Somehow need to overlap an LLM with vocal stream processing to identify semantically meaningful transition points to interrupt naturally instead of just looking

Somehow need to overlap an LLM with vocal stream processing to identify semantically meaningful transition points to interrupt naturally instead of just looking for any pause or sentence boundary.