How can I improve my agent latency?
There is no single source of latency, a turn will involve:
-
Determining when the user has finished talking, and are expecting a response from the agent
-
Speech-to-Text (STT), converting the transcribed audio into LLM tokens
-
Large Language Model (LLM), processing the tokens and replying
-
Text-to-Speech (TTS), converting the LLM response into natural-sounding audio.
For more information, please see this guide to understand and improve your agent latency: Understand and Improve Agent Latency | LiveKit