i enabled krips in livekit on the inbound trunk but I don’t see any difference, it actually seems to be worse!

Thanks for reaching out, luisa. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.

one call in develop

Thanks — that helps. A few quick, focused questions to narrow this down: 1. Which specific symptom did you observe on the single dev call? (higher latency, muffled/artifacted audio, dropouts/silence, agent slow to respond, or something else?) 2. Was

krisp_enabled=true

set on the inbound trunk? (yes/no) 3. When you enabled Krisp, did you test toggling it off again — did behavior return to normal? 4. Can you check and paste any short logs or metrics from that call (server or agent CPU at the time, and any audio/VAD warnings like “silence detected”)? If you prefer, start by toggling Krisp off to confirm it’s the cause and tell me whether quality returns to expected. Once I have your answers or logs I’ll recommend the next specific steps. , with interactive elements

yes krisp enabled set true on the trunk the agent was not responding at all

Thanks — helpful detail. Because Krisp was enabled and the agent stopped responding, here are the most likely causes and the immediate steps to isolate and fix it. Why this can happen • Krisp uses significant CPU on the server/agent worker; if the worker is CPU-starved the agent can stop processing audio (no responses). • Aggressive noise suppression can change audio enough to break voice-activity detection (VAD) or silence detection, preventing the agent from seeing speech. Immediate troubleshooting (do these now) 1. Disable Krisp on the inbound trunk and retry the call. If the agent responds again, Krisp is the cause. 2. While reproducing the problem, check CPU on the LiveKit server and the agent process (top/htop). Note spikes or sustained high usage. 3. Check agent logs for audio/VAD warnings such as “silence detected” or errors around the time the call started. 4. If you see high CPU but need Krisp, run a single-call test on a beefier machine or increase the agent worker resources. If disabling Krisp fixes it but you need noise suppression, next options • Use a less CPU-constrained host for the agent worker. • Keep Krisp off server-side and apply client-side noise suppression (if available) before sending audio. • Tune or replace VAD/silence detection settings if they’re exposed in your agent pipeline. If you can, paste a short snippet of agent logs around the failed call and a CPU snapshot; I’ll point to the most relevant next step. , with interactive elements

yes I disabled it and now it’s better! it ends up responding