For voice stacks you need per-stage visibility: STT, turn-taking, LLM TTFT, TTS, and end-to-end TTFA — not just one aggregate latency number.
Tag every span with provider so cost and latency can be compared across backends.
Connections
- part_of Voice Agent