New author alert! 🎉 @Seemanth_raju traces a voice agent's 3-second lag to hosts running at 96% CPU, not the model. Dedicated cores cut it to about 1 second, then silence recovery and turn-taking fixes lifted engagement. First DEV post! dev.to/seemanthraju/your-voi…
Your Voice Agent Isn't Slow Because of the LLM

Your Voice Agent Isn't Slow Because of the LLM

A production outbound voice agent was taking about three seconds to answer. We got it to about one....

dev.to

Nov 8, 2026 · 1:07 PM UTC

4
1
26
3,466
RelevantRecentLikes
The VAD cutting callers off mid-breath is the bit I'd hang on every voice team's wall. Plenty of real breaths between phrases run past 0.3 s. Good that you said to tune it on your own recordings.
22
ha, 96% cpu, classic. welcome to dev
14
Load more replies