player talks to...