Hey. Got a bilingual (PT/EN) voice agent on LK Agents sitting at ~2.3s p50 per turn. Stack is Gladia + Sonnet 4.6 + ElevenLabs flash_v2_5 (native plugin, WS streaming), all tuned over weeks with vendor help. LLM TTFT is ~1s and I can't seem to push it lower. The voice demo on livekit.io's homepage is obviously in a different class. Main question: is the homepage demo a classic STT/LLM/TTS chain or an end-to-end realtime speech model? Because if it's realtime speech, 1s LLM TTFT is a structural floor and I need to rethink. Full breakdown, per-component numbers, and config on the forum: https://community.livekit.io/t/stuck-at-2-3s-p50-after-weeks-of-tuning-is-the-livekit-io-homepage-demo-a-classic-chain-or-realtime-speech/851

Hi Nuno Pinto, ⚠️ This channel will be deleted soon ⚠️ Please use our new community at https://community.livekit.io for technical discussions and to post technical questions. The Slack community is not going away, but we are migrating technical questions to our new forum. Please see full details here.