This incident is resolved. Speech delivery has been normal since 7:51 AM PT (14:51 UTC).

Summary: Between 7:30 and 7:51 AM PT on July 31, an upstream text-to-speech provider experienced elevated latency. On affected calls (~7% during the window), the AI agent's responses were generated correctly but the audio was not delivered, so callers heard silence and calls ended in a silence timeout. Transcripts for affected calls may be missing agent responses.

The provider restored performance by adding server capacity, and we are following up with them on production SLOs. Our automatic failover to a backup voice provider did not engage during the degradation; we identified and fixed the defect in its latency detection, which will be deployed to production shortly. We apologize for the disruption to your calls.