Metric & Benchmark
Time to First Audio
TTS-specific latency from text-input to first audio packet — the voice analogue of TTFT.
Definition
TTFA measures how quickly a streaming TTS or voice agent begins playing audio after the user finishes speaking. Sub-300ms TTFA is the threshold for natural-feeling conversation, which is why SSM-based TTS like Cartesia Sonic targets it specifically.
Common use cases
- Voice agents
- Phone bots
- Real-time TTS