TTFB (Time To First Byte - Audio)

Category: science

The duration between the end of a user’s speech and the moment the AI agent begins streaming the audio response.

In voice AI, TTFB is the ultimate metric of perceived performance. Any latency over 500ms ruins the natural flow of conversation. Optimizing TTFB requires streaming TTS (Text-to-Speech) fragments, where the agent begins speaking the first sentence of a response while it is still generating the remainder of the thought in the background.

Common Examples

• We optimized our voice-agent pipeline to achieve a sub-300ms TTFB by parallelizing the LLM token generation with the initial TTS audio synthesis buffer.
• High TTFB values in conversational AI create "interruption fatigue," causing users to talk over the agent because they assume the connection has dropped.